Rodriguez, R., D. Jenkins, J.K. Leary and B.V. Mahnken ... · Rodriguez, R., D. Jenkins, J.K....

Rodriguez R D Jenkins JK Leary and BV Mahnken 2016 Herbicide Ballistic Technology Spatial Tracking Analysis of Operations Characterizing Performance of Target Treatment Trans ASABE Transactions of the ASABE 59(3) 803-809

Rodriguez R D Jenkins JK Leary 2015 Design and Validation of a GPS Logger System for Recording Aerial-Deployed Herbicide Ballistic Technology Operations IEEE Sensors 15 2078-2086

Leary JK B Mahnken L Cox A Radford J Yanagida T Penniman D Duffy and J Gooding 2014 Reducing Nascent Miconia (Miconia calvescens DC) Patches through Accelerated Interventions Utilizing Herbicide Ballistic Technology (HBT) Inv Plant Sci Mgmt 7 (1) 164-175 DOI 101614IPSM-D-13-000591 Awarded 2015 Outstanding Paper for IPSM

Leary JK J Gooding J Chapman B Mahnkan A Radford L Cox 2013 Calibration of an Herbicide Ballistic Technology (HBT) Helicopter Platform Targeting Miconia (Miconia calvescens DC) in Hawaii Inv Plant Sci Mgmt 6(2) 292-303 DOI 101614IPSM-D-12-000261 (with press release see popular media)

Transactions of the ASABE

Vol 59(3) 803-809 copy 2016 American Society of Agricultural and Biological Engineers ISSN 2151-0032 DOI 1013031trans5911474 803

HERBICIDE BALLISTIC TECHNOLOGY SPATIAL TRACKING ANALYSIS OF OPERATIONS CHARACTERIZING

PERFORMANCE OF TARGET TREATMENT

R Rodriguez III J J K Leary D M Jenkins B V Mahnken

ABSTRACT Since 2012 the Herbicide Ballistic Technology (HBT) platform has been deployed in helicopter operations with a mission to eliminate nascent populations of the invasive plant species miconia (Miconia calvescens DC) which is spreading across the East Maui Watershed in Hawaii The HBT platform is a refined pesticide application system that pneumatically delivers encapsulated herbicide projectiles (ie paintballs) from a long range (up to ~30 m) and varying attitude This onboard system provides accurate effective treatment of individual plant targets occupying remote inacces-sible portions of the forested landscape Statistics of operational performance are acquired through GIS analyses of rec-orded GPS data assigned to treated plant targets Recently we have developed a telemetry system for HBT applications (HBT-TS) to enhance the attribute data of target treatments The HBT-TS integrates a hardware sensor with the electro-pneumatic marker that is actuated by the trigger to generate time-stamped geo-referenced attribute data including target assignment azimuth tilt and range determined from the applicator position for every projectile discharged With target assignments the HBT-TS records the estimated dose applied to each target Furthermore the time stamps show that the actual time to administer projectiles (ie target treatment) is a minor component of the total time on target By tracking the orientation and distance of the discharged projectile we can calculate a precise offset target location relative to the applicator position and provide a more accurate interpretation of the herbicide use rate (g acid equivalent ha-1) based on the known amount of herbicide contained in each projectile and the final placement on the landscape We acknowledge the challenges of GPS inaccuracies while recording in a dynamic environment (ie a moving platform in extreme topog-raphy) albeit with increased precision Regardless the current state of the HBT-TS technology enhances operational in-telligence relevant to landscape-scale invasive species management

Keywords Aerial application GIS GPS Miconia calvescens DC Remote sensing Telemetry

he Hawaiian archipelago is the most isolated group of islands on Earth resulting in the evolu-tion of endemic highly specialized biotic com-munities (Carlquist 1970 Mueller-Dombois et

al 1981 Stone and Scott 1985) Currently approximately 1100 endemic plant species and 4600 introduced exotic species have been catalogged of which about 800 are con-sidered invasive (Stone and Scott 1985) These invasive plant incursions are threatening the ecological integrity and function of Hawaiirsquos forested watersheds (Cox 1999 Lin-denmayer et al 2000 Loope and Kraus 2009 Loope et al 2004 Mooney and Drake 1986 Stone et al 1992)

One of the most significant plant invasions in Hawaii is miconia (Miconia calvescens DC) in the East Maui Water-shed (EMW) The EMW is 55000 ha of forested landscape on the northeastern slope of Haleakala volcano Miconia was introduced as an ornamental nursery stock plant in Hana in 1970 Today the core infestation is approaching 1000 ha surrounding Hana with small nascent populations spread across 20000 ha of the EMW Interventions on these incipient outlier populations are critical to maintain-ing an effective containment strategy for protecting the most intact endemic forest areas The extreme topography and remote location of the EMW are major impediments to effective intervention with ground-based operations (Kueffer and Loope 2009) Herbicide Ballistic Technology (HBT) is a novel pesticide application technology (regis-tered as a 24(c) Special Local Need pesticide in Hawaii) that addresses these challenges with a capability to effec-tively treat individual plant targets with long-range accura-cy and flexible attitude from a helicopter while conducting surveillance operations (Leary et al 2014)

Since 2012 HBT interventions have eliminated over 14000 incipient miconia targets as recorded by GPS from the applicator position on the aircraft However with an effective range of ~30 m these coordinates are not the ac-tual target locations Furthermore the herbicide dose has

Submitted for review in August 2015 as manuscript number ITSC

11474 approved for publication by the Information Technology Sensorsamp Control Systems Community of ASABE in January 2016

The authors are Roberto Rodriguez III ASABE Member Graduate Student Department of Biosciences and Bioengineering University ofHawaii at Manoa Honolulu Hawaii James J K Leary Extension Specialist Department of Natural Resources and EnvironmentalManagement University of Hawaii at Manoa Kula Hawaii Daniel M Jenkins ASABE Member Professor Department of Biosciences andBioengineering University of Hawaii at Manoa Honolulu HawaiiBrooke V Mahnken GIS Specialist Maui Invasive Species CommitteeMakawao Hawaii Corresponding author Roberto Rodriguez III 218Agricultural Science Building 1955 East-West Road Honolulu HI96822 phone 727-415-6712 e-mail roberto6hawaiiedu

T

804 TRANSACTIONS OF THE ASABE

been calculated as a composite average of the total estimat-ed projectile inventory discharged to the total number of targets recorded To enhance analyses of operations a te-lemetry system was developed with sensor hardware inte-grated into the electro-pneumatic propulsion system (ie paintball marker) in which the data acquisition sequence is actuated by the trigger pull populating a data matrix with the azimuth tilt target ID time stamp and geodetic coor-dinates for each projectile discharged (Rodriguez et al 2015) In this article we report on a second-phase iteration of the telemetry system for HBT applications (HBT-TS) with a laser range finder augmenting the sensor hardware which introduces the capability to geometrically calculate the offset location of the target from the applicator position This further increases the resolution of the calculated met-rics pertinent to the operation including spatial statistics using the distance to the target to estimate the final position of each projectile and the herbicide use rate

In 2013 we developed our first iteration of a fully op-erational HBT-TS that could perform basic functions in recording geo-referenced locations with time stamps of all actions specific to an HBT operation (Rodriguez et al 2015) The telemetry hardware consists of a printed circuit board populated with several sensors including barometric altimeter gyroscope magnetic compass and triple-axis accelerometer that transmit data via Bluetooth to a devoted software application written in Android open source code We have since initiated phase II development of the HBT-TS with incorporation of a laser range finder (LRF) capable of quantifying the distance to target treatment (up to 40 m) As described by Rodriguez et al (2015) the HBT-TS sen-sor device is mounted parallel to the barrel on the side pi-catinny rail a series of extrusions with T-shaped cross-sections used to mount accessories onto a firearm or similar device of the electro-pneumatic marker (BT TM7 Kee Action Sports LLC Sewell NJ) with an interrupt-enabled IO pin connected to the open-drain contact controlling the firing-solenoid current When this signal is observed to go

to ground (ie projectile is discharged) the system records the current time along with the most current sensor data attributed to the projectile and target (fig 1)

MATERIALS AND METHODS A total of five aerial miconia interventions were conduct-

ed in February and March of 2015 for testing the second-generation HBT-TS with LRF in a dynamic operational set-ting The orientation sensors on the device were calibrated using a compass and clinometer (360PC360R Suunto Tan-dem Amer Sports Corp Helsinki Finland) prior to the first operation of each day The barometric altimeter was calibrat-ed before each operation using the known elevation of the heliport or landing zone All operations were conducted us-ing a Hughes 500D helicopter with the applicator positioned port side behind the pilot so the pilot and applicator shared the field of view for target detection and engagement Upon detection the applicator pressed the target acquisition button (fig 1) establishing a unique target ID to be assigned to all subsequent projectiles discharged until the next acquisition In the meantime the pilot approached the target for a clear line of sight within effective range and held the position while the applicator administered the projectile dose to the target In these operations all projectiles (x = 6845) were correctly assigned to targets (n = 365)

The distribution of the applicator GPS position and cal-culated offset coordinates for each target were calculated using the circular map accuracy standard (CMAS) and spherical accuracy standard (SAS) The CMAS is the mag-nitude of the radius of the smallest circle in the plane tan-gent to the Earth at the mean location corresponding to latitude and longitude containing 90 of the recorded po-sitions (USBB 1947) The SAS is the magnitude of the radius of the smallest sphere containing 90 of the record-ed positions (Greenwalt 1962) Assuming normally dis-tributed errors in the horizontal coordinates CMAS can be described mathematically as

cσ= 14602CMAS (1)

where σc is the circular standard error which can be ap-proximated by

( )yxc σ+σasympσ2

1 (2)

where σx and σy are the standard deviations in horizontal position corresponding to errors in latitude and longitude respectively assuming that the larger of the two is less than five times the magnitude of the other (Greenwalt 1962) SAS can be estimated as

sσ= 52SAS (3)

where σs is the spherical standard error approximated by

( )zyxs σ+σ+σasympσ3

1 (4)

where σz is the standard deviation in altitude assuming that Figure 1 HBT-TS mounted to electro-pneumatic marker with majorparts labeled

59(3) 59(3)803-809 805

the smallest of the three component standard deviations is not less than 035 times the magnitude of the largest (Greenwalt 1962)

The interquartile ranges of the spatial statistics are pre-sented along with the standard deviations which are com-posed of deliberate adjustments (eg offsets due to aiming the marker at different parts of the target to disperse herbi-cide) in addition to some small random variation in sensor response Since the azimuth (θ) is capable of spanning the full 360deg the directional mean and standard deviation are calculated from equations 5 through 8 (Berens 2009)

=

θ=n

iis

0

sin (5)

=

θ=n

iic

0

cos (6)

gtlt+

lt+

gtgt

=θ

0 0360arctan

0180arctan

0 0arctan

csc

s

cc

s

csc

s

(7)

( )22ln2 css +minus=θ (8)

Herbicide use rate (HUR) is the amount of pesticide ap-plied per unit area eg g acid equivalent (ae) ha-1 An HBT projectile contains 01994 g ae of the active ingredient triclopyr (HBT-G4U200 Nelson Paint Co Kingsford Mich) and is estimated to produce a circular spatter pattern that has a radius of ~1 m (ie 314 m2) Spatial analyses were performed to calculate the area of the footprint and HUR using ArcMap 101 (ESRI Redlands Cal) Area footprints were calculated for each projectile target and total net treated area using the Buffer Analyses tool with

1 m radius and the Dissolve Type parameter set to None List (using the identifier ldquoTarget_IDrdquo) and All commands respectively To determine the HUR distribution across the entire treated area a point density analysis was performed with a 00625 m2 cell size and a 1 m circular neighborhood to measure the magnitude of the herbicide dose (01994 g ae projectile-1 x = 6845) overlaid onto the total net area footprint created by the offset placement of the projectiles Aspect values were extracted from a 10 m digital elevation model layer of Maui (NOAA 2007) at each of the offset point coordinates to calculate the angular line of sight (L) to target with respect to the terrain aspect (mA)

AmL minusθ= (9)

RESULTS AND DISCUSSION An HBT application is an advancement of a surveillance

operation with the capability of immediate intervention on high-value incipient plant targets One of the critical attrib-utes of HBT is the capability to efficiently and effectively treat targets occupying extreme terrain that are otherwise inaccessible (fig 2) This includes situations where the ap-plicator can engage a target from different lines of sight (fig 2b) or multiple targets from a single applicator position with different trajectories (fig 2c) Target engagement (upon de-tection) is a systematic process that momentarily deviates from the normal search pattern to approach and treat the tar-get Consistency in these treatment actions includes a steady position of the applicator the attitude of the projectile range to target and treatment dose (table 1) Most of these statistics were slightly leptokurtic with a strong peak at the median along with a minority of values as large positive outliers making the averages of most distributions somewhat larger than the corresponding medians

The target dose had a median value of 15 projectiles with an interquartile range from 9 to 23 projectiles This is less than the median of an average 24-projectile dose over the last three years of operations but is consistent with

(a) (b) (c)

Figure 2 (a) Applicator (bull) and target offset (times) locations calculated from range azimuth and tilt values (gray lines) overlaid on 10 m digital elevation model (b) a target treated from multiple applicator positions and (c) multiple targets treated from a single applicator position


smaller targets becoming more common as more mature plants are eliminated (data not shown) The marker is a rapid fire semi-automatic system with discharge rates of ~22 projectiles s-1 (fig 3) Hence the median treatment time is lt7 s (data not shown) We suggest that the low CMAS and SAS precision values of the applicator position are predicated on this brief time period while the helicopter is holding hover (table 1) The accuracy of the target loca-tions is unknown as the inaccessibility of the terrain makes ground truthing impossible However previous static ex-periments indicated that the position accuracy is compara-ble to that of other commercial-grade GPS units with GPS error being the limiting factor (Rodriguez 2015)

The CMAS and SAS values for the target offset location are less precise than the applicator position where the de-viations of the projectilersquos range and attitude (eg tilt and azimuth) had a compounding effect (table 1) These varia-

tions in the recorded attitude are primarily due to operator adjustment to cover andor zero in on the target and account for 78 of the observed spread in estimated target offset locations in CMAS and 70 of the observed spread in SAS after accounting for random errors estimated from static measurements The CMAS and SAS of the target offset locations show weak dependence on herbicide dose and the SAS is not substantially larger than the CMAS (fig 4) When a 1st1st degree rational polynomial is fitted to the CMAS the resulting horizontal asymptote is 50 m This equates to a maximum treatment diameter of 10 m con-sistent with the diameter of miconia (Harling et al 1973) and a maximum treatment area of approximately 80 m2 The same process for SAS shows a similar asymptote at 62 m equating to a treatment volume of approximately 560 m3 If the treatment area is instead considered an annu-lar region comprising of 90 of the deviation observed in the azimuth and range then the total area calculated is 125 m2 Similarly considering the volume bounded by 90 of the deviation seen in the azimuth tilt and range the total volume would be 94 m3 Adding the CMAS and SAS to these values results in an average treatment area of 160 m2 and treatment volume of 191 m3 per treated plant This estimate of the treatment area more closely resembles that calculated by GIS analysis and better reflects the actual shape of the treatment area (fig 6b)

These aerial operations were conducted by searching for miconia occupying extreme terrain (eg slopes up to 70deg) Operational flight lines are typically transects running par-allel to the contour with a bias on the port side where the pilot and applicator share the same field of view ie ~ 270deg to 330deg relative to the aircraft bearing offering direct observation by both parties of the terrain search area When a target was detected the pilot approached the target to a median distance of 152 m with an interquartile range not exceeding 20 m and always approached from above the target where the median tilt of the applicator was -33deg with respect to the horizontal plane and with an interquartile range less than -20deg (table 1) From these positions targets were confronted with lines of sight almost directly opposite

Table 1 Statistical analysis of HBT application Range and tilt dataare from entire population of projectiles (x = 6845) All other dataincluding those preceded by SD (standard deviation of theparameter) are for projectiles grouped by target (n = 365) Becauseazimuth to target is semi-random for each target only the standarddeviation of the projectile azimuth for each target is reported forthese data

Metric Median Interquartile Range Average Applicator precision and spatial statistics CMASA (m) 07 01 to 15 11

SASA (m) 10 05 to 17 13 Range (m)[a] 152 133 to 178 156 SD range (m) 08 03 to 15 10 Tilt (deg)[b] -330 -434 to -231 -331 SD tilt (deg) 10 07 to 15 13 SD azimuth (deg)[c] 39 28 to 55 49

Target offset precision CMAST (m) 17 11 to 23 19

SAST (m) 19 13 to 28 23 Treatment statistics Projectile dose[d] 15 9 to 23 187

HUR (g ae ha-1)[e] 21030 16285 to 29205 24597 [a] Distance to target measured by laser rangefinder [b] Tilt measured with respect to the horizontal plane [c] Azimuth measured with respect to geographic north [d] Number of projectiles discharged during target treatment [e] Herbicide use rate with 01994 g ae per projectile and herbicide distri-

bution area calculated from a 1 m radius created on target impact

Figure 3 Treatment time dependence on projectile dose ( n = 338 y = 04572x R2 = 05575) Targets requiring reload andor applicatorrepositioning (times) result in larger treatment time (outliers n = 27)

Figure 4 CMAS ( ) and SAS (times) of predicted target offset locations versus projectile dose (n = 365 CMAS R2 = 019 SAS R2 = 015) Dashed line indicates CMAS trend and solid line indicates SAS trend

y = 04572xRsup2 = 05575

0

10

20

30

40

50

60

0 50 100

Tre

atm

ent

Tim

e (s

)

Projectile dose (No)

0

1

2

3

4

5

6

7

8

9

0 50 100

CM

AS

and

SA

S (

m)


59(3) 59(3)803-809 807

the terrain aspect with a median azimuth of 191deg (fig 5) The terrain aspect is the vector normal to the terrain surface projected into the horizontal plane and is also assumed to be the aspect for the plant colonizer such that an angle of 180deg in the plot indicates application directly into the con-tour of the hillside (ie plant target) The interquartile range was 163deg to 222deg where angles gt180deg could indicate efficiency in approaching the target immediately upon de-tection while angles lt180deg might relate to efforts to reposi-tion the aircraft for a better line of sight free of obstacles (eg tree canopy) Angles measured beyond perpendicular (ie gt270deg and lt90deg) are improbable and most likely rep-resent the compounding errors associated with the GPS and the terrain model particularly along ridges where aspect can change dramatically with position

Time on target (ToT) is the total time spent in the target acquisition and treatment process In a previous study Leary et al (2014) found total search effort to be dependent on target density where the slope value was equivalent to the added time for every target encountered (ie ToT) The actual treatment time recorded as the time interval between the first and last projectile discharged was a minor time element in these operations (table 2) The majority of the time appears to have been spent approaching and maneu-vering the applicator into position for target treatment Time spent reloading projectiles (191 plusmn83 s) and reposi-tioning the applicator (201 plusmn108 s) were other time ele-ments contributing to ToT although these activities were

not required for the majority of targets treated A single projectile creates a treatment footprint area es-

timated at 314 m2 with an equivalent HUR of 6347 g ae ha-1 which is 94 of the maximum allowable rate (ie 6720 g ae ha-1 year-1) The mean target HUR in these oper-ations was 2460 g ae ha-1 with 983 of targets below the maximum HUR (fig 6a) The average target dose and area footprint were 36 g ae and 151 m2 for targets below the maximum HUR (fig 6b) On the other hand for those six targets that exceeded the maximum HUR the average area footprint was only slightly higher at 172 m2 while the av-erage dose increased to 138 g ae These high doses are

Figure 5 Frequency plot of projectile azimuth with respect to theterrain aspect at the predicted target location (x = 6845)

Table 2 Time elements of the target acquisition and treatment process Name[a] Activity Characteristic Time (s)

Time on target Time spent at a target not in active search Total time 555 plusmn49 Treatment time Time to administer projectile dose to target Time between first and last projectile discharged 92 plusmn65

Reload time Reloading hopper with projectiles interrupting treatment action

Reload every ~160 projectiles 191 plusmn83

Reposition time Relocating to treat target from a different location Standard deviation of azimuth and range of target increase substantially

201 plusmn108

[a] Time on target was extrapolated from the slope of search effort dependent on target densities encountered as described by Leary et al (2014) Treatment time (n = 338) reload time (n = 20) and reposition time (n = 7) were direct empirical values derived from the time stamps

(a)

(b)

Figure 6 (a) Target herbicide use rate (HUR g ae ha-1 n = 365 x = 6845) calculated from the projectile dose and net area of the projectile offset locations buffered with a 1 m radius and (b) net target offset area created by projectile dose Green dashed line is the HUR for a single isolated projectile and red dashed line is the maximum allowa-ble HUR according to the registered pesticide label

90deg

0deg

180deg

270deg

80

60

40

0

3000

6000

9000

0 100 200 300 400

HU

R (

g ae

ha-

1 )

Target ID

0

10

20

30

40

50

60

70

80

90

0 10 20

Are

a (m

2 )

Projectile dose (g ae)


likely artifacts of treating larger (mature) targets or due to inaccuracies in treating smaller targets The total area foot-print of these excessive targets was 103 m2 or 19 of the total target area The total net area footprint was 4938 m2 which was saturated 4355 by the gross projectile foot-print area (ie 31416 m2 times 6845 projectiles) with a net HUR that was 411 of the maximum allowable rate In this case 61 of the total pixel area (ie 3711 m2) ex-ceeded the maximum HUR (fig 7) but was heterogeneous-ly scattered at a sub-meter scale The analyses of HBT-TS data highlight the fine-scale presentation of herbicide ap-plication and the opportunity for high-resolution analyses

CONCLUSION The HBT platform is a precision pesticide application

system customized for incipient species control where op-erational performance has both spatial and temporal rele-vance The second-phase HBT-TS with LRF has substan-tially enhanced the resolution of this data-driven process with the capability to (1) accurately measure treatment time (2) better interpret applicator position and approach to target (3) geometrically calculate target offset location and (4) determine HUR with high resolution While precision is a highlight of the HBT-TS accuracy remains a challenge for all technologies that rely on GPS in a dynamic envi-ronment Future work is proposed to investigate the transi-tion of this technology proven in manned aerial systems toward adoption in unmanned aerial systems

ACKNOWLEDGEMENTS This project was funded in parts by the USDA Forest

Service Special Technology Development Program (Award R5-2012-01) through collaboration with the Hawaii Depart-ment of Land and Natural Resources Forest Health Program the Hawaii Invasive Species Council the USDA Hatch Act Formula Grant Project 112H and the Maui County Office of Economic Development and Department of Water Supply

REFERENCES Berens P (2009) CircStat A MATLAB toolbox for circular

statistics J Stat Software 31(10) httpdxdoiorg1018637jssv031i10

Carlquist S J (1970) Hawaii A Natural History Geology Climate Native Flora and Fauna above the Shoreline Garden City NY American Museum of Natural History

Cox G W (1999) Alien Species in North America and Hawaii (1st ed) Washington DC Island Press

Greenwalt C R (1962) Principles of Error Theory and Cartographic Applications ACIC Technical Report 96 St Louis Mo Aeronautical Chart and Information Center

Harling G Sparre B Andersson L amp Persson C (1973) Flora of Ecuador Gothenborg Sweden University of Gothenborg Department of Systematic Botany

Kueffer C amp Loope L L (2009) Prevention early detection and containment of invasive nonnative plants in the Hawaiian Islands Current efforts and needs Honolulu Hawaii University of Hawaii at Manoa Department of Botany Pacific Cooperative Studies Unit

Leary J J Mahnken B V Cox L J Radford A Yanagida J Penniman T Gooding J (2014) Reducing nascent miconia (Miconia calvescens) patches with an accelerated intervention strategy utilizing herbicide ballistic technology Invasive Plant Sci Mgmt 7(1) 164-175 httpdxdoiorg101614IPSM-D-13-000591

Lindenmayer D B Margules C R amp Botkin D B (2000) Indicators of biodiversity for ecologically sustainable forest management Conserv Biol 14(4) 941-950 httpdxdoiorg101046j1523-1739200098533x

Loope L amp Kraus F (2009) Preventing establishment and spread of invasive species Current status and needs In Conservation of Hawaiian Forest Birds Implications for Island Birds New Haven Conn Yale University Press

Loope L Starr F amp Starr K I (2004) Protecting endangered plant species from displacement by invasive plants on Maui Hawaii Weed Tech 18(1) 1472-1474 httpdxdoiorg1016140890-037X(2004)018[1472PEPSFD]20CO2

Mooney H A amp Drake J A (Eds) (1986) Ecology of Biological Invasions of North America and Hawaii Ecological Studies Vol 58 New York NY Springer-Verlag httpdxdoiorg101007978-1-4612-4988-7

Mueller-Dombois D Bridges K W amp Carson H L (Eds) (1981) Island Ecosystems Biological Organization In Selected Hawaiian Communities Stroudsburg Pa Hutchinson Ross

NOAA (2007) Digital elevation models (DEMs) for the main eight Hawaiian Islands Washington DC National Oceanic and Atmospheric Administration Retrieved from httpsdatanoaagovdatasetdigital-elevation-models-dems-for-the-main-8-hawaiian-islands

Rodriguez R (2015) A custom GPS recording system for improving operational performance of aerially deployed herbicide ballistic technology MS thesis Honolulu Hawaii University of Hawaii at Manoa

Rodriguez R Jenkins D M amp Leary J J (2015) Design and validation of a GPS logger system for recording aerially deployed herbicide ballistic technology operations IEEE Sensors J 15(4) 2078-2086 httpdxdoiorg101109JSEN20142371896

Stone C amp Scott J M (1985) Hawaiirsquos Terrestrial Ecosystems Preservation and Management Honolulu Hawaii University of Hawaii Cooperative National Park Resources Studies Unit

Stone C amp Smith C amp Tunison J T (Eds) (1992) Alien Plant Invasions in Native Ecosystems of Hawaii Management and Research Honolulu Hawaii University of Hawaii Press

Figure 7 Cumulative distribution plot of pixel-scale (00625 m2) HUR for the entire area treated in this study Red dashed line is the maxi-mum allowable HUR

0

5000

10000

15000

000 025 050 075 100

HU

R (

g ae

ha-

1 )

pixelnet area

59(3) 59(3)803-809 809

USBB (1947) United States National Map Accuracy Standards Washington DC US Bureau of the Budget

NOMENCLATURE CMAS = circular mapping accuracy standard EMW = East Maui Watershed

HBT = Herbicide Ballistic Technology HBT-TS = HBT telemetry system HUR = herbicide use rate LRF = laser rangefinder SAS = spherical accuracy standard SD = standard deviation (for a single target) ToT = time on target

2078 IEEE SENSORS JOURNAL VOL 15 NO 4 APRIL 2015

Design and Validation of a GPS Logger Systemfor Recording Aerially Deployed Herbicide

Ballistic Technology OperationsRoberto Rodriguez III Student Member IEEE Daniel M Jenkins and James J K Leary

Abstractmdash Herbicide ballistic technology (HBT) is an electro-pneumatic delivery system designed for administering 173-mmherbicide-filled projectiles (eg paintballs) to visually acquiredweed targets Currently HBT is being deployed from a Hughes500D helicopter platform in aerial surveillance operations toeliminate satellite invasive weed populations in remote naturalwatershed areas of Maui (HI USA) In an effort to improveoperations we have integrated GPS and other sensor hardwareinto the electropneumatic device for instantaneous recording oftime origin and trajectory of each projectile discharged by theapplicator These data are transmitted wirelessly to a customandroid application that displays target information in real timeboth textually and graphically on a map

Index Termsmdash Data acquisition geospatial analysis globalpositioning system sensor integration and fusion

I INTRODUCTION

INVASIVE exotic plant species are threatening the endemicbiological integrity of the Hawaiian archipelago [1]ndash[7]

Moreover the islandsrsquo extreme topography and dense vegeta-tion are an impediment to effective mitigation [8] HerbicideBallistic Technology (HBT) is a concept to address thesechallenges with a novel capability to treat invasive plant targetswith long-range accuracy [9] In previous implementationsthe HBT platform enhances helicopter surveillance operationswith the capability of eliminating high-risk satellite popula-tions of invasive plant targets which are occupying remoteinaccessible areas of watershed [10]

II BACKGROUND

The HBT platform is an herbicide delivery system currentlyregistered in the state of Hawaii as a Special Local Need pes-

Manuscript received September 15 2014 revised November 3 2014accepted November 11 2014 Date of publication November 20 2014 dateof current version January 29 2015 This work was supported in part bythe US Department of Agriculture (USDA) Forest Service in part by theSpecial Technology Development Program under Award R5-2012-01 throughthe Hawaii Department of Land and Natural Resources Forest Health Programin part by the Hawaii Invasive Species Council under Award POC 40466 inpart by the USDA Hatch Act Formula Grant under Project 112H and inpart by the USDA Renewable Resources Extension Act The associate editorcoordinating the review of this paper and approving it for publication wasProf Octavian Postolache

R Rodriguez III and D M Jenkins are with the Department of MolecularBioscience and Biological Engineering University of Hawaii at ManoaHonolulu HI 96822 USA (e-mail roberto6hawaiiedu danieljehawaiiedu)

J J K Leary is with the Department of Natural Resources andEnvironmental Management University of Hawaii at Manoa HonoluluHI 96822 USA (e-mail learyhawaiiedu)

Color versions of one or more of the figures in this letter are availableonline at httpieeexploreieeeorg

Digital Object Identifier 101109JSEN20142371896

ticide for treating nascent miconia (Miconia calvescens DC)and strawberry guava (Psidium cattleianum) patches in remotenatural areas [9] The basic concept of HBT is the encapsula-tion of an active herbicide formulation into 173 mm soft-gelprojectiles (ie paintballs) for accurate long-range delivery toweed target via propulsion from an electro-pneumatic markerProjectiles can be delivered to a target with an estimatedeffective range of sim30m presenting novel opportunities formanaging areas with extreme inaccessible landscapes (egravines and cliff faces)

The adoption of satellite navigation (utilizing GPS andGLONASS) in HBT operations has allowed for the acquisitionof large data sets leading to explicit spatial and temporalperformance evaluation of an operation [10] Currentop erational procedures call for reference points to be recordedfor each target treated using a consumer-grade GPS receivercapturing latitude longitude elevation and timestamp At theend of each operation the applicator also records the estimatedtotal consumption of projectiles that is later translated intothe average dose rate for the total targets treated in thatoperation

Attitude and heading reference systems (AHRS) providethe roll pitch and yaw of a body and are commonly usedfor navigation purposes Recent development in micro-electro-mechanical systems (MEMS) based AHRS systems are beingused in a number of applications including unmanned aerialvehicles (UAV) [11] commercial airplanes [12] along withterrestrial [13] and submerged navigation [14] However manyof these systems have a large profile and require wiredconnections to data recording devices making them ill suitedfor implementation with the HBT platform

To enhance operational data acquisition a custom prototypeHBT-Logging System (HBT-LS) was developed for integrationwith the electro-pneumatic marker automating data acquisitionfor every projectile discharged Data acquired includes theabove mentioned spatial and temporal assignments alongwith added attributes for tilt and azimuth of the markerposition (Fig 1) New analyses of HBT-LS data sets include(i) accurate accounts of projectile dose rates assigned to eachtarget (ii) the time to deliver projectile dose to target and(iii) offset projections of the actual target location

Here in we report on the basic hardware and software spec-ifications of HBT-LS with static and dynamic calibrations incontrolled and operational settings respectively This includesa discussion on improvements in the data management processand technology limitations requiring further development

1530-437X copy 2014 IEEE Personal use is permitted but republicationredistribution requires IEEE permissionSee httpwwwieeeorgpublications_standardspublicationsrightsindexhtml for more information

RODRIGUEZ et al DESIGN AND VALIDATION OF A GPS LOGGER SYSTEM 2079

Fig 1 Spherical coordinate system to describe plant target locations withapplicator as the origin

III SENSOR DEVELOPMENT

A Hardware Design

The relevant data to be recorded include the position (lati-tude longitude and altitude) azimuth and tilt of the markerdepicted in Fig 1 as well as each discharge of a projectileand the unique target IDs and descriptions of each targetedplant Latitude and longitude are recorded within a customAndroid user interface from the built in GPS chip (BroadcomBCM4751) on the interfacing Android device Most non-GPSinformation is recorded on a custom circuit (Fig 2) withindividual sensors communicating to a simple 8-bit micro-controller (ATMEGA328P-MUR Atmel Corp San Jose CA)which transmits data wirelessly to the Android applicationthrough a serial Bluetooth module (RN42 Roving NetworksLos Gatos CA) Altitude is estimated from barometricpressure measurements based on empirical relationships(MPL3115A2 Freescale Semiconductor Austin TX) Azimuthand tilt are recorded from inertial (MPU-6050 InvenSenseSan Jose CA) and magnetic (LSM303DLHTR STMicro-electronics Geneva Switzerland) sensors as described belowProjectile discharge is identified by an interrupt triggered bythe transistor sinking the solenoid current in the marker andsimilarly a momentary contact switch triggers an interruptto indicate acquisition of a new target (for which a newtarget ID is automatically assigned) Descriptive informationfor each target is coded by user selection of radio buttons in theAndroid interface In our case descriptions are coded for thedevelopmental stage of the plant with options for ldquoJuvenilerdquo(not yet bearing fruitsseeds) ldquoMaturerdquo (flowerfruit bearing)or ldquoSurvivorrdquo (showing sub-lethal symptoms resulting froma previous herbicide application) Description reverts to thedefault ldquoJuvenilerdquo for each newly engaged target as these areby far the most commonly encountered plants in our operationsto eliminate incipient satellite populations

Fig 2 HBT-Logger System (HBT-LS) with annotations for major compo-nents (top) and mounted on rail attachment system (bottom)

1) Altitude Altitude is automatically inferred fromobserved barometric pressure (MPL3115A2 Freescale Semi-conductor) based on the following internal empirical relation-ship by setting the ldquoALTrdquo bit on Control Register 1 of thesensor

h = 4433077

[1 minus

(p

p0

)01902632]

+ hoff (1)

where h is the height (in meters) p is the pressure (in Pascals)p0 is the sea level pressure and hoff is the offset in thedetermined height To calibrate for variations in barometricpressure sensors are calibrated at the start of each operationbased on the known elevation at the landing site This knownelevation is used by the calibration activity of the Androiddevice to estimate the current sea level pressure based onthe current barometric pressure and its empirical relationshipwith altitude The new estimate of sea level pressure referenceis communicated to the microcontroller and loaded into theldquoBarometric Pressure Inputrdquo Registers of the MPL3115A2


2) Azimuth and Tilt The azimuth and tilt represented by θand ϕ in a spherical coordinate system (Fig 1) are recordedusing a triple axis gyroscopeaccelerometer (MPU-6050)and a triple-axis tilt-compensated magnetic compass(LSM303DLHTR) Redundant measurements of orientationwere made from these devices to facilitate identification andcompensation for drift in the azimuth estimated from thegyroscope or deviations in compass readings resulting fromlocal disturbances in geomagnetic field or interferences bymagnetic fields originating from or shielded by the aircraft

The calibration activity of the Android interface allowsthe user to overwrite the digitized values stored in deviceEEPROM corresponding to the maximum and minimummagnetic fields applied in each axis of the compass chip(which vary significantly between devices) This allows themeasurements in each axis to be scaled to internally consistentmeasures of the field strength This calibration requires theuser to align each of 6 reference orientations (+x minusx +y minusy +z minusz) of the sensor coordinates with the peakgeomagnetic field at the given location As a result the3-D vector of any arbitrary magnetic field is estimated accu-rately with respect to the marker coordinate system and vectoroperations are implemented by the microcontroller to projectthis vector onto a horizontal plane compensated for tilt androll of the marker estimated from acceleration measurementson the same chip These measurements provide an estimateof the orientation of the marker in the horizontal plane withrespect to magnetic north Marker orientation is estimated fromthe gyroscope readings using custom coded quaternion algebraon the quaternion output retrieved from the digital motionprocessor of the device using the manufacturerrsquos softwarelibrary (MotionAppsTM 50) The quaternion codes for thecurrent orientation of the sensor with respect to the initialorientation at the startup of the sensor

Magnetic declination is determined at the start of eachoperation by orienting the marker along a path aligned withtrue north (or any other given reference orientation) and com-municating this reference orientation to the microcontrollerso that the reference with respect to magnetic north can berecorded in EEPROM On device startup (or recalibration) themarker orientation estimated by the gyroscope projected ontothe horizontal plane is initialized to the current geographicalorientation based on the tilt compensated compass readingcorrected for magnetic declination The marker position(latitude longitude and altitude) azimuth and tilt as depictedin Fig 1 can be used to infer target position based onoverlaying projectile orientation and trajectories onto digitalelevation maps andor using an estimated range to targetrepresented by ρ

3) Trigger Detection The electro-pneumatic markerdischarges projectiles via an actuating solenoid Projectiledischarge is registered by an interrupt caused by a falling edgeof this signal using a custom implemented Schmitt Triggerto prevent multiple interrupts from a single transition andto shift the signal to the microcontroller logic level Thetrigger signal is buffered by a voltage follower to preventloading of the solenoid In order to prevent triggering fromartifacts of resonance in the inductive load associated with

the same trigger pull a latent period is enforced in softwareduring which subsequent falling edges are not registered Thislatent period is greater than the period for which oscillationsassociated with a single projectile discharge occur but issmaller than the period at which projectiles can be dischargedin rapid fire (ie gt10 projectiles secminus1)

B Software Design

1) Calculating Plant Target Coordinates The equations ofmotion for a spherical projectile travelling in air is [15]

m r = mv = minus1

2ρA ACD

∣∣v minus W∣∣(v minus W)+1

2ρA ACL

∣∣v minus W∣∣ [ ω times (v minus W )ω

]+ mg

(2)

where m is the mass of the projectile r is the position ofthe projectile in Cartesian coordinates relative to the originv is the velocity of the projectile relative to the Cartesiancoordinate system ρA is the air density A is the cross-sectional area of the projectile CD is the (dimensionless) dragcoefficient W is the velocity of the wind blowing relative tothe same coordinate system CL is the (dimensionless) liftcoefficient ω is the angular velocity of the projectile andmg is the gravitational force Due to the highly dynamic natureof wind conditions in close proximity to the helicopter theshort range of HBT application and the relatively large initialvelocity of the projectiles a simplification to linear trajectoriesis made for real time calculations However since all raw datais also recorded more rigorous analysis based on the data isalso possible

The azimuth tilt and range describe the location of thetarget relative to the applicator in a local spherical coordinatesystem centered at the applicator In order to superimpose thelocal coordinates of the target over the geodetic coordinatesof the applicator the local spherical coordinates are convertedinto local Cartesian coordinates with a North-East-Downreference frame centered at the applicator using the followingrelationships

x ArarrT = ρ cos ϕ cos θ (3)

yArarrT = ρ cos ϕ sin θ (4)

z ArarrT = ρ sin ϕ (5)

where x y and z (with subscript ArarrT) designate thecomponents of the vector from applicator to target in localCartesian coordinates and ρ θ and ϕ describe the vectorin local spherical coordinates The short range of anHBT application relative to the radius of the Earth allows forsimplified calculations using a local flat surface projectionThe final target offset coordinates are calculated from thefollowing relationships

φT = φA + x ArarrT

REarth(6)

λT = λA + yArarrT

REarth cos φA(7)

hT = h A + z ArarrT (8)


Fig 3 Errors due to assumptions of flat surface projection (black arrows)include discrepancies of the azimuth and longitude that are dependent on thelatitude Blue dashed lines represent the actual ldquolinesrdquo of constant longitudethat converge as latitude increases

where φ λ and h are geodetic coordinates (with subscripts Tfor target or A for applicator) and REarth is the radius of theEarth [16]

The flat surface approximations result in latitude-dependenterrors in the estimated values of the azimuth to and longitudeof the target (Fig 3) Using the prefix ldquordquo to designate theerror in a parameter errors in the target geodetic coordinatescan be estimated from basic uncertainty analysis assuming thaterrors in each measurement are random and independent

φT asympradic

(φA)2 + (ξφ(φA))2 (9)

λT asympradic

(λA)2 + (ξλ(φA)

)2(10)

hT =radic

(h A)2 + (ρ cos ϕ)2 + (ρ sin ϕϕ)2 (11)

where ξ (φA) is the error due to the flat surface approximationas a function of the applicatorrsquos latitude Due to the shortrange of the HBT application relative to the radius of theEarth ξ (φA) is relatively small compared to the horizontaland vertical errors in the GPS receiver For exampleat the approximate latitude of 208deg where these tests wereconducted with a range of 30 m and azimuth at 45deg the valueof ξ (φA) is less than 10minus6deg based on comparison with themore rigorous Vincenty formulae using the WGS-84 modelequivalent to about 01 m [17]

2) Microcontroller Machine code was generated usingthe Arduino IDE (Arduino Torino Italy) and programmedonto the microcontrollers (ATMEGA328P-MUR) using soft-ware (Atmel Studio 61) and programming hardware(AVRISP-mkII) from the manufacturer (Atmel) Duringnormal operation the microcontroller polls all of the sensorsfor new data each cycle of the program Any new data is sent toa paired Google Nexus 7 tablet (Asus Computer InternationalFremont CA) over Bluetooth Each set of data is preceded bya code uniquely identifying the type of data being sent (ieorientation tilt altitude projectile discharge or acquisition ofnew target) so that the Android Bluetooth Service can parsethe incoming data and update the user interface appropriately

Fig 4 User interface showing flight path and trajectories of projectilesfor a single HBT operation (note that the HBT-LS is not connected so thatmarker orientation is not displayed and target and projectile counts are resetto zero) Options available through the options button include connecting ordisconnecting to logger over Bluetooth starting or ending a record of datae-mailing or displaying recorded data on the map and launching otheractivities such as system calibration

In addition each batch of data is followed by a customcheck code to validate correct transmission of data Any datathat does not contain a valid code and a correct check codeis discarded by the Android device without further actionBecause data is continuously streamed to the Android deviceand only the most recent data is recorded to associate withperiodic trigger pulls and trackpoints no attempt to recoverincomplete data strings is made

3) User Interface An Android device is used to recorddata allow the user to send commands (ie for calibration)to the microcontroller and to display device data in realtime both textually and on a map (Fig 4) The customAndroid application was developed using the Eclipse IDEto record GPS coordinates parse incoming Bluetooth datatransmitted from the custom hardware record all data to acomma-delimited file and update the user interface (withtextual and map information) in real-time An options menuon the main activity interface allows the user to connect


Fig 5 Predicted locations and associated CMAS for Foretrex 401 (red) Google Nexus 7 (green) and HBT-LS (blue) Reference antenna location is markedwith a black X at the origin

wirelessly to different custom hardware start and end datalogging display recorded data on the map or launchother activities such as the calibration options describedpreviously

IV STATIC ACCURACY AND PRECISION OF HBT-LS

According to the National Standard for Spatial DataAccuracy (NSSDA) the accuracy of a GPS receiver is deter-mined by root mean square error (RMSE) which is the squareroot of the average of the sum of the squared differencesbetween the coordinate values from a GPS receiver relativeto another more accurate source of the same position used asa reference position [18] [19]

Precision of a GPS receiver in the horizontal x andy directions (aligning respectively with lines of constant lon-gitude and latitude) is described by the circular map accuracystandard (CMAS) which is based on the US National MapAccuracy Standards specifying that ldquono more than 10 ofthe points in the dataset will exceed a given errorrdquo [20]Therefore the CMAS is defined as the magnitude of the radiusof the smallest circle containing 90 of the recorded positions

Fig 6 Accuracy and precision statistics for Foretrex 401 Google Nexus 7and HBT-LS

Assuming normally distributed errors in the horizontal coor-dinates CMAS can be described mathematically as

C M AS = 21460σc (12)


TABLE I

HERBICIDE DOSES (GRAMS OF ACID EQUIVALENT TARGETminus1) DERIVED FROM

HBT-LS COMPARED TO COMPOSITE AVERAGES

Fig 7 Deviation in gyroscope bearing with respect to compass bearings (in degrees) over time for stationary logger unit with sensor z axis (times) or y axis(+) aligned with verticalgravitational force or recorded under dynamic conditions over the course of three separate airborne operations (circles squares andtriangles) Symbols without fill are trackpoint data during aerial operations and red filled symbols are associated with projectile discharges To remove clutterdata was decimated leaving only every 10th data point

where the circular standard error σc can be approximated by

σc sim 1

2

(σx+σy

)(13)

where σx and σy are the standard deviations in horizontalposition corresponding to errors in latitude and longituderespectively assuming that the larger of two is less than 5xthe magnitude of the other [21]

Precision of a GPS receiver in three dimensions includingthe z (vertical) direction is given by the Spherical Accu-racy Standard (SAS) which is defined as the magnitudeof the radius of the smallest sphere containing 90 of therecorded positions This value can be estimated as

S AS = 25σs (14)



σs sim 1

3

(σx+σy+σz

)(15)

where σz is the standard deviation in altitude assuming thatthe smallest of the three component standard deviations is notless than 035x the magnitude of the largest [21]

On July 20 2014 calibrations were performed with theHBT-LS and other consumer-grade GPS receivers against aGPS reference station (Trimble NetRS 21deg 17prime 5046primeprime N157deg 48prime 5895primeprime W 81378m above sea level) located inHonolulu HI The consumer-grade GPS units calibrated inthis study included a Google Nexus 7 tablet independent of theHBT-LS and the Foretrex 401 handheld (Garmin Olathe KS)These GPS units were placed at the base of the referencestation antenna on the roof and their positions recordedevery 5 seconds for 3 hours With the sensor board mountedto the marker the HBT-LS was mounted in an articulatedvise (PanaVise 350 Reno NV) at a horizontal distance ofapproximately 18 meters from and aligned azimuth to tar-get the reference position with azimuth 290deg In this atti-tude data was recorded by the HBT-LS every 5 secondsfor 3 hours Accuracy for all test receivers was determinedby RMSE relative to the known reference position whileprecision was determined by radial magnitudes for CMASand SAS as described above The reference position wasestimated from HBT-LS data using the known distance totarget to completely define the vector from marker to thereference

V DYNAMIC ACCURACY AND PRECISION IN OPERATIONS

Aerial surveillance operations were conducted in 2014targeting Miconia calvescens and Psidium cattleianum weedpopulations on Maui with the HBT-LS integrated to theelectro-pneumatic marker Operations were conducted usinga Hughes 500D helicopter with the applicator positioned onthe port side behind the pilot as described by Leary et al [10]In these operations the HBT-LS recorded all projectilesdischarged with timestamp and target assignment along withall other calibrated sensor information Operational data wereevaluated to determine the distribution of projectile trajectorieswith respect to the nose of the aircraft and the horizontal plane

For a data subset (n = 362) an additional HBT-LS waspositioned at the nose for calculation of the angle betweenthe applicator and aircraft azimuths and bearing for compar-ison with the horizontal target window (ie 270deg to 330deg)corresponding to the shared view of the pilot and applicatorTilt values were compared to the vertical target window(ie 0deg to minus50deg) Altitude above ground level (Altagl) wasestimated as the difference between the elevation inferredfrom the barometric pressure readings and the raster pixelvalue of the digital elevation model (DEM 10m resolution)corresponding to the recorded point location

Herbicide dose accuracy was measured as the percentdifference between mean target dose rate recorded by HBT-LS(average number of projectiles used to treat each uniquetarget ID) and composite averages in each operation estimatedby dividing the total number of pods (projectile containers)

Fig 8 Deviation in tilt estimate from gyro with respect to tilt estimatefrom compass for data in a single aerial operation Red filled symbols aredata associated with a projectile discharge and unfilled symbols are fromtrackpoint data Data are decimated to show only every 10th data point toremove clutter

consumed (sim160 projectiles per pod) by the total number oftargets treated in an operation with rounding errorsexpected [9]

VI RESULTS

A Accuracy and Precision of Target Coordinates

Both HBT-LS and tablet exhibited static circular and spher-ical precisions about seven times smaller than the corre-sponding values recorded for the handheld GPS unit (Fig 5)Accuracy (RMSE) of the tablet was also superior to the otherGPS unit in 2D and 3D while the HBT-LS was superior tothe handheld GPS in 3D accuracy (Fig 6) Moreover it isworth noting how the BCM4751 receivers of the HBT-LSand tablet were coinciding with much more stable plots ofthe coordinates in the short three hour period compared tothe widely dispersed random appearance of the handheld GPS(see Fig 6) The HBT-LS points are shifted further from thereference point which is likely due to compounding errorsexhibited by the tilt and azimuth sensors used for determiningthe target location and approximations made in installation andoffset calculation

B Operational Validation

The HBT-LS successfully acquired data from seven inde-pendent operations in 2014 for a total of 11132 projec-tiles assigned to 400 targets Android tablets never becamedisconnected textual data displayed on the app was con-stantly refreshed during operations and the system was neverobserved to miss a trigger pull during operation suggestingthat the data communication was robust By assigning targetIDs to each projectile with the HBT-LS we were able toqualitatively identify the target dose outliers (ie a large target


Fig 9 Altitude above ground at the point of application estimated from elevation measurements from the HBT-LS (n = 11 724) and Foretrex 401 (n = 11 745)with ground level inferred from a 10-m digital elevation model at the recorded location

treated with large projectile quantity) and also calculate meanvalues with standard deviations for each operation (Table I)In total the HBT-LS mean (plusmn standard deviation) calcula-tions were within 11 of the composite estimated valuesAnecdotally herbicide dose corresponds to weed target sizeso that HBT application data can be useful for assessingpopulation age and phenology across the geographical rangeof operations The standard deviations observed in this data setwere rather large relative to the mean suggesting diverse targetpopulations The capability of generating standard deviationswill enhance interpretations with measures of uniformity andfor segregating target outliers based on size class

Efforts to determine the accuracy of the azimuth offsetrelative to the target window deduced from the track vectorbearing were prone to large errors particularly when theaircraft was in a stationary hover and pivoting on its axis(data not shown) such that the track direction did not coincidewith the aircraft orientation Despite a large interquartile rangethe most frequent azimuth offset estimated for applicatorrelative to aircraft nose under these conditions was 311degVertical tilt estimates on the other hand were independentof aircraft bearing and so had a much tighter interquartilerange of minus20deg to minus41deg The more limited azimuth data subsetrelative to aircraft orientation recorded simultaneously by aseparate logger in the cockpit displayed an interquartile rangefrom 292deg to 335deg contained within the perceived targetwindow

The average time on target (time interval between the firstand last projectile discharged) was 123 plusmn 98 s

To determine the stability of the gyroscope readings ofazimuth and tilt (relative to those from the digital compassreadings) were recorded both under static and dynamic con-ditions and plotted against time (Figs 7 and 8 respectively)

Stationary data indicated that gyroscope readings have smallaxis-dependent biases in rotational velocity which doesnot completely explain the relatively large systematic driftobserved during aerial operations Furthermore the sys-tematic drift experienced during aerial operations is notexplained by random integration errors in the digital motionprocessor on the MPU-6050 This suggests that vibra-tion or other motion artifacts might superimpose significantbiases onto the rotational velocity measurements around thevertical axis for example by inducing sympathetic vibrationsin the gyroscope oscillators at odd harmonics The devia-tion in tilt is not subject to this systematic drift as thedigital motion processor in the gyroscope chip corrects fordrift in the tilt measurements based on acceleration (gravity)measurements in the different axes No systematic drift wasobserved in stationary measurements of the digital compassIn addition compass bearing to targets in operational datacorresponded closely to expectations in that application isprimarily against weed targets populating cliffsides so thattrajectories were more or less perpendicular to contoursof terrain maps in the direction of increasing elevationThese data suggest that the aircraft did not significantly distortor shield local geomagnetic fields and that at least in the areasof our operations there are little if any local distortions in thefield

The Altagl values (applicator elevation estimated from baro-metric pressure relative to ground surface elevation estimatedfor the corresponding GPS position on 10m digital elevationmodel) estimated from operational data recorded from theHBT-LS and consumer-grade GPS were highly variable andincluded some negative values (Fig 9) It should also be notedthat some dramatic weather changes occurred during several ofthe operations which likely confounded many of the altimeter


estimates based on the calibrated barometric pressure at thestart of operations and correspondingly contributed to errorsin Altagl

VII CONCLUSION

The HBT-LS provided satisfactory accuracy and precisionrelative to consumer-grade handheld GPS units currently inuse enhanced data acquisition with new spatial and quantifi-able attributes while maintaining a low profile relative to themarker Testing in an operational setting showed that resultsare consistent with anecdotal observation and expectationsImprovements in altimeter recordings may be necessary forbetter determination of offset locations to targets Implemen-tation of laser rangefinders to determine the range to theplant target and altitude will further improve the utility ofthe logger and enable direct estimation of target locationwithout requiring knowledge of elevation and reference todigital elevation maps Use of a single chip 9 axis sensor andadditional sensor fusion between the accelerometer gyroscopeand magnetic compass for improved bearing determinationshould also be implemented While this system was developedfor use with the HBT platform alternative applications includethe use of marking agents to track the positions of plantsand animals including invasive species or livestock surveyingdifficult to reach terrain and spatial analysis of video and stillimagery recorded by UAVs

ACKNOWLEDGMENT

The authors would like to thank Dr J Foster and the PacificGPS Facility for allowing use of their GPS reference stationfor accuracy and precision experiment

REFERENCES

[1] G W Cox Alien Species in North America and Hawaii 1st edWashington DC USA Island Press 1999

[2] L Loope and F Kraus ldquoPreventing establishment and spread of inva-sive species Current status and needsrdquo in Conservation of HawaiianForest Birds Implications for Island Birds New Haven CT USAYale Univ Press 2009

[3] L Loope F Starr and K Starr ldquoProtecting endangered plant speciesfrom displacement by invasive plants on Maui Hawaiirdquo Weed Technolvol 18 pp 1472ndash1474 Dec 2004

[4] H A Mooney and J A Drake ldquoEcology of biological invasionsof North America and Hawaiirdquo in Ecological Studies Analysis andSynthesis Berlin Germany Springer-Verlag 1986

[5] C P Stone and J M Scott Hawaiirsquos Terrestrial Ecosystems Preserva-tion and Management 1st ed Honolulu HI USA Univ Hawaii Press1985

[6] C P Stone C W Smith and J T Tunison Alien Plant Invasions inNative Ecosystems of Hawaii Management and Research HonoluluHI USA Univ Hawaii Press 1992

[7] D B Lindenmayer C R Margules and D B Botkin ldquoIndicators ofbiodiversity for ecologically sustainable forest managementrdquo ConservBiol vol 14 no 4 pp 941ndash950 2000

[8] C Kueffer and L Loope ldquoPrevention early detection and containmentof invasive nonnative plants in the Hawaiian Islands Current effortsand needsrdquo Dept Botany Univ Hawaii at Manoa Honolulu HI USATech Rep 166 Aug 2009

[9] J J K Leary J Gooding J Chapman A Radford B Mahnkenand L J Cox ldquoCalibration of an herbicide ballistic technology (HBT)helicopter platform targeting Miconia calvescens in Hawaiirdquo InvasivePlant Sci Manage vol 6 no 2 pp 292ndash303 Jan 2013

[10] J J K Leary et al ldquoReducing nascent miconia (Miconia calvescens)patches with an accelerated intervention strategy utilizing herbicideballistic technologyrdquo Invasive Plant Sci Manage vol 7 no 1pp 164ndash175 Jan 2014

[11] J S Jang and D Liccardo ldquoAutomation of small UAVs using a low costMEMS sensor and embedded computing platformrdquo in Proc IEEEAIAA25th Digital Avionics Syst Conf Oct 2006 pp 1ndash9

[12] M Steen P M Schachtebeck M Kujawska and P Hecker ldquoAnalysisand evaluation of MEMS INSGNSS hybridization for commercialaircraft and business jetsrdquo in Proc IEEEION Position Location NavigatSymp (PLANS) May 2010 pp 1264ndash1270

[13] J Kolecki and P Kuras ldquoLow cost attitude and heading sensorsin terrestrial photogrammetrymdashCalibration and testingrdquo ArchiwumFotogrametrii Kartografii Teledetekcji vol 22 pp 249ndash260 2011

[14] G Troni and L L Whitcomb ldquoPreliminary experimental evaluationof in-situ calibration methods for MEMS-based attitude sensors andDoppler sonars in underwater vehicle navigationrdquo in Proc IEEEOESAuto Underwater Veh (AUV) Sep 2012 pp 1ndash8

[15] G Robinson and I Robinson ldquoThe motion of an arbitrarily rotatingspherical projectile and its application to ball gamesrdquo Phys Scriptavol 88 no 1 p 018101 Jul 2013

[16] B L Decker ldquoWorld Geodetic System 1984rdquo Defense Mapping AgencyAerospace Center St Louis Mo USA Accession No ADA167570Apr 1986

[17] T Vincenty ldquoDirect and inverse solutions of geodesics on the ellipsoidwith application of nested equationsrdquo Surv Rev vol 23 no 176pp 88ndash93 1975

[18] Federal Geographic Data Committee ldquoNational standard for spatial dataaccuracy (NSSDA)rdquo Federal Geographic Data Committee WashingtonDC USA Tech Rep FGDC-STD-0073-1998 1998

[19] Information TechnologymdashSpatial Data Transfer StandardANSI-NCITS Standard 3201998 Jun 1998

[20] United States National Map Accuracy Standards US Bureau BudgetWashington DC USA 1947

[21] C R Greenwalt and M E Shultz Principles of ErrorTheory and Cartographic Applications St Louis MO USAAeronautical Chart and Information Center 1965

Roberto Rodriguez III (Mrsquo14) received the BS degree in biologicalengineering from the University of Hawaii at Manoa Honolulu HI USAwhere he is currently pursuing the MS degree in biological engineering

He has been a Research Assistant with the Department of MolecularBiosciences and Biological Engineering University of Hawaii at Manoa since2013 and a member of the American Society of Agricultural and BiologicalEngineers since 2014 His current research interests include the use of remotesensing for environmental and medical applications

Daniel M Jenkins received the BS and MEng degrees in agricultural andbiological engineering from Cornell University Ithaca NY USA in 1995and 1996 respectively and the PhD degree in biological and agriculturalengineering from the University of California at Davis Davis CA USA in2001

He has been a faculty member with the Department of MolecularBiosciences and Bioengineering University of Hawaii at Manoa HonoluluHI USA since 2002 a member of the American Society of Agricultural andBiological Engineers since 1998 and a member of the American ChemicalSociety since 2005 His primary interest is in molecular and other sensorsystems for agricultural and environmental applications

James J K Leary received the BS degree in horticulture with a minorin chemistry from Michigan State University East Lansing MI USA in1996 and the MS degree in horticulture and the PhD degree in molecularbiosciences and biological engineering with a specialization in weed scienceand molecular ecology from the University of Hawaii at Manoa HonoluluHI USA in 1999 and 2006 respectively

He has served as a faculty member with the Department of NaturalResources and Environmental Management University of Hawaii at Manoasince 2009 where he ranked as an Assistant Specialist with a research andextension split appointment His interests are applied research in invasive plantspecies management

Reducing Nascent Miconia (Miconiacalvescens) Patches with an AcceleratedIntervention Strategy Utilizing Herbicide

Ballistic TechnologyJames Leary Brooke V Mahnken Linda J Cox Adam Radford John Yanagida Teya Penniman David C Duffy and

Jeremy Gooding

The miconia (Miconia calvescens) invasion of the East Maui Watershed (EMW) started from a single introduction over

40 yr ago establishing a nascent patch network spread across 20000 ha In 2012 an accelerated intervention strategy was

implemented utilizing the Herbicide Ballistic Technology (HBT) platform in a Hughes 500D helicopter to reduce target

densities of seven nascent patches in the EMW In a 14-mo period a total of 48 interventions eliminated 4029 miconia

targets with an estimated 33 increase in operations and 168 increase in recorded targets relative to the adjusted means

from 2005 to 2011 data (prior to HBT adoption) This sequence of interventions covered a total net area of 1138 ha

creating a field mosaic of overlapping search coverage (saturation) for each patch (four to eight interventions per patch)

Target density reduction for each patch fit exponential decay functions (R2 088 P 005) with a majority of the

target interventions spatially assigned to the highest saturation fields The progressive decay in target density led to

concomitant reductions in search efficiency (min ha21) and herbicide use rate (grams ae ha21) in subsequent

interventions Mean detection efficacy (6 SE) between overlapping interventions (n 5 41) was 062 6 003 matching

closely with the probability of detection for a random search operation and verifying imperfect (albeit precise) detection

The HBT platform increases the value of aerial surveillance operations with 98 efficacy in target elimination Applying

coverage saturation with an accelerated intervention schedule to known patch locations is an adaptive process for

compensating imperfect detection and building intelligence with spatial and temporal relevance to the next operation

Nomenclature Miconia Miconia calvescens DC

Key words Adaptive management aerial surveillance GIS nascent patch network random search operation

mortality factor

An exotic plant invasion is a phenomenon carried outby species with the capability to occupy niches alreadyinhabited by other indigenous or endemic plant commu-

nities (Richardson et al 2000) This invasion phenomenonis also a main driver in habitat fragmentation leading tomodification of an ecosystemrsquos structure and function(Gilbert and Levine 2013) The resultant loss of endemicbiological diversity is particularly detrimental for isolatedisland ecosystems that exhibit high endemism (Denslow2003 Mack et al 2000 Reaser et al 2007)

A commitment to weed species eradication starts with aneffective containment strategy targeting the nascent patchnetwork that is expanding the invasion front (Cousensand Mortimer 1995 Moody and Mack 1988 Rejmanekand Pitcairn 2002) Reconnaissance and surveillance arefundamental activities for nascent weed detection andspatial delimitation (Baxter and Possingham 2011 Cachoet al 2007 Hester et al 2010 Lawes and McAllister 2006Panetta and Lawes 2005 Rew et al 2006 Taylor andHastings 2004) Reconnaissance tends to be a more cursory

DOI 101614IPSM-D-13-000591

First third and fifth authors Assistant Specialist Specialist and

Professor Department of Natural Resources and Environmental

Management University of Hawaii at Manoa Honolulu HI 96822

second fourth and sixth authors GIS Specialist OperationsManager and ManagerMaui Invasive Species Committee University

of Hawaii at Manoa Honolulu HI 96822 seventh author Director

Pacific Cooperative Studies Unit University of Hawaii at Manoa

Honolulu HI 96822 eighth author Liaison Pacific Islands Exotic

Plant Management Team National Park Service Biological Resource

Management Division Current address of first author Maui

Agricultural Research Center PO Box 269 Kula HI 96790

Corresponding authorrsquos E-mail learyhawaiiedu

Invasive Plant Science and Management 2014 7164ndash175

164 N Invasive Plant Science and Management 7 JanuaryndashMarch 2014

process of searching for new target locations whereassurveillance is distinguished by a more intensive process ofbuilding explicit intelligence on known target locationswith repeated visits (Anonymous 2007) In the early stagesof strategy development reconnaissance may be a necessarytrade-off to actual treatment activities when a lack ofintelligence compromises a sound priority decision process(Baxter and Possingham 2011) Otherwise surveillance is amore common practice of gathering intelligence usually asan ad hoc activity complementary to active managementand relegated to within the immediate vicinity of themanagement area (Cacho et al 2007 Fox et al 2009)Detection of invasive plant species residing in their naturalsettings is an imperfect process (Moore et al 2011 Reganet al 2011) Good intelligence regardless of the surveydesign should account for imperfect detectability as anadaptive process for optimizing future operations (Mooreet al 2011 Regan et al 2011 Thompson 2002)

Cacho et al (2007) coined the term lsquolsquomortality factorrsquorsquo todescribe management of individual weed targets accountingfor both detection and effective treatment as complementaryactions necessary to achieving target elimination With aneffective treatment technique detection then becomes thedeterminant outcome of an operation (Leary et al 2013Lodge et al 2006) Koopman (1946) introduced themathematical framework for estimating the probability ofdetection (Pd) in a random search operation that would be

expected within a terrestrial environment (Cooper et al2003) Search impediments (ie topography weathersurrounding vegetation) imposing even slight randomnessin coverage (c) have been shown to fit the exponentialdetection function (Equation 1) as a conservative estimate ofan imperfect search effort (Cacho et al 2007 Frost 1999)

pd~1ec frac121Herbicide Ballistic Technology (HBT) is a weed target

intervention platform that was developed in Hawaii toenhance helicopter surveillance operations with the capa-bility to dispatch satellite weed targets upon detection (Learyet al 2013) The HBT platform discretely administers small-aliquot herbicide projectiles through a pneumatic applica-tion device to treat individual weed targets with long-rangeaccuracy (ie 30 m [98 ft] effective range) from horizontalor vertical trajectories It expands the capability to treat alldetectable targets that might otherwise be untreatable withother conventional application methods An effective HBTtreatment application enhances surveillance operations withthe full complement of a mortality factor (Leary et al 2013)

The term lsquolsquointerventionrsquorsquo is used here to distinguish fromother surveillance operations with the inclusion of a mortalityfactor accommodated by the integration of an HBTplatform This study expands on platform calibrationsdescribed by Leary et al (2013) with the accelerateddeployment of an adaptive intervention strategy targetingthe primary nascent patch network of the miconia (Miconiacalvescens DC) invasion in the East Maui Watershed (EMWHawaii) We report on spatial and temporal parameters ofperformance measures associated with patch target densityreductions incurred by these intervention sequences

Materials and Methods

Target Species Miconia calvescens Miconia is a midstorytree 12 to 15 m tall native to Central and South AmericaIt was introduced to the Hawaiian Islands in 1961 and waspresumably introduced to the island of Maui by 1970where the first management program in the state wasinitiated 20 yr later (Chimera et al 2000 Medeiros et al1997) Miconia is an autogamous species that reachesmaturity in 4 to 5 yr A single plant has immense fecundityproducing millions of propagules in a single reproductivecycle (Meyer 1998) Its fruit are small and edible lendingitself to frugivorous dispersal by a diverse avian community(Chimera et al 2000 Spotswood et al 2013) In Australiathe dispersal range of a nascent patch had been estimatedto exceed 2000 m but with 95 of the recruitmentcontained within 500 m of the maternal source (Hardestyet al 2011 Murphy et al 2008) Viable seed persistencehas been measured beyond 16 yr (Meyer et al 2011)contributing to the long-term recruitment potential of alatent seed bank

Management ImplicationsThe herbicide ballistic technology (HBT) platform is a

herbicide delivery system registered as a 24(c) Special LocalNeed for use in Hawaii to treat nascent satellite miconia patches inremote natural areas that require helicopter operations We definean intervention as a weed management operation with thecombined actions of target detection immediately followed byeffective elimination In 14 mo a total of 48 interventions wereconducted for seven nascent patches (1) eliminating 4029miconia targets and (2) encompassing a total net area of1138 ha while (3) administering 1 of the maximumallowable herbicide use rate Target density reduction for the patchnetwork was 86 fitting an exponential decay function Thisresulted in a threefold improvement in search efficiency (min ha21)and a 10-fold reduction in herbicide use rate (g ae ha21)Expansion of the total net area corresponded to improvements insearch efficiency less time was needed to saturate known targetlocations with reduced densities These interventions are describedas random search operations with imperfect detection Efficacy ofan HBT application to miconia can be confirmed in less than1 mo allowing for the acceleration of intervention schedules thatcan compensate for imperfect detection by saturating coveragewith compounding sequential interventions The goals of thisaccelerated intervention strategy are to rapidly reduce nascentpatch populations to manageable or undetectable levels and buildspatially and temporally explicit intelligence where expectedprojections are accurate with observed values

Leary et al Reducing nascent miconia with HBT N 165

Suitable Habitat The EMW The EMW encompassesover 55000 ha (136000 ac) on the windward slope ofHaleakala Crater on Maui The core miconia infestationoccupies 1000 ha on the easternmost portion of EMW(20u479440N 156u009520W) while the spatial distribu-tion of the nascent patch network is spread across anestimated 20000 ha (Figure 1) The invasion is below1000 m above sea level (masl) with invasion frontsadvancing south 8 km (5 mi) and west 18 km from thecore infestation (CORE) respectively The mean annualprecipitation range within this area is 3000 to 6000 mm(118ndash236 in) (Giambelluca et al 2013) The mean annualtemperature from Hana Airport (24 masl) is 234uC (741F 1950 to 2005 WRCC 2013) with a lapse rate of2064uC per 100 m elevation (Jacobson 2005) Theseclimatic conditions are consistent with the description of asuitable habitat in French Polynesia (Pouteau et al 2011)The dominant vegetation types include a wet coastal forestwith mixed exotic canopy species (eg African tulip tree[Spathodea campanulata P Beauv] eucalyptus [Eucalyptusrobusta Sm] guava [Psidium spp]) that transitions to a wetmontane forest at higher elevations represented byendemic assemblages (eg ohirsquoa [Metrosideros polymorphaGaudich] and koa [Acacia koa Gray]) (Wagner et al 1999)that also serve as critical habitat for 59 other threatened andendangered plant species (Anonymous 2000) Miconia iscapable of invading EMW critical habitat above 1000 masl

(Medeiros et al 1997 Meyer 1996 Meyer and Florence1996 Pouteau et al 2011) making mitigation of its spreadinto these areas a strategic priority

Over the last decade systematic surveys of the entirewatershed have identified all major nascent patch popula-tions along with hundreds of incipient satellites Thesepatches are comparably smaller and phenologically distin-guishable from the CORE with plant maturity observedless frequently They are most often identified as juvenileunderstory saplings (ca 3 m tall) that may be aggregatinginto monotypic midstory canopies These patches also tendto colonize the extreme topography associated with theravines and drainages of EMWrsquos heterogeneous landscape(see Pouteau et al 2011 for similar topographic features)

Experimental Units The Miconia Patch Network Sevenof the most prominent patches within the network wereselected to test an accelerated intervention schedule (seebelow) Nuuailua (NULU) Waiakuna Pond (WAKU)Keanae Wall (KANE) Wailua nui (WALU) Wailua iki(WALI) Hanawi (HAWI) and Waihiumalu (WAHI)(Figure 1) Five of the patches (NULU WAKU KANEWALU and WALI) were located along the western frontof the invasion 12 to 18 km from the CORE Each of thewestern-front patches was within a 1000-m radius of thenext adjacent patch creating a viable interpatch networkfor frugivorous dispersal (Hardesty et al 2011 Murphyet al 2008) The HAWI patch was approximately 3000 mfrom the nearest patch (WALI) However several targetsinterspersed between these patches had been dispatched inseparate operations that were not a part of this study TheWAHI patch was located on the southern front of theinvasion and was closest to the CORE (ca 8 km fromcenters) One other known patch was just south of WAHIand was also being managed independently of this study

The Accelerated Intervention Schedule The strategyimplemented for this study consisted of an accelerateddeployment schedule of sequential interventions A total of48 interventions were conducted on eight separate datesover a 14-mo period with all patches receiving at leastfour interventions and three of the seven receiving eightinterventions (Table 1) The time intervals between inter-ventions ranged from 18 to 182 d with the medianmodeinterval at 32 d Based on previous recorded operationsconducted in these patch areas (2005 to 2011) theintervention schedules for this study were accelerated by33 (Supplemental Table 1 httpdxdoiorg101614IPSM-D-13-00059TS1) All helicopter operations wereconducted with a full fuel load providing 70 to 100 min ofoperation flight time Starting locations were typically at thelowest elevation point for each patch proceeding with anadaptive process of deliberating search efforts to knowntarget locations based on intelligence derived from previousinterventions (Fox et al 2009 Thompson 2002) Anyremaining flight time would accommodate expansion ofsearch coverage to surrounding adjacent areas typically intohigher elevations Flight lines were anisotropic running

Figure 1 Map of East Maui Watershed with mean center ofthe miconia core infestation (CORE black diamond) and thefinal net areas of the patch units (shaded buffers withcorresponding four-letter acronyms black) See ExperimentalUnits in the Materials and Methods section 300-m contoursdisplayed on the base map


along the contours of the steep topography and withiso-altitudinal transects spaced 50 m of elevation apart forensuring visual search overlap from valley floor to ridgeline

Helicopter HBT Platform and Intervention ProceduresAll aerial operations were conducted in a Hughes 500Dhelicopter by a three-person crew with a portside biascreating a 210u field of view and a detectable sight rangeestimated at 50 m Operation speeds were maintained at 20 km h21 while actively searching Miconia targetswere treated with 173-mm soft-gel projectiles encapsulat-ing 1994 mg ae (0007 oz) of the active herbicideingredient triclopyr Applicators administered three to fiveprojectiles to each stem axial point (eg larger targets withmore axial points received higher doses) The applicatorseated portside behind the pilot administered an effectiveherbicide dose to target within a 30-m effective rangeOperation data were recorded with a FortrexH 301 globalpositioning system (GPS) data logger (GarminH OlatheKS) with flight lines recorded as 5-sec vertex track logs anddispatched targets recorded as waypoints which wereactually recorded from the position of the application andnot the actual location of the target (ie within 30 m) Thisprotocol was consistently applied to all recorded points dueto flight safety considerations Projectile consumption wasdetermined at the end of each operation For further detailson these protocols refer to Leary et al (2013)

Geographic Information System Data Analyses All GPSdata were projected in the NAD 1983 UTM Zone 4Ncoordinate system and processed in ArcMAPH 101 (EsriHRedlands CA) Tracks logs were manually spliced intooperation flight segments that were 16 km h21 Search

coverage areas were calculated from these operation flightsegments using buffer tool procedures with a full 50-mradius and dissolved overlap Operation times werecalculated by tabulating the number of vertices in thesesame flight segments (eg 12 vertices min21) All buffersfor each patch were superimposed with a union toolprocedure creating a field mosaic of the cumulative netarea Coverage saturation levels for each field were scoredby the number of overlapping buffers As an example anarea mosaic created by eight interventions contained fieldsrepresenting all levels of saturation (ie 1 to 8) All targetwaypoints were spatially joined to their respective bufferunions for saturation field assignment

Operation data sets consisted of (1) area (2) time (3)targets and (4) projectile consumption which were used tocalculate target density (targets ha21) search efficiency(min ha21) and herbicide use rate (g ae ha21) The dataand calculations were each assigned to three separatesubcategories1 momentary intervention assignment (momentary)mdashan

independent intervention data set with no saturationfield assignments generating gross area and operationperformance values of the moment

2 cumulative intervention assignment (cumulative)mdashacompilation of sequential data sets scaled by the numberof interventions with progressive assignments made to achanging saturation field mosaic generating net areaand compounded operation performance values and

3 historic intervention assignment (historic)mdashall sequen-tial data sets were retroactively assigned to the finalsaturation field mosaic generating gross area andoperation performance values for each momentaryoperation in retrospect

Table 1 Intervention data recorded for the seven nascent miconia patches

Patch Interventiona Gross areab (ha) Net areac (ha) Targets OFTd (min) PCe

HAWI 4 3020 1308 471 5741 14698WAHI 6 2758 944 465 5743 11431WAKU 7 3291 1940 412 5628 12167WALI 7 4781 1750 401 6849 14127WALU 8 4129 1544 676 7647 17174NULU 8 5867 1746 508 7858 14294KANE 8 6888 2149 1096 11698 31674NETWf 48 30734 11381 4029 51164 115565

a Interventions conducted in February 2012 May 2012 June 2012 October 2012 November 2012 December 2012 February2013 March 2013 and April 2013

b Momentary gross areac Cumulative net aread Abbreviations OFT operation flight time PC projectile consumption HAWI) Hanawi WAHI Waihiumalu WAKU

Waiakuna Pond WALI Wailua iki WALU Wailua nui NULU Nuuailua KANE Keanae Wall NETW networke Projectile consumption estimated from the pod inventory (ca 140 projectiles pod21)f Total values for the entire patch network


Momentary data sets were used to calibrate searchefficiency and herbicide use rate corresponding to encoun-tered target densities from independent flight segments (n 5103) contributing to the sequential interventions (n 5 48)

Cumulative data sets were used to calculate weighted mean

saturation ( Csatn ) from the net area field mosaic with thesum of each (ith) saturation level adjusted by the respectiveproportion of assignments divided by the total fieldassignments within the entire mosaic (Equation 2) In thisstudy xi includes either area or target assignments with thehighest (ith) level of saturation equivalent to (n) interventions

Csatn ~Xn

i~1ixieth THORNX

xi frac122

Cumulative target densities (Ctdn) were adjusted by dividingthe compounded values with their respective number (n)of contributing interventions (Equation 3) The influenceof sequential interventions on the adjusted target density wasfit with an exponential decay function for each patch(Equation 4)

Ctdn~(Can=Ctn)=n frac123

f (x)~yoeln frac124

Momentary and historic target densities only representoutcomes of the moment which is dependent on the areasearched and targets detected within that interventionCumulative target densities on the other hand account fortargets dispatched in all previous interventions for the entirenet area providing a more spatially accurate presentation ofprogress

Detection efficacy (de) as described by Leary et al(2013) is the ratio of targets (x) recorded betweensequential interventions (n) spatially assigned to the samesaturation fields (Equation 5)

de~xn=xnzxnz1 frac125Retroactive proportions of cumulative target and net

area totals (Cpropx) of each intervention were calculatedfrom the final total (Equation 6)

Cpropx~Cxn

Cxfinal frac126

Cumulative target proportions were plotted against theircorresponding coverage (c) which was calculated as theweighted mean net area saturation (Equation 4) multipliedby the cumulative net area proportion for each intervention(Equation 7)

c~ Csatn|Cpropn frac127

Coverage of the last interventions (Cpropx 5 1) of eachpatch were equal to the highest weighted mean saturationlevels These values were plotted with search theory

concepts postulating probabilities of detection (Pd) for arandom search operation (Equation 1) (Koopman 19461980) and also a theoretically perfect lsquolsquodefinite rangersquorsquosensor (Equation 8) where Pd is proportional up tocomplete coverage (c 5 1) resulting in perfect detectionthat cannot be exceeded with added coverage

Limc1

(Pdc) frac128

Polynomial linear and two-parameter exponential decayfunctions were used to fit momentary and cumulativesequential data plots respectively using SigmaPlotH(version 120 (Systat Software Inc San Jose CA)

Results

In a 14-mo period starting in February 2012 a total of48 interventions were administered to the seven patchescovering 11381 net ha and dispatching 4029 targets(Table 1) In total 51164 min of operation flight time(ca 853 h) and 115565 projectiles were utilized toaccomplish these results HAWI was the last patch onwhich the accelerated strategy was initiated and it wasadministered only four interventions whereas WALUNULU and KANE were administered twice as manyExcept for WAHI all patch net areas were 100 ha withKANE being the largest at 2149 ha which also recordedthe highest number of dispatched targets (ie 1096targets) Effective applications of previous interventionswere visually confirmed even for the shortest time interval(ie 18 d) Only 88 survivors were retreated indicating98 efficacy with the HBT platform (data not shown)

Linear fits for search efficiency and herbicide use ratewere highly significant when plotted against target densities

Figure 2 A scatter plot with best-fit lines and 95 confidenceintervals (dashed lines) for surveillance efficiency (dark circlesmin ha21 P 0001 R2 5 078) and herbicide use rate (opendiamonds g ae ha21 P 0001 R2 5 091) corresponding toencountered target densities (targets ha21) derived frommomentary operation flight segments (n 5 103)


encountered (P 0001 n 5 103) (Fig 2) Searchefficiency was measured at 090 min ha21 dispatch time at053 min target21 and herbicide dose was estimated at 32projectiles target21 All herbicide use rates were 1 ofthe maximum allowable rate (ie 6720 g ae ha21) Thehigh coefficients of determination illustrated a strongdependence to target density and validated platformconsistency with previously published results (Leary et al2013) despite the use of multiple pilotapplicator teamsacross seven different sites

All patches displayed reductions in cumulative targetdensities (adjusted by the number of interventions)following their respective intervention sequences Allreductions were significantly fit to exponential decayfunctions (P 005 R2 088 coefficient of variationfor the root mean square residuals between observed andfitted values 0393) despite the variability of initial targetdensities (14 to 70 target ha21) and intervention sequences(four to eight interventions) (Figure 3 Table 2) Twogeneral observations were made (1) the sharpest reductionsin target density occurred within the first three interventionsand (2) targets were dispatched in every operation Similarreductions were also observed for momentary target densitysequences but with greater peak deviations from the best-fitdecay functions (data not shown) Cumulative targetdensities are comprehensive values of the entire patch withreduction observed when target density of the last

intervention was less than the net value of all previousinterventions These decay functions were influenced by adiminishing number of targets recorded in the latterinterventions with the curve asymptotically approachingzero which will continue even beyond the point ofeventually recording momentary undetectable levels

Mean momentary search efficiency and herbicide userate were reduced with each sequential interventioncorresponding to lower target densities encountered(Figure 4) Search efficiency is optimized when no targetsare detected (Leary et al 2013) As target density reducedto 1 target ha21 (ie operations 7 and 8) mean searchefficiency was approximate to the calculated y-interceptcoefficient (see Figure 2) suggesting some discrepancybetween these independent derivations Herbicide use rateshowed a similar reduction trend with the final mean valueequivalent to 006 of the maximum allowable use rateThus the efficacy of previous interventions contributed toimproved performance efficiencies for subsequent inter-ventions on these accelerated schedules

Mean net area increased with every sequential operation(Figure 5) This is due in part to improved searchefficiencies making operational flight time available forexpanding area coverage (see Figure 4) Net area expansioninfluenced the exponential decay of the cumulative targetdensities with fewer targets being dispatched in thesesurrounding areas

Figure 3 Adjusted cumulative target densities (targets ha21 solid black lines) of each patch and the entire network (6 SE n 5 7 forone to four interventions n 5 6 for five or six interventions n 5 5 for seven interventions and n 5 3 for eight interventions) with best-fitexponential decay functions (black dot-dash lines P 005 R2 088) Notice KANE with a larger target density scale on the y-axis


Mean saturation for net area and target assignmentsincreased with each sequential operation (Figure 6) Thepositive trends are an indication of the strategyrsquos adaptivequality for revisiting known target locations The rate ofcumulative target saturation advanced nearly three timesfaster than net area saturation as determined by slopecoefficients of best-fit linear models (P 0001 R2 098 data not shown) Cumulative target saturation wasinfluenced by progressive reassignment of former dis-patched targets to their next level of saturation along with

continued detection of new targets in these same saturatedfields Net area saturation was influenced by the sameadaptive paradigm of saturation reassignment and wasalso counter influenced by net area expansions with newunsaturated fields where less targets were detected as statedabove (see Figure 4)

Distinct differences in saturation field assignments wereobserved between final net areas and historic target pointswere distinct for all patches (Figures 7 and 8) Net areaassignments were overrepresented by the lowest saturationlevel whereas historic target assignments were overrepre-sented by the highest saturation levels For instance theexpanded area for the entire patch network was almost sixtimes larger than the highest saturation levels of each patchcombined By comparison historic target assignments in

Table 2 Parameters of the exponential decay function estimated for each patch

Patch TDiab (targets ha21) lc Pd R2 CV(RMSR) TDf (targets ha21)

HAWI 45 0808 0041 092 0222 09WAHI 34 0322 0003 091 0165 08WAKU 50 1170 0001 089 0392 03WALI 14 0286 0001 096 0113 03WALU 30 0239 0001 090 0176 05NULU 30 0434 0003 090 0259 04KANE 70 0488 0001 095 0215 06NETWe 39 6 02 0476 0001 094 0229 05 6 01

a Abbreviations TD target density CV(RMSR) coefficient of variation for the root mean square residuals between observed and fittedvalues TDf final adjusted cumulative target densities HAWI Hanawi WAHI Waihiumalu WAKU Waiakuna Pond WALIWailua iki WALU Wailua nui NULU Nuuailua KANE Keanae Wall NETW network

b Initial adjusted cumulative target densitiesc Decay function exponentd P value for nonlinear regressione The mean TDindashf 6 SE of the entire patch network with the other calculations based the best-fit exponential decay curve

Figure 4 Mean search efficiencies (dark triangle) and herbicideuse rates (light diamond) derived from momentary interventions(6 SE n 5 7 for one to four interventions n 5 6 for five or sixinterventions n 5 5 for seven interventions and n 5 3 for eightinterventions) with mean momentary target density serving as abackdrop for each sequential operation Backdrops appearoversized and transparent for the sole purpose of referencingtarget densities to the intervention performance values in focusand do not constitute any variance or deviation of the mean

Figure 5 Mean cumulative net areas of sequential interventions(6 SE n 5 7 for one to four interventions n 5 6 for five or sixinterventions n 5 5 for seven interventions and n 5 3 for eightinterventions)


the highest saturation levels were 20 times greater than totalrepresentation in the lowest saturation level In practicalterms this highlights the basic elements of an adaptivesurveillance process for delimiting satellite populationsby continuing to revisit known target locations as long asnew targets are being detected while also progressivelyexpanding surveillance coverage into new territories wherefinding targets is a more anomalous occasion

As an example Figure 9 shows the spatial dynamics ofthe sequential interventions imposed on the KANE patchSimilar to all of the other patches this adaptive sequencedepicts (1) target reduction and (2) net area expansion andincreased complexity of the saturation field mosaic Alsotarget detections are recorded throughout the net area butwith a majority of targets consistently recorded in thehighest-saturation fields

Mean detection efficacy of all interventions ranged from056 to 076 (Figure 10) The phenomenon of imperfectdetection is likely the artifact of two conditions (1) falsenegative (type II) error with a failure to detect or (2)biological recruitment of target individuals achievingstature that exceeds the threshold of detectability Meandetection efficacy of the entire patch network was 062 603 which is conspicuously close to the probability ofdetection (Pd) of a random search operation (ie 063)where coverage (c) equals 1 (Koopman 1946 1980)Detection efficacy as measured in this study is anempirical value between sequential interventions wherecoverage of the overlapped field is equal to 1 Cumulativetarget proportions (n 5 41) plotted against theircorresponding coverage values also fit closely with theexponential detection function for a random searchoperation (Figure 11)

Discussion

Miconia has been naturalizing in the EMW for over40 yr with the extant of the invasion approaching20000 ha making containment the current managementgoal (Duncan and Leary 2013) Over the last two decadesfluctuations in funding have guided management decisionsEfforts to reduce high-density infestations have beenprioritized in well-funded years although more often thepriority is to focus efforts on reducing low-density nascentfoci (Taylor and Hastings 2004) The HBT platform isspecifically designed for treating individual weed targetsand is ideal for administering interventions to low-densitypatch populations Adoption of this technology hasinvigorated the miconia containment strategy with accel-erated interventions measurably reducing the nascent patchnetwork within a short period of time In the 14-mo periodreported here operations increased by 33 over theaverage number of operations (per 14 mo) conducted for

Figure 6 Weighted mean saturation for cumulative target(dark squares) and net area (light circles) field mosaicassignments (6 SE n 5 7 for one to four interventions n 5

6 for five or six interventions n 5 5 for seven interventions andn 5 3 for eight interventions)

Figure 7 Proportion of the final cumulative net areaassignments to the saturation field mosaic of each patch NETWrepresents the proportion of the total assignments for allseven patches

Figure 8 Proportion of historical target assignments to thefinal saturation field mosaic of each patch NETW represents theproportion of the total assignments for all seven patches


Figure 9 Intervention sequence (1 to 8) for KANE patch showing targets dispatched (white circles) and the net area field mosaic with increasinglevels of saturation (yellow-to-red) Targets (T) and area (A) are presented (momentarycumulative) along with weighted mean saturation levelsshown in parentheses 30-m contours displayed on the base map (Color for this figure is available in the online version of this paper)


the entire patch network from 2005 to 2011 Consequentlythe number of targets eliminated increased by 168(Supplemental Table 1 httpdxdoiorg101614IPSM-D-13-00059TS1) It is worth reiterating that these patchnetworks are in remote areas accessible only to aerialoperations In these cases the HBT platform improves targetaccessibility with capabilities in delivering an effectiveherbicide dose with horizontal trajectory and long rangeaccuracy relative to current aerial treatment options that mustline up directly over the target (eg long line sprayer) We arealso continuing to validate high treatment efficacy of HBT Inreference to the mortality factor the HBT platform providesassurance that detectability is the limiting factor (Leary et al2013) This study shows how detectability can be improvedby accelerating the intervention schedule Eventually futureoperations will begin recording momentary undetectabletarget levels with intermittent detection of recruitment eventsleading to exhaustion of a latent seed bank

The entire spatial distribution of a weed invasion mustbe properly delimited to ensure effective containment(Panetta and Lawes 2005) The surveillanceinterventionapproach used in this study shares some of the qualitiesfor a delimiting process described by Leung et al (2010)in which they highlight the (1) approach (2) decline and (3)delimitation of a patch as an inside-out tactic for determiningan effective containment boundary All of the patches in thisstudy are in the lsquolsquodeclinersquorsquo stage where the least number oftargets are assigned to the lowest-saturation fields (Figure 8)which also tend to be spatially located on the peripheries ofthe respective net area mosaics (see Figure 9) This strategywill not proceed to the lsquolsquodelimitrsquorsquo stage until all targets havebeen assigned to multi-saturated fields and net area hasexpanded beyond the most distal target assignments

The detection efficacies recorded for these interventionsfit the description of a random search operation (Cooperet al 2003) A theoretically perfect search of a lsquolsquodefiniterangersquorsquo sensor detects 100 of all targets with complete

uniform coverage making any further investment in searcheffort a waste of resources (Koopman 1946 1980)However any randomness in a search effort will reduce Pd

(Cacho et al 2006 2007 Cooper et al 2003 Frost 1999)In fact for a random search operation it would take threesuccessive interventions with no detections recorded in orderto assume a high probability of 098 that no targets are in thearea Only two of the seven patches have a final coverage 30 in this study However targets continued to berecorded up to the last intervention which strongly suggeststhe possibility of target recruitment taking place during thecourse of this study This is further supported by a significantlinear decline in detection efficacy from 18 to 182 d betweeninterventions (P 0039 R2 5 0127 data not shown)suggesting a greater possibility of new recruitment withlonger time intervals With the combination of imperfectdetection and active recruitment future interventions thatrecord undetectable levels should not serve as absoluteconfirmation of population extirpation However it doesprovide the opportunity to more accurately record inter-mittent detections as intrapatch recruitment (ie spatiallyassigned to high-saturation fields) or stochastic dispersalevents (ie spatially assigned to low-saturation fields)

In a weed management scenario randomness resultingin type II errors might include (1) physical stature of aplant target (2) spatial arrangement of cohorts within thelandscape or (3) search capability (Cacho et al 2004)Individual miconia targets can grow 1 m in height in1 yr reaching a detectable threshold within that timeperiod Thus what was undetectable in the previousinterventions will eventually become detectable in subse-quent interventions These patch populations tend to bewell hidden in topographic drainages and ravines as anunderstory to a complex forest canopy structure creatingimpediments to clear sight lines As a frugivore-dispersed

Figure 10 Mean detection efficacy (6 SD n 5 no ofinterventions 2 1 refer to Table 1) of each patch including theentire patch network (NETW mean of all patches 6 SE n 5 7)

Figure 11 Sequential cumulative target proportions (n 5 48)plotted against their respective weighted mean saturation valuesalong with Pd of a theoretical lsquolsquodefinite rangersquorsquo sensor (solid blackline Equation 8) and a random search operation (black dot-dashEquation 1) (Koopman 1946 1980)


propagule miconia targets have been commonly found hid-ing behind the trunks of canopy trees that have presumablyserved as bird defecation perches The helicopter operationsperformed in these extreme three-dimensional spaces are alsolikely incurring randomness during the search effort Theoperation flight lines are not perfectly spaced parallel tracks ona two-dimensional plane (as in ocean search and rescue) butinstead are anisotropic to the terrestrial landscape with intentby the pilot to overlap the fields of view for each track withaltimeter adjustments and visual cues in the landscape Theexponential detection function (Pd) is solely dependent on theamount of search effort applied uniformly to an arearegardless of whether it is applied all at once or incrementally(Koopman 1946 1980) Admittedly the adaptive approachto saturating coverage in known target locations is not auniform process yet net coverage derived from the weightedmean saturation values still show good correspondence to thePd of a random search operation Furthermore if recruitmentand growth to detectable target size is actively occurring totalsearch effort applied to a single intervention would likely beineffective The early development of herbicide symptoms onmiconia was critical to distinguishing new (untreated) targetsin subsequent interventions This feature alone allowed for theacceleration of intervention deployments leading to anincreased probability of target detection in rapid succession

In this study there are three potential outcomes to anintervention1 expansionmdashinitial area coverage naive to any target

encounters which could be the first operation or asubsequent sequential operation expanding beyondknown target locations

2 saturationmdashsequential interventions covering known lo-cations with measurable reductions in target detection or

3 confirmationmdashthe final sequence of interventions mon-itoring momentary undetectable target levels with thelikelihood of intermittent detections characterized asrecruitment events particularly in known target locationsAs the net area surveyed for each patch continues to

expand major portions of the network will amalgamate into acontiguous and complex saturation field mosaic Continuingto build on this intelligence will provide clear depictions ofactive target locations surrounded by confirmed protectedareas The fit of empirical target values to exponential decayand detection functions prove that this accelerated interven-tion strategy is outpacing the biological recruitment of thishighly invasive species As progress continues new strategieswill lead to a deceleration of interventions and a refinement ofcoverage optimizing future operations within the frameworkof an efficient long-term containment strategy

Acknowledgments

This project was funded in parts by US Department ofAgriculture (USDA) Forest Service Special Technology

Development Program Award R5-2012-01 through a collab-oration with the Hawaii Department of Land and NaturalResources Forest Health Program the USDA Hatch ActFormula Grant project 112H and Maui County Offices ofEconomic Development and Department of Water SupplyThe USDA the State of Hawaii and the County of Maui areequal opportunity providers and employers We would alsolike to give our special appreciation to Mr Chuck Chimerafor his editorial improvements and local expertise Theauthors claim no conflict of interest in this report

Literature Cited

Anonymous (2000) Endangered and Threatened Wildlife and PlantsDeterminations of Prudency and Designations of Critical Habitat forPlant Species from the Islands of Maui and Kahoolawe Hawaii USFish and Wildlife Service 65 FR 79192 CFR 50 CFR 17 RIN1018-AH70 Doc no 00-31078 Pp 79192ndash79275

Anonymous (2007) Intelligence Surveillance and ReconnaissanceOperations Air Force Doctrine Document 2ndash9 66 p

Baxter PWJ Possingham HP (2011) Optimizing search strategies forinvasive pests learn before you leap J Appl Ecol 4886ndash95

Cacho OJ Hester S Spring D (2007) Applying search theory todetermine the feasibility of eradicating an invasive population innatural environments Aust J Agric Res Econ 51425ndash433

Cacho O Spring D Pheloung P Hester S (2004) Weed Search andControl Theory and Application Agricultural and ResourceEconomics University of New England Armidale NSW 2351Work-ing paper No 2004-11 17 p

Cacho OJ Spring D Pheloung P Hester S (2006) Evaluating thefeasibility of eradicating an invasion Biol Invasions 8903ndash917

Chimera CG Medeiros AC Loope LL Hobdy RH (2000) Status ofmanagement and control efforts for the invasive alien tree Miconiacalvescens DC (Melastomataceae) in Hana East Maui Honolulu HIUniversity of Hawaii Pacific Coop Studies Unit Tech Rep 128 58 p

Cooper DC Frost JR Robe RQ (2003) Compatibility of land SARprocedures with search theory Department of Homeland SecurityUS Coast Guard Operations Technical Rep DTCG32-02-F-000032 177 p

Cousens R Mortimer M (1995) Dynamics of Weed Populations NewYork Cambridge University Press 348 p

Denslow JS (2003) Weeds in paradise thoughts on the invasibility oftropical islands Ann Mo Bot Gard 90119ndash127

Duncan C Leary JK (2013) Protecting Paradise through Partnershipshttptechlinenewscomarticles2013313protecting-paradise-through-partnerships Accessed June 10 2013

Fox JC Buckley YM Panetta FD Bourgoin J Pullar D (2009)Surveillance protocols for management of invasive plants modellingChilean needle grass (Nassella neesiana) in Australia Divers Distrib15577ndash589

Frost JR (1999) Principles of search theory part II effort coverage andPOD Response 178ndash15

Giambelluca TW Chen Q Frazier AG Price JP Chen YL Chu PSEischeid JK Delparte DM (2013) Online rainfall atlas of HawairsquoiBull Am Meterol Soc 94313ndash316

Gilbert B Levine JM (2013) Plant invasions and extinction debts ProcNatl Acad Sci U S A 1101744ndash1749

Hardesty BD Metcalfe SS Westcott DA (2011) Persistence and spreadin a new landscape dispersal ecology and genetics of Miconiainvasions in Australia Acta Oecol 37657ndash665

Hester SM Brooks SJ Cacho OJ Panetta FD (2010) Applying asimulation model to the management of an infestation of miconia


(Miconia calvescens DC) in the wet tropics of Australia Weed Res 50269ndash279

Jacobson MK (2005) Fundamentals of Atmospheric Modeling 2ndedn New York Cambridge University Press 828 p

Koopman BO (1946) Search and Screening OEG Report No 56 TheSummary Reports Group of the Columbia University Division ofWar Research) Alexandria Virginia Center for Naval Analyses 172 p

Koopman BO (1980) Search and Screening General Principles withHistorical Applications Revised New York Pergamon 400 p

Lawes RA McAllister RRJ (2006) Using networks to understand sourceand sink relationships to manage weeds in riparian zonePages 466ndash469 in Preston C Watts JH Crossman ND edsProceedings of the 15th Australian Weed Conference ManagingWeeds in a Changing Climate Adelaide Australia Weed Manage-ment Society of South Australia

Leary JK Gooding J Chapman J Radford A Mahnken B Cox LJ(2013) Calibration of an herbicide ballistic technology (HBT)helicopter platform targeting Miconia calvescens DC in HawaiiInvasive Plant Sci Manag 6292ndash303

Leung B Cacho OJ Spring D (2010) Searching for non-indigenousspecies rapidly delimiting the invasion boundary Divers Distrib 16451ndash460

Lodge DM Williams SL MacIsaac H Hayes K Leung B Reichard SMack RN Moyle PB Smith M Andow DA Carlton JT McMichaelA (2006) Biological invasions recommendations for US policy andmanagement Ecol Appl 162035ndash2054

Mack RN Simberloff D Lonsdale WM Evans HC Clout M BazzazFA (2000) Biotic invasions causes epidemiology global consequenc-es and control Ecol Appl 10 689ndash710

Medeiros AC Loope LL Conant P McElvaney S (1997) Statusecology and management of the invasive plant Miconia calvescens DC(Melastomataceae) in the Hawaiian Islands Bishop Mus Occas Pap4823ndash36

Meyer J-Y (1996) Status of Miconia calvescens (Melastomataceae) adominant invasive tree in the Society Islands (French Polynesia) PacSci 5066ndash76

Meyer J-Y (1998) Observations on the reproductive biology of Miconiacalvescens DC (Melastomataceae) an alien invasive tree on the islandof Tahiti (South Pacific Ocean) Biotropica 30609ndash624

Meyer J-Y Florence J (1996) Tahitirsquos native flora endangered by theinvasion of Miconia calvescens DC (Melastomataceae) J Biogeog 23775ndash781

Meyer J-Y Loope LL Goarant AC (2011) Strategy to control theinvasive alien tree Miconia calvescens in Pacific islands eradicationcontainment or something else Pages 91ndash96 in Veitch CR CloutMN Towns DR eds Island Invasives Eradication and ManagementGland Switzerland International Union for Conservation of Nature

Moody ME Mack RN (1988) Controlling the spread of plant invasionsthe importance of nascent foci J Appl Ecol 251009ndash1021

Moore JL Hauser CE Bear JL Williams NSG McCarthy MA (2011)Estimating detection-effort curves for plants using search experimentsEcol Appl 21601ndash607

Murphy HT Hardesty BD Fletcher CS Metcalfe DJ Westcott DABrooks SJ (2008) Predicting dispersal and recruitment of Miconiacalvescens (Melastomataceae) in Australian tropical rainforests BiolInvasions 10925ndash936

Panetta FD Lawes R (2005) Evaluation of weed eradication programsthe delimitation of extent Divers Distrib 11435ndash442

Pouteau R Meyer J-Y Stoll B (2011) A SVM-based model forpredicting the distribution of the invasive tree Miconia calvescens intropical rainforests Ecol Model 2222631ndash2641

Regan TJ Chades I Possingham HP (2011) Optimally managing underimperfect detection a method for plant invasions J Appl Ecol 4876ndash85

Reaser JK Meyerson LA Cronk Q DePoorter M Eldrege LG GreenE Kairo M Latasi P Mack RN Mauremootoo J OrsquoDowd D OrapaW Sastroutomo S Saunders A Shine C Thrainsson S Vaiutu L(2007) Ecological and socioeconomic impacts of invasive alien speciesin island ecosystems Environ Conserv 3498ndash111

Rejmanek M Pitcairn MJ (2002) When is eradication of exotic pestplants a realistic goal Turning the tide the eradication of islandinvasives Pages 249ndash253 in Veitch CR Clout MN Towns DR edsIsland Invasives Eradication and Management Gland SwitzerlandInternational Union for Conservation of Nature

Rew LJ Maxwell BD Dougher FL Aspinall R (2006) Searching for aneedle in a haystack evaluating survey methods for non-indigenousplant species Biol Invasions 8523ndash539

Richardson DM Allsopp N DrsquoAntonio C Milton SJ Rejmanek M(2000) Plant invasionsmdashthe role of mutualisms Biol Rev 7565ndash93

Spotswood EN Meyer J-Y Bartolome JW (2013) Preference for aninvasive fruit trumps fruit abundance in selection by an introducedbird in the Society Islands French Polynesia Biol Invasions DOI101007s10530-013-0441-z

Taylor CM Hastings A (2004) Finding optimal control strategies forinvasive species a density-structured model for Spartina alternifloraJ Appl Ecol 411049ndash1057

Thompson SK (2002) Sampling 2nd edn New York J Wiley 400 pWagner WL Herbst DR Sohmer SH (1999) Manual of the Flowering

Plants of Hawaii Revised edn Honolulu University of Hawaii PressBishop Museum Press

[WRCC] Western Region Climate Center httpwwwwrccdrieduAccessed June 10 2013

Received August 2 2013 and approved November 18 2013


Calibration of an Herbicide BallisticTechnology (HBT) Helicopter PlatformTargeting Miconia calvescens in HawaiiJames J K Leary Jeremy Gooding John Chapman Adam Radford Brooke Mahnken and Linda J Cox

Miconia (Miconia calvescens DC) is a tropical tree species from South and Central America that is a highly invasive

colonizer of Hawaiirsquos forested watersheds Elimination of satellite populations is critical to an effective containment

strategy but extreme topography limits accessibility to remote populations by helicopter operations only Herbicide

Ballistic Technology (HBT) is a novel weed control tool designed to pneumatically deliver encapsulated herbicide

projectiles It is capable of accurately treating miconia satellites within a 30 m range in either horizontal or vertical

trajectories Efficacy was examined for the encapsulated herbicide projectiles each containing 1994 mg ae triclopyr

when applied to miconia in 5-unit increments Experimental calibrations of the HBT platform were recorded on a

Hughes 500-D helicopter while conducting surveillance operations from November 2010 through October 2011 on

the islands of Maui and Kauai Search efficiency (min ha21 n 5 13 R2 5 0933 P 0001) and target acquisition

rate (plants hr21 n 5 13 R2 5 0926 P 0001) displayed positive linear and logarithmic relationships

respectively to plant target density The search efficiency equation estimated target acquisition time at 251 sec and a

minimum surveillance rate of 678 s ha21 when no targets were detected The maximum target acquisition rate for

the HBT platform was estimated at 143 targets hr21 An average mortality factor of 0542 was derived from the

product of detection efficacy (0560) and operational treatment efficacy (0972) in overlapping buffer areas

generated from repeated flight segments (n 5 5) This population reduction value was used in simulation models to

estimate the expected costs for one- and multi-year satellite population control strategies for qualifying options in

cost optimization and risk aversion This is a first report on the performance of an HBT helicopter platform

demonstrating the capability for immediate rapid-response control of new satellite plant detections while

conducting aerial surveillance of incipient miconia populations

Nomenclature Miconia Miconia calvescens DC MICA20

Key words Hawaii Herbicide Ballistic Technology

The loss of endemic biological diversity due to exoticplant invasions is particularly detrimental on isolatedislands (Denslow 2003 Mack et al 2000 Reaser et al2007) Either eradication or containment of the invasivespecies can serve as viable mitigation strategies dependingon which option has the greatest potential for success

(Panetta 2009 Panetta and Cacho 2012 Taylor andHastings 2004 Wittenberg and Cock 2001) Regardless ofthe approach detection and control must be effectivelyapplied to the entire population particularly with the mostisolated satellites (Brooks et al 2009 Cacho et al 2006Hulme 2006 Myers et al 2000 Panetta 2009 Panetta andLawes 2005) Archiving knowledge of the target species isalso critical to determine the management approach andwould include studies on the biology (eg growth andfecundity) ecology (ie propagule dispersal) and physiog-raphy (ie suitable habitat) of the target species (Chimera etal 2000 Florence 1993 Hardesty et al 2011 Kuefer et al2010 Pouteau et al 2011) An effective species mitigationstrategy combines (1) practical knowledge (2) sustainedresources and (3) proven actions that progress towards ameasurable reduction of target density and contraction ofthe delimited invasion perimeter

DOI 101614IPSM-D-12-000261

First and sixth authors Assistant Specialist and Specialist

Department of Natural Resources and Environmental ManagementUniversity of Hawaii at Manoa PO Box 269 Kula HI 96790

second author Liaison Pacific Islands Exotic Plant Management

Team National Park Service PO Box 880896 Pukalani HI 96788

third author Operations PlannerAnalyst Kauai Invasive Species

Committee PO Box 1998 Lihue HI 96766 fourth and fifth

authors Operations Manager and GIS Specialist Maui Invasive

Species Committee PO Box 983 Makawao HI 96768 Corre-

spondending authorrsquos Email learyhawaiiedu


292 N Invasive Plant Science and Management 6 AprilndashJune 2013

Miconia (Miconia calvescens DC) is a mid-story tree 12to 15 m tall native to Central and South America It haslarge bicolored leaves that are up to 80 cm in length whichmade this species desirable to botanical hobbyists andhorticultural professionals This led to purposeful intro-ductions to other suitable habitats throughout the Pacific(Meyer 1996) It is currently listed as one of the 100 worstglobal invasive species (Lowe et al 2000) is a class 1 weedin Queensland Australia (Hardesty 2011) and a statenoxious weed in Hawaii USA (Medeiros et al 1997) Themiconia infestation in Tahiti has been well characterizedAfter its introduction in 1937 it became the dominantvegetation to over 65 of the forest in less than 60 yr(Florence 1993 Meyer 1996) High densities of thisshallow rooted species are known to shade out theunderstory vegetation and further suspected to promotesoil surface erosion particularly on steeper terrain(Giambelluca et al 2010 Medeiros et al 1997 Meyer1996)

Miconia is an autogamous species that reaches maturityin 4 to 5 yr A single plant has immense fecundity with theability to produce millions of propagules in a singlereproductive cycle (Meyer 1998) Miconia produces asmall edible fruit approximately 59 mm in diam (023 in)lending itself to frugivorous dispersal by a generalist avianpopulation (Chimera et al 2000) In Australia bothHardesty et al (2011) and Murphy et al (2008) infer that95 of dispersal events occur within 500 m but with amaximum dispersal range that could go beyond 2000 mThe current recommendation is for maintaining radial

management buffers that are at least 500 m but preferably1000 m (Hardesty et al 2011) The most recent reportfrom Tahiti has validated seed bank viability to be over16 yr (Meyer et al 2011) Thus preventing satellitemiconia populations from reaching maturity (ie elimina-tion) is critical to mitigating the invasion

Miconia was introduced to the Hawaiian Islands in 1961(Medeiros et al 1997) Thirty years later the firstmanagement program was initiated on Maui and by1996 management programs existed on the islands ofKauai Oahu and Hawaii (Chimera et al 2000)Population reduction (ie containment over eradication)was recently determined to be the optimal managementpolicy for minimizing expected costs of control on OahuMaui and Hawaii On Kauai deferment of control wassuggested due to the higher costs associated with searchingin a much lower population density (Burnett et al 2007)Currently however all islands including Kauai areoperating under a more risk averse policy that implementssurveillance and treatment operations focused on knownincipient populations This approach may have higheroperational costs but also presents greater opportunity tomitigate detrimental uncertainties regarding the extent ofthese invasions Reduction of satellite populations can havea more mitigating effect on an invasion compared tocontrol efforts in a higher density core infestation (Moodyand Mack 1988) Search effort particularly in remotenatural settings is a costly procedure making plantdetectability critical to effective containment (Cacho etal 2007 Hester et al 2010) Costs are further compound-ed when plant detection and treatment activities areperformed in separate operations which has often beenthe case for controlling miconia in Hawaii

The herbicide active ingredient triclopyr is lethal tomiconia in low doses as basal bark or foliar applications(Chimera et al 2000 Medeiros et al 1998) The aeriallong line spray system currently used to treat miconia wasoriginally developed by the US Drug Enforcement Agencyfor marijuana (Cannabis sativa L) control A typicalassembly consists of a 95 L tank (25 gal) mounted to thecargo hook of a Hughes 500-D helicopter and a tethered30 m by 9 mm hose with a distal nozzle configuration fordirected applications administered by the pilot Most aerialoperations are relegated to inaccessible locations and oftenrequire separate reconnaissance and treatment flights due toencumbrance of the long line sprayer (Burnett et al 2007)This spray system relies on pilot dexterity to safely positionthe nozzle assembly directly overhead and is typicallylimited to treating miconia in open tree canopy gaps andon shallow slopes However miconia is also able to resideon much steeper slopes up to 75u (Pouteau et al 2011) andunder impeding tree canopy (Meyer 1994) making itdifficult or impossible for the long line sprayer to treat alltargets This limitation ultimately compromises the success

Management ImplicationsHerbicide Ballistic Technology (HBT) is a novel application

technique designed to deliver encapsulated herbicide projectileswith long-range accuracy and precision We report on theperformance of an HBT platform providing immediate controlof miconia (Miconia calvescens DC) satellite plants detected whileconducting helicopter surveillance calibrations in Hawaiirsquos re-mote watersheds Flight calibrations (n 5 13) generated efficiencyparameters related to the functionality of the platform Plant targetdensity was a significant variable for determining search efficiency(min ha21) target acquisition rate (plants hr21) and herbicide use(g ae ha21) The product of detection efficacy and treatmentefficacy estimated population mortality (ie reduction) as an-other operational parameter used in simulation models to projectfeasibility and expected cost of different population reductionstrategies based on cost optimization and risk aversion Thisresearch is critical to our technology transfer program that includesdevelopment of the standard operating procedure for safe use ofthe HBT platform which has been approved by the PacificCooperative Studies Unit University of Hawaii and approval bythe Hawaii Department of Agriculture for a FIFRA Section 24cSpecial Local Needs registration for HBT-G4U200 with GarlonH4 Ultra (EPA SLN Reg No HI-120001) with miconia listed as atarget species

Leary et al Herbicide Ballistic Technology Targeting Miconia N 293

of an effective containment strategy (Myers et al 2000Panetta 2009)

Herbicide Ballistic Technology (HBT) is a novelherbicide delivery technique designed to discretely admin-ister encapsulated herbicide aliquots through a pneumaticdevice to individual weed satellites with long-rangeaccuracy The effective treatment range is 30 m in eitherhorizontal or downward vertical trajectories while main-taining submeter accuracy The high velocity impact of theprojectile to the plant (ca 50 m s21) creates a circularspatter pattern that is approximately 1 m2 with ourobservation that a majority of the fluid is retained at thepoint of impact This precision delivery platform isuniquely suited to treating satellite miconia residing onextreme topography or under tree canopy that wouldotherwise impede the long line sprayer

The objectives of this study were to determine triclopyrefficacy when delivered as HBT projectiles to miconia andevaluate the utility of an HBT aerial platform in helicoptersurveillance calibrations The empirical performance mea-sures were further utilized in model simulations withexpected cost analyses to project effective containmentstrategies of incipient miconia populations


The HBT Projectile Batch processing of HBT projectileswas conducted by the Nelson Paint Company (EPA EstNo 86199-MI-001) using standard in-house proceduresfor producing spherical soft gelatin capsules (173 mm dia)with a 26 ml (009 fl oz) liquid fill capacity The HBT-TCP200 herbicide formulation is a simple bipartite blendof triclopyr (356-trichloro-2-pyridinyloxyacetic acidbutoxyethyl ester GarlonH 4 Ultra EPA Reg No 62719-527 DowH Agrosciences LLC Indianapolis IN) dilutedwith a modified vegetable oil concoction of surfactants andcoupling agents to produce a projectile unit with 1994 mgae The HBT-IMZ31 herbicide formulation is imazapyr(2-[45-dihydro-4-methyl-4-(1-methylethyl)-5-oxo-1H-imidazol-2-yl]-3-pyridinecarboxylic acid ArsenalH Power-lineTM EPA Reg No 241-431 BASFH Corp TrianglePark NC) blended with the same adjuvant concoction toproduce a projectile unit with 312 mg ae HBT-TCP200and HBT-IMZ31 formulations are comparable to 16and 5 vv respectively

Treatment Efficacy Validation Two ground-based fieldtrials were established in February 2010 to compare efficacyof the HBT-TCP200 and HBT-IMZ31 on miconia withinthe East Maui infestation (20u459460N 156u019170W)The first experiment was conducted as a completelyrandomized design replicated three times with a 2 by 2factorial treatment set comparing formulations (TCP200vs IMZ31) at two different application rates (5-unit vs 10-

unit) All experimental targets were juvenile miconia ofrelatively uniform size with a single leader stem between 3to 5 m tall Projectiles were administered at the lowest axialpoint from a 3 m range with a marker calibrated todischarge a projectile with a 100 m s21 muzzle velocity Asecond experiment compared the formulations as a 5-unitapplication rate targeting the base of the main leader stemapproximately 30 cm from the soil surface Visualconfirmation of treatment lethality was conducted inSeptember 2010 (227 DAT)

The Onboard HBT Platform The three basic compo-nents of the onboard HBT platform include (1) the HBTprojectile inventory subdivided into pods (ca 140projectiles) serving as retention devices during transfer tothe (2) electro-pneumatic application marker (BTHTM7TM Kee Action Sports LLC Sewell NJ) consistingof a microswitch-controlled solenoid actuating a 1380 kPacompressed air discharge for bolt-action propulsion ofprojectiles powered by (3) a regulated 1180 cm

3

aluminumreservoir tank pressurized up to 20700 kPa The markerwas calibrated to launch projectiles through a 30 cm longceramic-coated smooth bore barrel calibrated for a muzzlevelocity of 100 m s21 and a terminal range of just beyond50 m The tanks were connected to the marker via coiledhigh pressure remote line with quick connect coupler andsliding check valve for depressurized tank replacement Thecomplete onboard assembly consisted of forty pods (ca5600 units) six tanks and two markers The platform wasdesigned to match resource consumption (ie projectilesand compressed air) with operational flight time (ca100 min) and accommodate redundancy in the event of aminor component malfunction

HBT Calibration Flight Segments and Site LocationsHBT platform calibrations were derived from flightsegments of helicopter surveillance operations that weredistinguished by date time and site Up to two segmentswere recorded from a single operation although mostoperations had only one segment recorded A total ofthirteen calibration flight segments were recorded fromOctober 2010 to November 2011 in three sites on theisland of Maui Wailua Nui (20u509060N 156u089060Wthree flight segments) Waiokamilo (20u509030N156u089280W two sites with three flight segments each)and one site on the Island of Kauai Opaekaa (22u049500N159u249110W four flight segments)

In-flight HBT Calibration Protocols and RecordedParameters All calibrations were conducted onboard aHughes 500-D helicopter with doors removed and a three-person crew configured with the applicator seated portsideposterior to the pilot and an additional spotter seated in thefront starboard side All crewmembers were responsible forsafety monitoring and miconia target detection The target


acquisition process between the pilot and the applicatorwas initiated by positive identification of the miconia targetfollowed by the aircraft safely approaching to within a 30 mrange and clear line of sight (PCSU 2011 see Figure 1) Atarget window was designated for the applicator to safelydischarge projectiles within a 270u to 300uhorizontaltrajectory (ie 9 to 10 orsquoclock position) and a 200u to270u vertical trajectory (ie landing skid to eye level) Withthe aircraft in a stationary position the applicator obtainedpermission from the pilot to treat the target dischargedprojectiles recorded GPS waypoint and cued the pilot tocontinue with the operation Detection from the helicopterwas typically limited to miconia plants that were at least1 m tall with fully expanded leaves (ie assumed to be atleast 2nd-yr juveniles) All applications were administeredas a 5-unit treatment to each visible axial point of the plantJuvenile plants were supported by a single leader stem thatcould range from 1 to 4 m tall and typically had 5 axialpoints while larger (potentially mature) plants woulddisplay a spreading canopy with $ 5 axial points All planttarget waypoints and corresponding logs of the flight pathwere recorded with a GPS device (ForetrexH 301 GarminHOlathe KS) set to record geographical position timestampand velocity on 30-s intervals Discrepancy between therecorded applicator location and the actual offset distanceof the target should be noted In most cases closelyapproaching the target would have been hazardous orimpractical and would also have confounded the timeparameter of the calibration All of the targets were locatedwithin 30 m of the recorded waypoint Survivors of earliertreatments were also recorded for repeat calibrations andwere identified as symptomatic targets with viable intactcanopy and were administered retreatment to those living

portions Pod (ca 140 projectiles) inventory consumptionwas recorded for each flight segment

Platform Performance Calculations Operational flightsegments containing first and last recorded miconia targetswere spliced from raw track logs by removing the ferryportions (determined by flight segments to and from thelanding zone that exceeded 20 km hr21) using MapsourceH(version 6124 GarminH Olathe KS) Net surveillanceareas were calculated with a 50 m buffer on each side ofthe operational flight segments with overlapped portionsdissolved using the buffer analysis tool in ArcGISH (version100 ESRIH Redlands CA) The time interval between thetimestamps of the start and end points were recorded Planttarget density was the product of targets acquired dividedby the segment area reported as targets ha21 Targetacquisition rate was the product of the targets acquireddivided by the segment time reported as targets hr21Search efficiency was the product of segment time dividedby segment area reported as min ha21 Detection efficacywas based on the assumption that all targets acquired in thesubsequent overlapping segment were not detected in theprevious segment and was calculated as the product oftargets acquired in the previous segment divided by thecomposite of all targets acquired in the previous andsubsequent segments This is a conservative but reasonableassumption that newly recorded miconia targets is morelikely the result of crew detection errors in the previousoperation than actual recruitment within the short timeintervals between flight segments (ie 89 to 171 days)Operational treatment efficacy was the product ofeffectively treated targets divided by the total targetsacquired Effectively treated targets were deciphered bysubtracting the number of survivors identified in thesubsequent segment Mortality factor as described by Cachoet al (2007) is the product of the probability of detectionmultiplied by treatment efficacy For this study themortality factor was empirically derived as the product ofdetection efficacy multiplied by operational treatmentefficacy calculated from the repeated overlap areas Targetherbicide dose was calculated as the product of podinventory consumption divided by the targets acquired ina flight segment Similarly herbicide use was calculated asthe product of pod inventory consumption divided by thebuffered surveillance area Both projectile consumptionparameters were reported in triclopyr acid equivalents basedon a known quantity of each projectile (eg 1994 mg ae)

Simulation Models and Expected Cost Analyses Simu-lations of 1-yr population reduction strategies withinbuffered isotropic management areas (1 km radius 5314 ha) were performed to compare different managementfrequencies (1) quarter-annual (four operations) (2) semi-annual (two operations) and (3) annual (one operation)The population density range was 0 to 315 targets A

Figure 1 An HBT operator positioned in the portside rear seatof a Hughes 500-D with a pneumatic device engaging twoincipient miconia targets within effective range and a clear line ofsight Photo credit to Josh Atwood


mortality factor of 0542 (the average derived from therepeated flight segments) was imposed on each opera-tionThe undetected targets were calculated from thesubtracted product of the mortality factor which servedas the target population for the subsequent operations forquarter- and semi-annual strategies respectively

A structured matrix model developed for miconia byHester et al (2010) was adopted in this study to simulatestrategies for eradicating incipient miconia populationsThe model estimates annual population growth based onthe probabilities of survival and succession of the (1) seedbank (2) four distinct juvenile stages and (3) small andlarge mature plants generating positive-feedback by fruitproduction augmenting the seedbank The model wasmodified using Meyer (1998) fecundity data where theaverage fruit per panicle was 208 with 195 seeds per fruitFor this model a small mature plant only produced twopanicles which is equivalent to 81120 seed while a largemature plant produced 50 panicles with 2028000 seedThe population vector (Xt) of this matrix model startedwith a seed bank of 668075 propagules 210 juvenileplants and 2 small mature plants (see Appendix 1) Theaverage mortality factor of the HBT platform (0542 seeTable 5) was imposed on the matrix model targeting thejuvenile stages 2 to 4 yr and adult stages with theassumption that first-yr juveniles were undetectablePopulation densities (maturejuvenile) were selected asmanagement starting points that included (1) 248 (2)2074 and (3) 2091518 representing increasing levels ofinvasion identified along the growth model The manage-ment frequencies were as described above and also includedbi- and triennial strategies

Expected cost estimations for both models were based ona helicopter flight cost of $1000 USD hr21 HBT

projectile inventory consumption with a projectile priceof $031 USD (Nelson Paint Company personal commu-nication) and crew costs of $25 USD person-hr21Operational flight times were determined by solving forsearch efficiency (min ha21) dependent on plant targetdensity and the lowest cost determined by the number ofhelicopters needed to accommodate that operation alongwith corresponding ferry times and crew labor (Table 1)Similarly projectile inventory consumption was based onthe equation that solves for herbicide dose which wasestimated to be 25 units per target The 1-yr populationreduction strategies were reported as basic cost estimatesthat would accommodate a manager submitting an annualbudget request The matrix model simulation calculatedthe net present values (NPV) of projected eradicationtimelines at a 6 discount rate applied to the annual costof operations (Hester et al 2010)

Regression analyses were performed using ordinary leastsquares to determine the effect of plant target density onempirically derived search efficiency target acquisition rateand herbicide use (n 5 13) and on expected cost estimatesfor the 1-yr population reduction strategies (SPSS 2009PASW Statistics 18 Release Version 1800 SPSS IncChicago IL)

Results and Discussion

The herbicide efficacy experiment determined HBT-TCP200 to be lethal with 5- and 10-unit treatmentsapplied either to canopy axial points or the base of theleader stem (Table 2) The HBT-IMZ31 treatments werenot effective The high velocity projectiles penetrated thethin epidermis of branches with a fraction of the herbicidecreating a lsquolsquowater soakedrsquorsquo mark at the point of impact and

Table 1 Time and costs estimates ($USD) for helicopter operations

------------------------------OFTa ----------------------------- ----------------------------- Ferryb ---------------------------- ------------------------------ Crewc -----------------------------

13 ops 23 ops Full ops 13 ops 23 ops Full ops 13 ops 23 ops Full ops

1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)

a Operational flight time is the flight time dedicated to surveillance and target acquisition calculated from a fuel cycle of a Hughes500D performing low-level hovering tactics estimated at 120 total min minus 20 min round-trip ferry to and from the LZ A full opssession accommodates 3 fuel cycles providing 5 10 or 15 hours of operational flight time for 1 2 or 3 helicopters respectively Utilityhelicopter flight services are $1000 hr21

b Ferry time is non-operational flight time for round-trip transport of aircraft from the heliport (08 hrs rt) to the LZ and from theLZ to the management containment area (033 hrs rt) Each aircraft is committed to one round-trip heliport ferry and each ops fuelcycle is committed to one round-trip LZ ferry

c Crew management consists of an operations manager plus two crew members for each aircraft Logistical responsibilities includeonboard HBT platform assembly replenishment refueling flight following navigation surveillance and application Wage is $25person-hr21


the remaining portion scattered on to the leaf canopyincluding the undersides The option to effectively treateither the base or canopy of miconia is useful in anoperational setting where a clear line of site at the base ofthe plant may not be available due to adjacent impedingvegetation This herbicide efficacy experiment allowed us toproceed with further evaluating the utility of HBT in ahelicopter platform using the HBT-TCP200 projectiles

A total of thirteen HBT calibration flight segments wererecorded with a single applicator three different pilots andsix spotters Segment time intervals ranged from 13 to92 min (Table 3) Segment lengths ranged from 508 to

16276 m with corresponding buffer areas ranging from49 to 1214 ha Target acquisitions ranged from 4 to 164targets per segment and HBT projectile consumption from2 to 27 pods per segment Search efficiency targetacquisition rate and herbicide use were all correlated toplant target density showing significant positive trendsSimple linear equations produced best fits for searchefficiency (Figure 2 R25 0933 P005 0001) andherbicide use (Figure 4 R2 5 0966 P005 0001) whiletarget acquisition rate was best fit with a logarithmicequation (Figure 3 R2 5 0927 P005 0001)

According to the linear search efficiency equation(Figure 2) a helicopter surveillance operation was estimat-ed to search one hectare in 113 min (ca 68 s) where notargets were detected within the buffered area while theslope coefficient estimated thecomplete target acquisitionprocess to take 0418 min (ca 25 s) (Table 4) Thus inareas with $ 3 targets ha21 the target acquisition timeexceeded surveillance time According to the logarithmicequation (Figure 3) the maximum target acquisition ratewas projected to be 143 targets hr21 which was achieved ata density of 164 targets ha21 but exceeded 100 targetshr21 when the density $ 8 targets ha21 This wasrepresented in 5 of the 13 calibration flight segments(Figure 3) According to the linear herbicide use equationeach miconia plant received a mean herbicide dose of 479 gae triclopyr (017 oz) which back calculates to an estimateof 24 projectiles per target This would suggest an averageplant size with 4 to 5 axial points assuming 100 accuracyof the designated 5-unit application rate A majority of the

Table 2 Miconia survival after treatment with TCP200 andIMZ31 at the main stem axial points with 5- and 10-unitapplications and basal treatments with 5-unit applicationsRecorded 224 DAT

ri rii riii

Alive Dead Alive Dead Alive Dead

Axial

TCP-5 X X XTCP-10 X X XIMZ-5 X X XIMZ-10 X X X

Basal

TCP-5 X X XIMZ-5 X X X

Table 3 Recorded HBT calibration flight segment data

IslandSite Date Ta Lngta Areab Targets Podsc

min m ha--------------------------------------------------------------------------------------------------------------- Maui --------------------------------------------------------------------------------------------------------------Wailua Nui 1110 61 1896 78 134 27

0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

a Flight segment time and length recorded from the start and end points of the track logb Area calculated from a 50 m buffer on both sides with overlap areas dissolved (see Materials and Methods)c Estimated 140 projectiles per pod


targets had 3 to 4 axial points and the accuracy of theapplication was not measured during the calibrations but itwas less than 100 Furthermore an axial point may havereceived 5 units but not more than 10 units Incomparison to other registered triclopyr products thehighest use rate of HBT-TCP200 was 109 of themaximum allowable rate of 896 kg ae ha21(8 lbs ae acre-1)while the remaining twelve calibration flight segments hadcalculated use rates that were 1 As described belowthese low use rates resulted in 94 treatment efficacyThese calibrations highlight the efficiency of a surveillanceoperation that combines effective target control which isfundamental to an effective containment strategy andfurther contributes to a limited knowledge base relatingcontrol effort to weed density (Buddenhagen and Yanez2005 Cacho et al 2007 Campbell et al 1996 Hester2010 Panetta 2009 Panetta and Lawes 2005)

Previously undetected targets were identified and treatedin subsequent overlapping operations The range ofdetection efficacy calculated among the sites was 0427to 0708 with a mean of 0560 (Table 5) Operationaltreatment efficacy was determined in repeat segments byidentifying survivors of a previous application Typicalsymptoms included severe defoliation and necrosis associ-ated with the herbicide treatment but with intact lateralbranches retaining photosynthetic leaf canopy Thussurvivorship was not likely due to a malfunction of theHBT-TCP200 projectiles but rather a condition of theapplication under a simulated operational setting Regard-less the lowest operational treatment efficacy recorded was0941 at Opaekaa with a mean of 0972 for all of thesegments Mortality factors (ie product of detection andtreatment efficacies) among the segments had a range of0424 to 0667 with a mean value of 0542 (Table 5)These calibrations suggest that detection efficacy is a moreinfluential correlate of mortality factor highlighting thedifficulty of plant detection in these natural environmentsbut also validates the consistency of this herbicideapplication platform (eg 5-unit dose)

This study recognizes mortality factor (Cacho et al2006) as an important parameter generated from theseHBT calibrations A mean of 0542 would suggest a needto improve miconia detection through better surveillancetechniques Panetta and Cacho (2012) suggest that astructured search effort with uniform coverage and overlapshould optimize target detectability According to Cachoet al (2007) coverage is a product of speed time anddetectable sight distance This study utilized three differentpilots and six different spotters for conducting thecalibration flight segments (n 5 13) All participants hadseveral years of experience in helicopter surveillanceoperations and miconia detection The speeds observed

Figure 2 A scatter plot with best fit line (R25 0927 P 0001) with 95 confidence interval (dash lines) of searchefficiency (min ha21) versus target density (targets ha21) forHBT calibration flight segments (n 5 13) conducted in Wailuanui (black) Waiokamilo 1 (white) Waiokamilo 2 (stripe) andOpaekaa (grey) from November 2010 to October 2011 SeeTable 4 for regression coefficients

Figure 3 A scatter plot with best fit line (R2 5 0924 P 0001) with 95 confidence interval (dash lines) of targetacquisition rate (targets hr21) versus target density (targets ha21)for experimental HBT helicopter operations (n 5 13) conductedin Wailua nui (black) Waiokamilo 1 (white) Waiokamilo 2(stripe) and Opaekaa (grey) from November 2010 to October2011 See Table 4 for regression coefficients

Figure 4 A scatter plot with best fit line (R2 5 0966 P 0001) with 95 confidence interval (dash lines) of herbicideacid equivalent amount (grams ae ha21) versus target density(targets ha21) for experimental HBT helicopter operations (n 5

13) conducted in Wailua nui (black) Waiokamilo 1 (white)Waiokamilo 2 (stripe) and Opaekaa (grey) from November 2010to October 2011 See Table 4 for regression coefficients


for these flight segments could be described as a slow hoverwithin a lsquolsquocomfortrsquorsquo zone for efficient target detectionwhich was estimated at 53 km hr21 (see y-intercept forsearch efficiency in Table 4) Speed reduction mightimprove target detection but with an added expense andthe possibility of increasing pilot fatigue

Miconia detectability is impeded by heavy vegetationand extreme topography which are typical of tropical wetforest ecosystems Plant size was also a factor indetectability A majority of the acquired plant targets wereat least 1 m tall indicating a minimum age of 2 yr while amajority of the incipient populations may consist ofundetectable 1-yr juveniles This highlights two points (1)miconia detection is likely to be less than 100 and (2)recruitment of new detectable miconia can be expected inmanagement areas within 1 yr following initial populationreduction These two conditions warrant a commitment to

frequent repeated surveillance operations of a managementarea that extends beyond reaching an undetectable level ifthe ultimate goal is to actually achieve complete eradicationof the incipient population Ideally this would occur undera policy that minimizes expected costs based on knowledgeof the target speciesrsquo biology and performance of theoperational strategy (Burnett et al 2007)

Operational costs were projected for simulated 1-yrpopulation reduction strategies with different surveillancefrequencies and plant target densities (Figure 5 Table 4)Costs for all strategies exhibited significant positive lineartrends to plant target density (Table 4 R25 0987ndash0996P005 0001) with semi- and quarter-annual strategiescosting two- and four-fold higher than the annual strategyrespectively where no targets were present Howevertreatment costs (ie slope) progressively increased fromlowest to highest frequency With the cost of the herbicide

Table 4 Regression coefficients as performance analytics for search efficiency target acquisition rate and herbicide dose from empiricalcalibrations (n 5 13) and cost components for simulated 1-yr extirpation strategies of different management frequencies all relative totarget density (T ha21)

Dependent variable------------------------------Coefficients ----------------------------

R2 Pm b

Empiricala

Search efficiency (min ha21) 04183 11307 0933 0001Target acquisition rate (targets hr21) 21533ln 57885 0927 0001Herbicide dose (g ae t21) 47937 0 0966 0001

Simulationb

Qrt MF 096 ($USD) 2381 39111 0996 0001Semi MF 079 ($USD) 2039 19468 0994 0001Annu MF 054 ($USD) 1340 9811 0987 0001

a Refer to Figures 2 3 and 4 for search efficiency target acquisition rate and herbicide dose respectivelyb Refer to Figure 5 for simulations

Table 5 Empirical Estimates for Detection Efficacya (DE) and Operational Treatment Efficacyb (OTE) to calculate Mortality Factors(MF) from overlapping areas with repeated calibrations

Wailua Nui (78 ha) Waiokamilo 1 (73 ha) Waiokamilo 2 (57 ha) Opaekaa (1214 ha)

Segments 1 2 3 1 2ndash3c 1 2ndash3c 1ndash3c 4Targets 134 110 65 140 188 33 35 17 7Survivors mdash 4 5 mdash 1 mdash 0 mdash 1DE mdash 0549 0629 mdash 0427 mdash 0485 mdash 0708OTE mdash 0970 0955 mdash 0993 mdash 1000 mdash 0941MF mdash 0533 0600 mdash 0424 mdash 0485 mdash 0667

a DE is calculated as the ratio of the previous targets from the total combined targets with the subsequent operation and it is assumedthat all new targets in the subsequent operation were previously undetected and not new recruits

b OTE is the ratio of effectively treated targets from the total targets which is derived from the number of confirmed survivorsidentified in the subsequent operation

c Combined calibrations overlapping the entirety of the of the previoussubsequent calibration and were conducted within 1 week ofeach other with targets and survivors reported as a composite value


dose remaining static at $775 per plant the cost increasescorresponded to the extra flight time for overlappingcoverage of the same area regardless of whether the targetwas undetected in a prior operation or with treatmentconfirmation in the subsequent operation The increases inpopulation reduction potential (ie mortality factor) forthe more frequent strategies is less than the correspondingcost increases which reflects the difficulties of treating thelast remaining targets to achieve undetectable levels andcorroborates with Burnett et al (2007) suggestion to defermanagement at extremely low target densities The cost tosearch a 314-ha management area with no targets detectedalong with an added surveillance operation to confirm notargets was estimated to be $19657 USD and served as areference to opportunity cost for the other managementoptions (Figure 5) For instance this cost is comparable toa single surveillance operation of the same size managementarea but with the opportunity to treat up to 734 targets ata mortality factor of 054 while conversely the higherfrequency strategies lose opportunities to confirm unde-tectable levels in other management areas

The matrix model simulated these same strategies andincluded bi- and triennial schedules over a multi-yr perioduntil complete eradication was achieved by exhaustingpredetermined seed banks (Table 6) For all simulationsNPVs decreased proportionally to operational frequencyand were largely influenced by the total number ofoperations and the discount rate applied each year acrossthe timeline The quarter-annual strategy achieved eradi-cation with the shortest timelines but also with the highest

number of operations and at the least discounted ratesthus resulting in the highest NPVs The triennial strategyeradicated the smallest population with the lowest NPVresulting from the longest timeline with the mostdiscounted rates However this strategy failed to eradicatethe two larger vector populations with reduction beingoutpaced by recruitment The biennial strategy eradicatedthese larger vector populations with the longest timelinesand most discounted rates resulting again in the lowestNPVs This illustrates the value of maximizing discountedrates with long-term extensions to strategies However thiscould only be accounted for with a reliable commitment tosustained resources and if those resources were actuallyinvested on years where management was not scheduledMiconia management programs in Hawaii operate underno such mandate and experience fiscal fluctuations withannual renewals

These short- and long-term projections (Figure 5Tables 4 and 6) are deterministic by producing expectedoutcomes which can vary from actual results but shouldcontinue to improve with updates in quantitative ecologyand operations research (Hester et al 2010) Cost estimatesand NPVs reported in this study are within range ofreported projections from Australia (Hester et al 2010)These expected cost and NPV projections assume completesurveillance of a 314-ha isotropic buffer but is likely to bean impractical management unit in heterogeneous envi-ronments (Murphy et al 2008) Spatial analyses ofpopulation distribution and suitable habitat will becomevaluable contributions in management area prioritizationand resource allocation For example ravines and gulliesare known to serve as conduits in frugivorous dispersal andgravitational migration of propagules (Metcalf et al 1998Murphy et al 2008) New information from Tahiti hasidentified several parameters including topgraphic slopeand aspect as identifiers of suitable habitat for miconia(Pouteau et al 2011) The projections from this study arelikely to over-estimate resource requirements to accomplishpopulation reduction goals although the performance ofthe HBT platform should be consistent regardless ofmanagement area designation Future operational usepatterns of the platform are expected to improve thecalibrations with larger data sets generated by multipleapplicators and a broader range of scenarios The protocolsadopted in this study for recording basic operationalparameters and calibrating performance should be universalto most weed management techniques in natural areaswhere search effort and treatment efficacy are both criticalfeatures to successful operations (Cacho et al 2007) TheHBT platform is presented as a complement to existingground and aerial efforts for miconia management inHawaii Accurate calibrations of these conventionaltechniques will further support decisions for futureassignments where HBT might be best suited

Figure 5 Expected operational cost projections of simulated1-yr extirpation strategies within a 314-ha buffer managementarea with incipient population densities up to 1 target ha21 Eachoperation is assigned an MF 5 0542 with different managementfrequencies Qrt (4 ops MF 5 096) Semi (2 ops 079) andAnnu (1 operation MF 5 054) Calculations based on single-helicopter operations The solid line represents a semi-annualhelicopter surveillance strategy with two operations where notargets are detected to confirm extirpation See Table 4 forregression coefficients


A comprehensive miconia management strategy islimited by available fiscal resources which force decisionsto be made between risk aversion and budget optimization(Burnett et al 2007 Hester et al 2010) An optimal policyfor miconia management minimizes the present values ofmanagement costs and residual damages along an infinitetimeline that would even dictate management defermentfor low-density populations where the marginal cost tosearch and treat these remaining individuals is extremelyhigh (Burnett et al 2007) Murphy et al (2008)recommended the establishment of management unitswith a minimum 1000 m radial buffer but in the samereport identified 95 of the targets treated within a 500 mbuffer According to Burnett et al (2007) and in this study(see Figure 5) elimination of that last 5 becomesexceedingly more difficult and costly This study suggestsa triennial schedule as an optimum strategy for the smallestincipient population although it must accommodate a 58-yr timeline with an 8-fold recruitment of the maturepopulation Anecdotally we have encountered multipleoccasions confirming the establishment of mature standswhen reentry to a site exceeds 2 yr From a practitionerrsquosstandpoint a budget optimization approach is likely tobe viewed as too risky particularly when funding pro-

visions fluctuate annually with no permanent mandate ofsupport

For invasive weed management marginal costs can beinterpreted different ways with incremental units based ontarget numbers or net treated area (Buddenhagen andYanez 2005 Campbell et al 1996 Rejmanek and Pitcairn2002) Resources dedicated to control inputs may benegligible compared to resources dedicated to search effort(Hester et al 2010) Helicopter flight time was a dominantcost component for analyzing HBT platform performanceWe identified surveillance operations confirming undetect-able levels (ie no targets detected) to have the lowestmarginal cost that could be applied to a management areaDelimiting incipient populations is critical to effectivecontainment of miconia and should include expandingsurveys into unknown areas despite the probability of notdetecting new populations (Brooks et al 2009 Cacho et al2010) However the strategy must acknowledge theconundrum of an opportunity lost in reducing (ie targettreatment) known incipient populations

This study does not attempt to provide concretedecisions in favor of extreme risk aversion or expected costminimization However miconia is an autogamous specieswith rapid maturity high fecundity and a large dispersal

Table 6 Projected long-term management strategies to eradicate incipient miconia populations within a 314 ha isotropic buffermanagement area with an operational mortality factor (MF) 5 0542

Xta (adultjv) Scheduleb Yearsopsc Adultjvd Mean MFe NPVf ($10003)

248 Qrt 1338 2264 0977 27287Semi 1323 2277 0819 15890Annu 1313 3329 0565 8746Bi 1910 17455 0325 5848Tri 5820 1231736 0225 5717

2074 Qrt 1546 20543 0971 31712Semi 1527 24628 0809 18022Annu 1616 34856 0556 10269Bi 2915 1101653 0306 7299Tri mdash mdash mdash mdash


a Initial vector population of adults and detectable juveniles (yr 2 to 4)b Operation frequency Qrt - four operations yr21 Semi- two operations yr21 Annu- one operation yr21 Bi- one operation two

yrs21 Tri- one operation three yrs21c Total yr and operations projected to achieve incipient population extirpationd Total adults and juvenile (yr 2 to 4) targets acquired to achieve incipient population extirpatione Mean annual mortality factor based on an operational mortality factor of 0542 compounded by the number of operationsf Net present value investment ($USD) based on cost of helicopter operations (see Table 2) projectile inventory consumption and

annual discount rate of 6


range making a single plant in a remote area a high valuetarget that can only be detected with resources dedicated tofrequent overlapping surveillance operations The bestutility identified so far for the HBT platform is throughintegration into aerial surveillance operations wherenominal herbicide use rates translate into significant flighttime cost savings and provides streamlined efforts towardseffective miconia containment

Acknowledgments

Funding for this project was provided in part by the MauiCounty Office of Economic Development Special GrantsProgram Hawaii Invasive Species Council Technology GrantsProgram USDA-CSREES Tropical Subtropical AgricultureResearch Program USDA-NIFA Rural Resources ExtensionAct and the USDA- NRCS Conservation Innovation GrantWe thank Nelson Paint Company Dow Agrosciences LLCBASF Corp and The Wilbur Ellis Company for their currentand past collaborations We thank Keren Gundersen LloydLoope Chuck Chimera and two anonymous reviewers forimproving this manuscript with thoughtful suggestions

Literature Cited

Brooks S J F D Panetta and T A Sydes 2009 Progress towardsthe eradication of three melastome shrub species from northernAustralian rainforests Plant Protect Quart 24(2)71ndash78

Buddenhagen C and P Yanez 2005 The costs of Quinine Cinchonapubescens control on Santa Cruz Island Galapagos Galapagos Res6332ndash36

Burnett K B Kaiser and J Roumasset 2007 Economic Lessons fromControl Efforts for an Invasive Species Miconia calvescens in HawaiiJ For Econ 13(2ndash3)151ndash167

Cacho O J S Hester and D Spring 2007 Applying search theory todetermine the feasibility of eradicating an invasive population innatural environments Aus J Agric and Res Econ 51425ndash433

Cacho O J D Spring P Pheloung and S Hester 2006 Evaluatingthe feasibility of eradicating an invasion Bio Inv 8903ndash917

Cacho O J D Spring S Hester and N MacNally 2010 Allocatingsurveillance effort in the management of invasive species a spatially-explicit model Environ Modelling and Software 25444ndash454

Campbell S D C L Setter P L Jeffrey and J Vitelli 1996Controlling dense infestations of Prosopis pallida Pages 231ndash232 inRCH Shepherd ed Proc 11th Aus Weeds Conf Weed ScienceSociety of Victoria Frankston

Chimera C G A C Medeiros L L Loope and R H Hobdy 2000 Statusof management and control efforts for the invasive alien tree Miconiacalvescens DC (Melastomataceae) in Hana East Maui Honolulu HIUniversity of Hawaii Pacific Coop Studies Unit Tech Rep 128 53 p

Denslow J S 2003 Weeds in paradise thoughts on the invasibility oftropical islands Ann MO Bot Gard 90119ndash127

Florence J 1993 La vegetation de quelques iles de Polynesie Planches54ndash55 in F Dupon coord ed Atlas de la Polynesie franc aiseORSTOM Paris France

Giambelluca T W R A Sutherland K Nanko R G Mudd andA D Ziegler 2010 Effects of Miconia on hydrology a firstapproximation Pages 1ndash7 in L L Loope J Y Meyer B DHardesty and C W Smith eds Proceedings of the InternationalMiconia Conference Keanae Maui Hawaii May 4ndash7 2009Honolulu HI University of Hawaii Maui Invasive Species Com-

mittee and Pacific Coop Studies Unit wwwhearorgconferencesmiconia2009pdfsgiambellucapdf Accessed March 2012

Hardesty B D S S Metcalfe and D A Westcott 2011 Persistenceand spread in a new landscape dispersal ecology and genetics ofMiconia invasions in Australia Acta Oecol 37657ndash665

Hester S M S J Brooks O J Cacho and F D Panetta 2010 Applyinga simulation model to the management of an infestation of Miconiacalvescens in the wet tropics of Australia Weed Res 50(3)269ndash279

Hulme P E 2006 Beyond control wider implications for themanagement of biological invasions J Appl Ecol 43835ndash847

Kueffer C C C Daehler C W Torres-Santana C Lavergne J-YMeyer R Otto and L Silva 2010 A global comparison of plantinvasions on oceanic islands Persp Plant Ecol Evol Syst 12145ndash161

Lowe S M Browne S Boudjelas and M De Poorter 2000 100 ofthe Worldrsquos Worst Invasive Alien Species A selection from the GlobalInvasive Species Database The Invasive Species Specialist Group(ISSG) a specialist group of the Species Survival Commission (SSC)of the World Conservation Union (IUCN) 12 p

Mack R N D Simberloff W M Lonsdale H C Evans M Cloutand F A Bazzaz 2000 Biotic invasions causes epidemiology globalconsequences and control Ecol Appl 10689ndash710

Medeiros A C L L Loope P Conant and S McElvaney 1997Status ecology and management of the invasive plant Miconiacalvescens DC (Melastomataceae) in the Hawaiian Islands B PBishop Museum Occasional Papers 4823ndash36

Medeiros A C L L Loope and R W Hobdy 1998 Interagencyefforts to combat Miconia calvescens on the island of Maui HawairsquoiPages 45ndash51 in J-Y Meyer and C W Smith eds Proceedings of thefirst regional conference on Miconia control August 26ndash29 1997Centre ORSTOM de Tahiti

Metcalfe D J P J Grubb and I M Turner 1998 The ecology ofvery small-seeded shade-tolerant trees and shrubs in lowland rainforest in Singapore Plant Ecol 134131ndash149

Meyer J-Y 1994 Mecanismes drsquoinvasion de Miconia calvescens enPolynesie Francaise PhD thesis IrsquoUniversite de Montpellier I1Sciences et Techniques du Langueduc Montpellier France 122 p

Meyer J-Y 1996 Status of Miconia calvescens (Melastomataceae) adominant invasive tree in the Society Islands (French Polynesia) PacSci 5066ndash76

Meyer J-Y 1998 Observations on the reproductive biology of Miconiacalvescens DC (Melastomataceae) an alien invasive tree on the islandof Tahiti (South Pacific Ocean) Biotropica 30609ndash624

Meyer J-Y L L Loope and A C Goarant 2011 Strategy to controlthe invasive alien tree Miconia calvescens in Pacific islandseradication containment or something else Pages 91ndash96 in C RVeitch M N Clout and D R Towns eds 2011 Island InvasivesEradication and Management Gland Switzerland IUCN

Moody M E and R N Mack 1988 Controlling the spread of plantinvasions the importance of nascent foci J Appl Ecol 251009ndash1021

Murphy H T B D Hardesty C S Fletcher D J Metcalfe D AWestcott and S J Brooks 2008 Predicting dispersal andrecruitment of Miconia calvescens (Melastomataceae) in Australiantropical rainforests Biol Inv 10925ndash936

Myers J H D Simberloff A M Kuris and J R Carey 2000Eradication revisited dealing with exotic species Trends Ecol Evol15316ndash20

PCSU (Pacific Cooperative Studies Unit) 2011 Standing OperatingProcedure for Herbicide Ballistic Technology Operations Groundand Aerial Herbicide Application Safety Management ProgramRCUH-PCSU SOP no 32 19 p

Panetta F D 2009 Weed eradication an economic perspective InvPl Plant Sci Manag 2(4)360ndash368


Panetta F D and O J Cacho 2012 Beyond fecundity control whichweeds are most containable J Appl Ecol doi 101111j1365-2664201102105

Panetta F D and R Lawes 2005 Evaluation of weed eradicationprograms the delimitation of extent Div Distr 11(5)435ndash42

Pouteau R J-Y Meyer and B Stoll 2011 A SVM-based model forpredicting the distribution of the invasive tree Miconia calvescens intropical rainforests Ecol Model 2222631ndash2641

Reaser J K L A Meyerson Q Cronk M DePoorter L G EldregeE Green M Kairo P Latasi R N Mack J Mauremootoo DOrsquoDowd W Orapa S Sastroutomo A Saunders C Shine SThrainsson and L Vaiutu 2007 Ecological and socioeconomicimpacts of invasive alien species in island ecosystems EnvironConserv 3498ndash111

Rejmanek M and M J Pitcairn 2002 When is eradication of exoticpest plants a realistic goal Pages 249ndash253 in C R Vietch and MN Clout eds Turning the Tide The Eradication of IslandInvasive Gland Switzerland IUCN SSC Invasive Species SpecialistGroup

Taylor C M and A Hastings 2004 Finding optimal control strategiesfor invasive species a density-structured model for Spartinaalterniflora J Appl Ecol 411049ndash1057

Wittenberg R and MJW Cock eds 2001 Invasive AlienSpecies A Toolkit of Best Prevention and Management PracticesWallingford Oxon UK CAB International 241 p

Received April 2 2012 and and accepted January 7 2013

Appendix 1 Stage matrix (H) and vector populations (Xt) adopted from Hester et al 2010

H 5 0 0 0 0 0 0 Fsa Fla Xt 5 248 2074 2091518Pfruit Psb 0 0 0 0 0 0 SBa 2080 4799 67545

0 G 0 0 0 0 0 0 jv1 3418 4872 954400 0 Pjv1 0 0 0 0 0 Jv2 37 44 10120 0 0 Pjv2 0 0 0 0 Jv3 8 11 2570 0 0 0 Pjv3 P6 0 0 Jv4 3 19 2490 0 0 0 0 Pjv4 P7 0 sa 0 8 830 0 0 0 0 0 Psa Pla la 2 12 126

a Seed bank values reported in 3 106


Transactions of the ASABE

Vol 59(3) 803-809 copy 2016 American Society of Agricultural and Biological Engineers ISSN 2151-0032 DOI 1013031trans5911474 803

HERBICIDE BALLISTIC TECHNOLOGY SPATIAL TRACKING ANALYSIS OF OPERATIONS CHARACTERIZING

PERFORMANCE OF TARGET TREATMENT

R Rodriguez III J J K Leary D M Jenkins B V Mahnken

ABSTRACT Since 2012 the Herbicide Ballistic Technology (HBT) platform has been deployed in helicopter operations with a mission to eliminate nascent populations of the invasive plant species miconia (Miconia calvescens DC) which is spreading across the East Maui Watershed in Hawaii The HBT platform is a refined pesticide application system that pneumatically delivers encapsulated herbicide projectiles (ie paintballs) from a long range (up to ~30 m) and varying attitude This onboard system provides accurate effective treatment of individual plant targets occupying remote inacces-sible portions of the forested landscape Statistics of operational performance are acquired through GIS analyses of rec-orded GPS data assigned to treated plant targets Recently we have developed a telemetry system for HBT applications (HBT-TS) to enhance the attribute data of target treatments The HBT-TS integrates a hardware sensor with the electro-pneumatic marker that is actuated by the trigger to generate time-stamped geo-referenced attribute data including target assignment azimuth tilt and range determined from the applicator position for every projectile discharged With target assignments the HBT-TS records the estimated dose applied to each target Furthermore the time stamps show that the actual time to administer projectiles (ie target treatment) is a minor component of the total time on target By tracking the orientation and distance of the discharged projectile we can calculate a precise offset target location relative to the applicator position and provide a more accurate interpretation of the herbicide use rate (g acid equivalent ha-1) based on the known amount of herbicide contained in each projectile and the final placement on the landscape We acknowledge the challenges of GPS inaccuracies while recording in a dynamic environment (ie a moving platform in extreme topog-raphy) albeit with increased precision Regardless the current state of the HBT-TS technology enhances operational in-telligence relevant to landscape-scale invasive species management

Keywords Aerial application GIS GPS Miconia calvescens DC Remote sensing Telemetry

he Hawaiian archipelago is the most isolated group of islands on Earth resulting in the evolu-tion of endemic highly specialized biotic com-munities (Carlquist 1970 Mueller-Dombois et

al 1981 Stone and Scott 1985) Currently approximately 1100 endemic plant species and 4600 introduced exotic species have been catalogged of which about 800 are con-sidered invasive (Stone and Scott 1985) These invasive plant incursions are threatening the ecological integrity and function of Hawaiirsquos forested watersheds (Cox 1999 Lin-denmayer et al 2000 Loope and Kraus 2009 Loope et al 2004 Mooney and Drake 1986 Stone et al 1992)

One of the most significant plant invasions in Hawaii is miconia (Miconia calvescens DC) in the East Maui Water-shed (EMW) The EMW is 55000 ha of forested landscape on the northeastern slope of Haleakala volcano Miconia was introduced as an ornamental nursery stock plant in Hana in 1970 Today the core infestation is approaching 1000 ha surrounding Hana with small nascent populations spread across 20000 ha of the EMW Interventions on these incipient outlier populations are critical to maintain-ing an effective containment strategy for protecting the most intact endemic forest areas The extreme topography and remote location of the EMW are major impediments to effective intervention with ground-based operations (Kueffer and Loope 2009) Herbicide Ballistic Technology (HBT) is a novel pesticide application technology (regis-tered as a 24(c) Special Local Need pesticide in Hawaii) that addresses these challenges with a capability to effec-tively treat individual plant targets with long-range accura-cy and flexible attitude from a helicopter while conducting surveillance operations (Leary et al 2014)

Since 2012 HBT interventions have eliminated over 14000 incipient miconia targets as recorded by GPS from the applicator position on the aircraft However with an effective range of ~30 m these coordinates are not the ac-tual target locations Furthermore the herbicide dose has

Submitted for review in August 2015 as manuscript number ITSC

11474 approved for publication by the Information Technology Sensorsamp Control Systems Community of ASABE in January 2016

The authors are Roberto Rodriguez III ASABE Member Graduate Student Department of Biosciences and Bioengineering University ofHawaii at Manoa Honolulu Hawaii James J K Leary Extension Specialist Department of Natural Resources and EnvironmentalManagement University of Hawaii at Manoa Kula Hawaii Daniel M Jenkins ASABE Member Professor Department of Biosciences andBioengineering University of Hawaii at Manoa Honolulu HawaiiBrooke V Mahnken GIS Specialist Maui Invasive Species CommitteeMakawao Hawaii Corresponding author Roberto Rodriguez III 218Agricultural Science Building 1955 East-West Road Honolulu HI96822 phone 727-415-6712 e-mail roberto6hawaiiedu

T








cσ= 14602CMAS (1)



1 (2)


sσ= 52SAS (3)



1 (4)


59(3) 59(3)803-809 805



=

θ=n

iis

0

sin (5)

=

θ=n

iic

0

cos (6)

gtlt+

lt+

gtgt

=θ

0 0360arctan

0180arctan

0 0arctan

csc

s

cc

s

csc

s

(7)




AmL minusθ= (9)




(a) (b) (c)
















y = 04572xRsup2 = 05575

0

10

20

30

40

50

60

0 50 100

Tre

atm

ent

Tim

e (s

)


0

1

2

3

4

5

6

7

8

9

0 50 100

CM

AS

and

SA

S (

m)


59(3) 59(3)803-809 807











201 plusmn108


(a)

(b)


90deg

0deg

180deg

270deg

80

60

40

0

3000

6000

9000

0 100 200 300 400

HU

R (

g ae

ha-

1 )

Target ID

0

10

20

30

40

50

60

70

80

90

0 10 20

Are

a (m

2 )



























0

5000

10000

15000

000 025 050 075 100

HU

R (

g ae

ha-

1 )

pixelnet area

59(3) 59(3)803-809 809









I INTRODUCTION



II BACKGROUND
















A Hardware Design




h = 4433077

[1 minus

(p

p0

)01902632]

+ hoff (1)








B Software Design


m r = mv = minus1

2ρA ACD


2ρA ACL


]+ mg

(2)








REarth(6)

λT = λA + yArarrT

REarth cos φA(7)






φT asympradic

(φA)2 + (ξφ(φA))2 (9)

λT asympradic

(λA)2 + (ξλ(φA)

)2(10)

hT =radic

(h A)2 + (ρ cos ϕ)2 + (ρ sin ϕϕ)2 (11)














C M AS = 21460σc (12)


TABLE I





σc sim 1

2

(σx+σy

)(13)



S AS = 25σs (14)



σs sim 1

3

(σx+σy+σz

)(15)









VI RESULTS















VII CONCLUSION


ACKNOWLEDGMENT


REFERENCES






























Jeremy Gooding


















mortality factor




DOI 101614IPSM-D-13-000591


















































Csatn ~Xn

i~1ixieth THORNX

xi frac122



f (x)~yoeln frac124





Cpropx~Cxn

Cxfinal frac126





Limc1

(Pdc) frac128


Results


























Discussion
























Acknowledgments



Literature Cited










































































DOI 101614IPSM-D-12-000261




























3





















1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited




















































cσ= 14602CMAS (1)



1 (2)


sσ= 52SAS (3)



1 (4)


59(3) 59(3)803-809 805



=

θ=n

iis

0

sin (5)

=

θ=n

iic

0

cos (6)

gtlt+

lt+

gtgt

=θ

0 0360arctan

0180arctan

0 0arctan

csc

s

cc

s

csc

s

(7)




AmL minusθ= (9)




(a) (b) (c)
















y = 04572xRsup2 = 05575

0

10

20

30

40

50

60

0 50 100

Tre

atm

ent

Tim

e (s

)


0

1

2

3

4

5

6

7

8

9

0 50 100

CM

AS

and

SA

S (

m)


59(3) 59(3)803-809 807











201 plusmn108


(a)

(b)


90deg

0deg

180deg

270deg

80

60

40

0

3000

6000

9000

0 100 200 300 400

HU

R (

g ae

ha-

1 )

Target ID

0

10

20

30

40

50

60

70

80

90

0 10 20

Are

a (m

2 )



























0

5000

10000

15000

000 025 050 075 100

HU

R (

g ae

ha-

1 )

pixelnet area

59(3) 59(3)803-809 809









I INTRODUCTION



II BACKGROUND
















A Hardware Design




h = 4433077

[1 minus

(p

p0

)01902632]

+ hoff (1)








B Software Design


m r = mv = minus1

2ρA ACD


2ρA ACL


]+ mg

(2)








REarth(6)

λT = λA + yArarrT

REarth cos φA(7)






φT asympradic

(φA)2 + (ξφ(φA))2 (9)

λT asympradic

(λA)2 + (ξλ(φA)

)2(10)

hT =radic

(h A)2 + (ρ cos ϕ)2 + (ρ sin ϕϕ)2 (11)














C M AS = 21460σc (12)


TABLE I





σc sim 1

2

(σx+σy

)(13)



S AS = 25σs (14)



σs sim 1

3

(σx+σy+σz

)(15)









VI RESULTS















VII CONCLUSION


ACKNOWLEDGMENT


REFERENCES






























Jeremy Gooding


















mortality factor




DOI 101614IPSM-D-13-000591


















































Csatn ~Xn

i~1ixieth THORNX

xi frac122



f (x)~yoeln frac124





Cpropx~Cxn

Cxfinal frac126





Limc1

(Pdc) frac128


Results


























Discussion
























Acknowledgments



Literature Cited










































































DOI 101614IPSM-D-12-000261




























3





















1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited













































59(3) 59(3)803-809 805



=

θ=n

iis

0

sin (5)

=

θ=n

iic

0

cos (6)

gtlt+

lt+

gtgt

=θ

0 0360arctan

0180arctan

0 0arctan

csc

s

cc

s

csc

s

(7)




AmL minusθ= (9)




(a) (b) (c)
















y = 04572xRsup2 = 05575

0

10

20

30

40

50

60

0 50 100

Tre

atm

ent

Tim

e (s

)


0

1

2

3

4

5

6

7

8

9

0 50 100

CM

AS

and

SA

S (

m)


59(3) 59(3)803-809 807











201 plusmn108


(a)

(b)


90deg

0deg

180deg

270deg

80

60

40

0

3000

6000

9000

0 100 200 300 400

HU

R (

g ae

ha-

1 )

Target ID

0

10

20

30

40

50

60

70

80

90

0 10 20

Are

a (m

2 )



























0

5000

10000

15000

000 025 050 075 100

HU

R (

g ae

ha-

1 )

pixelnet area

59(3) 59(3)803-809 809









I INTRODUCTION



II BACKGROUND
















A Hardware Design




h = 4433077

[1 minus

(p

p0

)01902632]

+ hoff (1)








B Software Design


m r = mv = minus1

2ρA ACD


2ρA ACL


]+ mg

(2)








REarth(6)

λT = λA + yArarrT

REarth cos φA(7)






φT asympradic

(φA)2 + (ξφ(φA))2 (9)

λT asympradic

(λA)2 + (ξλ(φA)

)2(10)

hT =radic

(h A)2 + (ρ cos ϕ)2 + (ρ sin ϕϕ)2 (11)














C M AS = 21460σc (12)


TABLE I





σc sim 1

2

(σx+σy

)(13)



S AS = 25σs (14)



σs sim 1

3

(σx+σy+σz

)(15)









VI RESULTS















VII CONCLUSION


ACKNOWLEDGMENT


REFERENCES






























Jeremy Gooding


















mortality factor




DOI 101614IPSM-D-13-000591


















































Csatn ~Xn

i~1ixieth THORNX

xi frac122



f (x)~yoeln frac124





Cpropx~Cxn

Cxfinal frac126





Limc1

(Pdc) frac128


Results


























Discussion
























Acknowledgments



Literature Cited










































































DOI 101614IPSM-D-12-000261




























3





















1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited



























































y = 04572xRsup2 = 05575

0

10

20

30

40

50

60

0 50 100

Tre

atm

ent

Tim

e (s

)


0

1

2

3

4

5

6

7

8

9

0 50 100

CM

AS

and

SA

S (

m)


59(3) 59(3)803-809 807











201 plusmn108


(a)

(b)


90deg

0deg

180deg

270deg

80

60

40

0

3000

6000

9000

0 100 200 300 400

HU

R (

g ae

ha-

1 )

Target ID

0

10

20

30

40

50

60

70

80

90

0 10 20

Are

a (m

2 )



























0

5000

10000

15000

000 025 050 075 100

HU

R (

g ae

ha-

1 )

pixelnet area

59(3) 59(3)803-809 809









I INTRODUCTION



II BACKGROUND
















A Hardware Design




h = 4433077

[1 minus

(p

p0

)01902632]

+ hoff (1)








B Software Design


m r = mv = minus1

2ρA ACD


2ρA ACL


]+ mg

(2)








REarth(6)

λT = λA + yArarrT

REarth cos φA(7)






φT asympradic

(φA)2 + (ξφ(φA))2 (9)

λT asympradic

(λA)2 + (ξλ(φA)

)2(10)

hT =radic

(h A)2 + (ρ cos ϕ)2 + (ρ sin ϕϕ)2 (11)














C M AS = 21460σc (12)


TABLE I





σc sim 1

2

(σx+σy

)(13)



S AS = 25σs (14)



σs sim 1

3

(σx+σy+σz

)(15)









VI RESULTS















VII CONCLUSION


ACKNOWLEDGMENT


REFERENCES






























Jeremy Gooding


















mortality factor




DOI 101614IPSM-D-13-000591


















































Csatn ~Xn

i~1ixieth THORNX

xi frac122



f (x)~yoeln frac124





Cpropx~Cxn

Cxfinal frac126





Limc1

(Pdc) frac128


Results


























Discussion
























Acknowledgments



Literature Cited










































































DOI 101614IPSM-D-12-000261




























3





















1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited













































59(3) 59(3)803-809 807











201 plusmn108


(a)

(b)


90deg

0deg

180deg

270deg

80

60

40

0

3000

6000

9000

0 100 200 300 400

HU

R (

g ae

ha-

1 )

Target ID

0

10

20

30

40

50

60

70

80

90

0 10 20

Are

a (m

2 )



























0

5000

10000

15000

000 025 050 075 100

HU

R (

g ae

ha-

1 )

pixelnet area

59(3) 59(3)803-809 809









I INTRODUCTION



II BACKGROUND
















A Hardware Design




h = 4433077

[1 minus

(p

p0

)01902632]

+ hoff (1)








B Software Design


m r = mv = minus1

2ρA ACD


2ρA ACL


]+ mg

(2)








REarth(6)

λT = λA + yArarrT

REarth cos φA(7)






φT asympradic

(φA)2 + (ξφ(φA))2 (9)

λT asympradic

(λA)2 + (ξλ(φA)

)2(10)

hT =radic

(h A)2 + (ρ cos ϕ)2 + (ρ sin ϕϕ)2 (11)














C M AS = 21460σc (12)


TABLE I





σc sim 1

2

(σx+σy

)(13)



S AS = 25σs (14)



σs sim 1

3

(σx+σy+σz

)(15)









VI RESULTS















VII CONCLUSION


ACKNOWLEDGMENT


REFERENCES






























Jeremy Gooding


















mortality factor




DOI 101614IPSM-D-13-000591


















































Csatn ~Xn

i~1ixieth THORNX

xi frac122



f (x)~yoeln frac124





Cpropx~Cxn

Cxfinal frac126





Limc1

(Pdc) frac128


Results


























Discussion
























Acknowledgments



Literature Cited










































































DOI 101614IPSM-D-12-000261




























3





















1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited






































































0

5000

10000

15000

000 025 050 075 100

HU

R (

g ae

ha-

1 )

pixelnet area

59(3) 59(3)803-809 809









I INTRODUCTION



II BACKGROUND
















A Hardware Design




h = 4433077

[1 minus

(p

p0

)01902632]

+ hoff (1)








B Software Design


m r = mv = minus1

2ρA ACD


2ρA ACL


]+ mg

(2)








REarth(6)

λT = λA + yArarrT

REarth cos φA(7)






φT asympradic

(φA)2 + (ξφ(φA))2 (9)

λT asympradic

(λA)2 + (ξλ(φA)

)2(10)

hT =radic

(h A)2 + (ρ cos ϕ)2 + (ρ sin ϕϕ)2 (11)














C M AS = 21460σc (12)


TABLE I





σc sim 1

2

(σx+σy

)(13)



S AS = 25σs (14)



σs sim 1

3

(σx+σy+σz

)(15)









VI RESULTS















VII CONCLUSION


ACKNOWLEDGMENT


REFERENCES






























Jeremy Gooding


















mortality factor




DOI 101614IPSM-D-13-000591


















































Csatn ~Xn

i~1ixieth THORNX

xi frac122



f (x)~yoeln frac124





Cpropx~Cxn

Cxfinal frac126





Limc1

(Pdc) frac128


Results


























Discussion
























Acknowledgments



Literature Cited










































































DOI 101614IPSM-D-12-000261




























3





















1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited













































59(3) 59(3)803-809 809









I INTRODUCTION



II BACKGROUND
















A Hardware Design




h = 4433077

[1 minus

(p

p0

)01902632]

+ hoff (1)








B Software Design


m r = mv = minus1

2ρA ACD


2ρA ACL


]+ mg

(2)








REarth(6)

λT = λA + yArarrT

REarth cos φA(7)






φT asympradic

(φA)2 + (ξφ(φA))2 (9)

λT asympradic

(λA)2 + (ξλ(φA)

)2(10)

hT =radic

(h A)2 + (ρ cos ϕ)2 + (ρ sin ϕϕ)2 (11)














C M AS = 21460σc (12)


TABLE I





σc sim 1

2

(σx+σy

)(13)



S AS = 25σs (14)



σs sim 1

3

(σx+σy+σz

)(15)









VI RESULTS















VII CONCLUSION


ACKNOWLEDGMENT


REFERENCES






























Jeremy Gooding


















mortality factor




DOI 101614IPSM-D-13-000591


















































Csatn ~Xn

i~1ixieth THORNX

xi frac122



f (x)~yoeln frac124





Cpropx~Cxn

Cxfinal frac126





Limc1

(Pdc) frac128


Results


























Discussion
























Acknowledgments



Literature Cited










































































DOI 101614IPSM-D-12-000261




























3





















1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited


















































I INTRODUCTION



II BACKGROUND
















A Hardware Design




h = 4433077

[1 minus

(p

p0

)01902632]

+ hoff (1)








B Software Design


m r = mv = minus1

2ρA ACD


2ρA ACL


]+ mg

(2)








REarth(6)

λT = λA + yArarrT

REarth cos φA(7)






φT asympradic

(φA)2 + (ξφ(φA))2 (9)

λT asympradic

(λA)2 + (ξλ(φA)

)2(10)

hT =radic

(h A)2 + (ρ cos ϕ)2 + (ρ sin ϕϕ)2 (11)














C M AS = 21460σc (12)


TABLE I





σc sim 1

2

(σx+σy

)(13)



S AS = 25σs (14)



σs sim 1

3

(σx+σy+σz

)(15)









VI RESULTS















VII CONCLUSION


ACKNOWLEDGMENT


REFERENCES






























Jeremy Gooding


















mortality factor




DOI 101614IPSM-D-13-000591


















































Csatn ~Xn

i~1ixieth THORNX

xi frac122



f (x)~yoeln frac124





Cpropx~Cxn

Cxfinal frac126





Limc1

(Pdc) frac128


Results


























Discussion
























Acknowledgments



Literature Cited










































































DOI 101614IPSM-D-12-000261




























3





















1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited
















































A Hardware Design




h = 4433077

[1 minus

(p

p0

)01902632]

+ hoff (1)








B Software Design


m r = mv = minus1

2ρA ACD


2ρA ACL


]+ mg

(2)








REarth(6)

λT = λA + yArarrT

REarth cos φA(7)






φT asympradic

(φA)2 + (ξφ(φA))2 (9)

λT asympradic

(λA)2 + (ξλ(φA)

)2(10)

hT =radic

(h A)2 + (ρ cos ϕ)2 + (ρ sin ϕϕ)2 (11)














C M AS = 21460σc (12)


TABLE I





σc sim 1

2

(σx+σy

)(13)



S AS = 25σs (14)



σs sim 1

3

(σx+σy+σz

)(15)









VI RESULTS















VII CONCLUSION


ACKNOWLEDGMENT


REFERENCES






























Jeremy Gooding


















mortality factor




DOI 101614IPSM-D-13-000591


















































Csatn ~Xn

i~1ixieth THORNX

xi frac122



f (x)~yoeln frac124





Cpropx~Cxn

Cxfinal frac126





Limc1

(Pdc) frac128


Results


























Discussion
























Acknowledgments



Literature Cited










































































DOI 101614IPSM-D-12-000261




























3





















1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited



















































B Software Design


m r = mv = minus1

2ρA ACD


2ρA ACL


]+ mg

(2)








REarth(6)

λT = λA + yArarrT

REarth cos φA(7)






φT asympradic

(φA)2 + (ξφ(φA))2 (9)

λT asympradic

(λA)2 + (ξλ(φA)

)2(10)

hT =radic

(h A)2 + (ρ cos ϕ)2 + (ρ sin ϕϕ)2 (11)














C M AS = 21460σc (12)


TABLE I





σc sim 1

2

(σx+σy

)(13)



S AS = 25σs (14)



σs sim 1

3

(σx+σy+σz

)(15)









VI RESULTS















VII CONCLUSION


ACKNOWLEDGMENT


REFERENCES






























Jeremy Gooding


















mortality factor




DOI 101614IPSM-D-13-000591


















































Csatn ~Xn

i~1ixieth THORNX

xi frac122



f (x)~yoeln frac124





Cpropx~Cxn

Cxfinal frac126





Limc1

(Pdc) frac128


Results


























Discussion
























Acknowledgments



Literature Cited










































































DOI 101614IPSM-D-12-000261




























3





















1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited

















































φT asympradic

(φA)2 + (ξφ(φA))2 (9)

λT asympradic

(λA)2 + (ξλ(φA)

)2(10)

hT =radic

(h A)2 + (ρ cos ϕ)2 + (ρ sin ϕϕ)2 (11)














C M AS = 21460σc (12)


TABLE I





σc sim 1

2

(σx+σy

)(13)



S AS = 25σs (14)



σs sim 1

3

(σx+σy+σz

)(15)









VI RESULTS















VII CONCLUSION


ACKNOWLEDGMENT


REFERENCES






























Jeremy Gooding


















mortality factor




DOI 101614IPSM-D-13-000591


















































Csatn ~Xn

i~1ixieth THORNX

xi frac122



f (x)~yoeln frac124





Cpropx~Cxn

Cxfinal frac126





Limc1

(Pdc) frac128


Results


























Discussion
























Acknowledgments



Literature Cited










































































DOI 101614IPSM-D-12-000261




























3





















1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited





















































C M AS = 21460σc (12)


TABLE I





σc sim 1

2

(σx+σy

)(13)



S AS = 25σs (14)



σs sim 1

3

(σx+σy+σz

)(15)









VI RESULTS















VII CONCLUSION


ACKNOWLEDGMENT


REFERENCES






























Jeremy Gooding


















mortality factor




DOI 101614IPSM-D-13-000591


















































Csatn ~Xn

i~1ixieth THORNX

xi frac122



f (x)~yoeln frac124





Cpropx~Cxn

Cxfinal frac126





Limc1

(Pdc) frac128


Results


























Discussion
























Acknowledgments



Literature Cited










































































DOI 101614IPSM-D-12-000261




























3





















1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited














































TABLE I





σc sim 1

2

(σx+σy

)(13)



S AS = 25σs (14)



σs sim 1

3

(σx+σy+σz

)(15)









VI RESULTS















VII CONCLUSION


ACKNOWLEDGMENT


REFERENCES






























Jeremy Gooding


















mortality factor




DOI 101614IPSM-D-13-000591


















































Csatn ~Xn

i~1ixieth THORNX

xi frac122



f (x)~yoeln frac124





Cpropx~Cxn

Cxfinal frac126





Limc1

(Pdc) frac128


Results


























Discussion
























Acknowledgments



Literature Cited










































































DOI 101614IPSM-D-12-000261




























3





















1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited















































σs sim 1

3

(σx+σy+σz

)(15)









VI RESULTS















VII CONCLUSION


ACKNOWLEDGMENT


REFERENCES






























Jeremy Gooding


















mortality factor




DOI 101614IPSM-D-13-000591


















































Csatn ~Xn

i~1ixieth THORNX

xi frac122



f (x)~yoeln frac124





Cpropx~Cxn

Cxfinal frac126





Limc1

(Pdc) frac128


Results


























Discussion
























Acknowledgments



Literature Cited










































































DOI 101614IPSM-D-12-000261




























3





















1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited























































VII CONCLUSION


ACKNOWLEDGMENT


REFERENCES






























Jeremy Gooding


















mortality factor




DOI 101614IPSM-D-13-000591


















































Csatn ~Xn

i~1ixieth THORNX

xi frac122



f (x)~yoeln frac124





Cpropx~Cxn

Cxfinal frac126





Limc1

(Pdc) frac128


Results


























Discussion
























Acknowledgments



Literature Cited










































































DOI 101614IPSM-D-12-000261




























3





















1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited















































Jeremy Gooding


















mortality factor




DOI 101614IPSM-D-13-000591


















































Csatn ~Xn

i~1ixieth THORNX

xi frac122



f (x)~yoeln frac124





Cpropx~Cxn

Cxfinal frac126





Limc1

(Pdc) frac128


Results


























Discussion
























Acknowledgments



Literature Cited










































































DOI 101614IPSM-D-12-000261




























3





















1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited

















































































Csatn ~Xn

i~1ixieth THORNX

xi frac122



f (x)~yoeln frac124





Cpropx~Cxn

Cxfinal frac126





Limc1

(Pdc) frac128


Results


























Discussion
























Acknowledgments



Literature Cited










































































DOI 101614IPSM-D-12-000261




























3





















1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited


































































Discussion
























Acknowledgments



Literature Cited










































































DOI 101614IPSM-D-12-000261




























3





















1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited






























































Acknowledgments



Literature Cited










































































DOI 101614IPSM-D-12-000261




























3





















1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited




































































































DOI 101614IPSM-D-12-000261




























3





















1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited





























































3





















1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited





























































1 heli 17 ($1667) 33 ($3300) 50 ($5000) 11 ($1133) 15 ($1467) 18 ($1800) 120 ($300) 195 ($488) 270 ($675)2 heli 33 ($3300) 67 ($6700) 100 ($10000) 23 ($2267) 29 ($2933) 36 ($3600) 200 ($500) 325 ($813) 450 ($1125)3 heli 50 ($5000) 100 ($10000) 150 ($15000) 34 ($3400) 44 ($4400) 54 ($5400) 280 ($700) 455 ($1138) 630 ($1575)










ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited


















































ri rii riii


Axial


Basal





0511 69 2372 81 114 190811 64 3125 156 71 15

Waiokamilo 1 0411 60 1402 82 140 220811 45 1343 87 96 150811 92 3849 198 164 23

Waiokamilo 2 0511 26 756 63 33 50811 13 508 49 16 30811 13 571 51 19 3

---------------------------------------------------------------------------------------------------------------Kauai --------------------------------------------------------------------------------------------------------------Opaekaa 0411 39 8447 587 8 3

0611 73 9048 700 5 20711 48 5984 518 4 21011 69 16276 1214 7 2

















R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited



























































R2 Pm b

Empiricala


Simulationb






























Acknowledgments


Literature Cited

































































Acknowledgments


Literature Cited













































Rodriguez, R., D. Jenkins, J.K. Leary and B.V. Mahnken ... · Rodriguez, R., D. Jenkins, J.K....

Documents

Transcript of Rodriguez, R., D. Jenkins, J.K. Leary and B.V. Mahnken ... · Rodriguez, R., D. Jenkins, J.K....