MODIS Vegetation Indices (MOD13)

MODIS Vegetation Indices (MOD13)Ramón Solano-Barajas

[email protected]

Laboratorio de Geomática, Universidad de ColimaCoquimatlán, Col. Mexico 28400

mailto:[email protected]

mailto:[email protected]

WHY TO CARE ABOUT VIS?

• VIs are estimates of the APAR of plants hence correlated with photosynthetic capacity (Sellers 1985, Myneni 1995)

• Then VI over time ~ GPP

• Related to vegetation “greeness” and “health”

• Allow time series analysis (growing season analysis)

• Better understanding of carbon cycle and sink

X

BIOPHYSICAL BASIS

• VIs are a function of the optical and physical properties of leaves, canopy structure, soil background, and atmosphere load.

• VIs are typically based on the absorption of blue and red light by chlorophyll, and reflection of near-infrared light by the leaves’s structure

CHALLENGES

• Observing vegetation from space has several challenges, mainly due to atmospheric and soil effects

• Additional issues may be introduced by aggregating techniques (e.g. MVC) and observation geometry (BRDF effects)

VIS FORMULATION

• Several forms of VIs have been proposed over time. Original goals were to estimate vegetation status and biomass (Kriegler 1969 , Miller 1972, Rouse 1974, Tucker 1977). Now many applications

• Each VI instance try to address a specific issue, such as atmospheric or soil effects

• Examples:

• NDVI - Normalized Difference VI

• ARVI - Atmospheric Resistant VI

• SAVI - Soil Adjusted VI

• EVI - Enhanced VI

NDV I =NIR�R

NIR+R

EV I = G · NIR�R

NIR+ C1 ·R� C2 ·B + 1

ARV I =NIR�RB

NIR+RB

RB = R� �(B �R)

MODIS VI (MOD13)

• The MODIS VI product (MOD13) is a chain of VI products at varying spatial (250m, 1km, 0.05 degree) and temporal (16-day, monthly) resolutions

• It contains both NDVI and EVI indices:

• NDVI as the continuity index (+30 yr data). Simple, well-known, APAR-correlated. Shows “saturation” and atmospheric and soil issues

• EVI as an enhanced index, addressing some NDVI concerns (although at some “costs”)

• A 2-band EVI (no blue band) is used as backup algorithm under anomalous areas

MODIS VI (MOD13)

Product Family

•MOD13Q1: 16 d, 250 m, tiled

•MOD13A1: 16 d, 500 m, tiled

•MOD13A2: 16 d, 1 km, tiled

•MOD13A3: monthly, 1 km, tiled

•MOD13C1: 16 d, 0.05 deg (CMG)

•MOD13C2: monthly, 0.05 deg (CMG)

Tile h09v05

MODIS Sinusoidal 10x10 deg tiles

MOD13 AGGREGATION

• MODIS VI products rely on the upstream daily Surface Reflectance (SR, MOD09) series

• The VI algorithms temporally composite the SR to generate the VI products.

• The 1-km VI product first aggregates 250- and 500-m pixel sizes to 1 km.

• The CMG products are generated through spatial averaging of the 1-km versions

• Monthly products are time-weighed averages of their 16-day versions

MOD13 PRODUCT CHAIN

Surface(flectance(L2G(

Surface(Reflectance(

MODPRAGG(

Aggregated(1km(Surface(

Reflectance(

MOD13A2(1km(16day(

ComposiBng(

Temporal(Averaging(

MOD13A3(1km(Monthly(

ComposiBng(

MOD13(Q1/A1((250/500m(16day(

SpaBal(Averaging(

Temporal(Averaging(

MOD13C1(0.05deg(16day(

MOD13C2(0.05deg(Monthly(

MOD09GHK MOD09GQK MOD09GST MOD09GAD MOD0PTHKM MODPTQKM

Figure 1: Overview of MODIS VI product series.

products for the Southwest USA are included at the end of this document (Fig. 8).

5.1 Algorithm Description

The VI algorithm operates on a per-pixel basis and requires multiple observations (days) togenerate a composited VI. Due to orbit overlap, multiple observations may exist for one dayand a maximum of four observations may be collected. In theory, this can result in a max-imum of 64 observations over a 16-day cycle, however, due to the presence of clouds andthe actual sensor spatial coverage, this number will range between 64 and 0 with decreas-ing observations from polar to equatorial latitudes. The MOD13A1 algorithm separates allobservations by their orbits providing a means to further filter the input data.

Once all 16 days are collected, the MODIS VI algorithm applies a filter to the data basedon quality, cloud, and viewing geometry (Fig. 3). Cloud-contaminated pixels and extremeoff-nadir sensor views are considered lower quality. A cloud-free, nadir view pixel with noresidual atmospheric contamination represents the best quality pixel. Only the higher qual-

7

MOD13 AGG ALGORITHM

• MODIS VI algorithm applies a filter to the data based on quality, cloud, and viewing geometry

• Contaminated and extreme off-nadir data are considered lower quality. A cloud-free, nadir view pixel with no residual atmospheric contamination represents the best quality pixel.

• Compositing VI technique uses a Constrained View angle - Maximum Value Composite (CV-MVC)

• It uses a simple MVC when no enough data exist

MOD13 AGG ALGORITHM

!"#$%&'()'&*%+&'(!*%&8(>$(7;(6%AD<(A9%5'@(

+(5>>6(G4(3AL'D@(

M(-(

N'@(O>(

+(%E%ADK(3AL'D@(M(7(

N'@(O>(

!'D'&*(P'@*(?AL'D("@A+5(=QR

0Q=(

!'D'&*(P'@*(?AL'D("@A+5(

0Q=(

=>9?"*'(SADD(E%D"'("@A+5((

FA@*>#A&(%E'#%5'(

/'@"DB+5(&>9?>@A*'(BD'(

3AL'D(P<(?AL'D(%+%D<@A@(

T9%5'(A+*'5#%*'6(?AL'D(P<(?AL'D(

Figure 3: MODIS VI Compositing algorithm data flow.

is similar to that used in the AVHRR-NDVI product, in which the pixel observation withthe highest NDVI value is selected to represent the entire period (16 days). Furthermore,the algorithm will choose the orbit observation with the highest NDVI if presented withmultiple observations for the same day (multiple orbits).

The CV-MVC is an enhanced MVC technique, in which the number of observations n (nbeing set to 2 at the moment) with the highest NDVI are compared and the observationwith the smallest view angle, i.e. closest to nadir view, is chosen to represent the 16-daycomposite cycle.

All compositing methodologies result in spatial discontinuities, which are inevitable andresult from the fact that disparate days can always be chosen for adjacent pixels over the 16-day period. Thus, adjacent selected pixels may originate from different days, with differentsun-pixel-sensor viewing geometries and different atmospheric and residual cloud/smokecontamination.

5.2 Scientific Data Sets

The 250m/500-m VI product has the following characteristics (Table 1).

9

MOD13 CONTENTS (SDS)

Science&Data&Set& Units& Data&type&

Valid&Range&

Scale&factor&

XYZm%16%days%NDVI% NDVI% int16% 32000,%10000%

0.0001%

XYZm%16%days%EVI% EVI% int16% 32000,%10000%

0.0001%

XYZm%16%days%VI%Quality%detailed%QA% Bits% uint16% 0,%65534% NA%XYZm%16%days%red%reflectance%(Band%1)% Reflectance% int16% 0,%10000% 0.0001%XYZm%16%days%NIR%reflectance%(Band%2)% Reflectance% int16% 0,%10000% 0.0001%XYZm%16%days%blue%reflectance%(Band%3)% Reflectance% int16% 0,%10000% 0.0001%XYZm%16%days%MIR%reflectance%(Band%7)% Reflectance% int16% 0,%10000% 0.0001%XYZm%16%days%view%zenith%angle% Degree% int16% 39000,%

9000%0.01%

XYZm%16%days%sun%zenith%angle% Degree% int16% 39000,%9000%

0.01%

XYZm%16%days%relative%azimuth%angle% Degree% int16% 33600,%3600%

0.1%

XYZm%16%days%composite%day%of%the%year% Day%of%year% int16% 1,%366% NA%XYZm%16%days%pixel%reliability%summary%QA% Rank% int8% 0,%3% NA%

!

MOD13 QA SCHEME

Product Level

SDS Level

Field Level

MOD13&QA&

DETAILED&QA&SDS&&9&Fields&(16&bits)&

SUMMARY&QA&SDS&

VI&USEFULNESS&Field&2&(BITS&2A5)&

MOD13 QUALITY ASSURANCE

Rank%Key% Summary%QA% Description%61% Fill/No%Data% Not%Processed%0% Good%Data% Use%with%confidence%1% Marginal%data% Useful,%but%look%at%other%QA%information%2% Snow/Ice% Target%covered%with%snow/ice%3% Cloudy% Target%not%visible,%covered%with%cloud%

!MOD13Q1/A1 Pixel Reliability SDS

MOD13&QA&


SUMMARY&QA&SDS&


Bits% Parameter%Name% Value% Description%051% VI%Quality%(MODLAND%QA%

Bits)%00% VI%produced%with%good%quality%

%% %% 01% VI%produced,%but%check%other%QA%%% %% 10% Pixel%produced,%but%most%probably%cloudy%%% %% 11% Pixel%not%produced%due%to%other%reasons%than%

clouds%255% VI%Usefulness% 0000% Highest%quality%%% %% :% :%%% %% 1101% Quality%so%low%that%it%is%not%useful%%% %% 1110% L1B%data%faulty%%% %% 1111% Not%useful%for%any%other%reason/not%processed%657% Aerosol%Quantity% 00% Climatology,%Low%5%High%8% Adjacent%cloud%detected% 0% No%%% %% 1% Yes%9% Atmosphere%BRDF%

Correction%0% No/Yes%

10% Mixed%Clouds% 0% No/Yes%11513% Land/Water%Mask% 000% Shallow%ocean,%Land,%inland%water,%etc%14% Possible%snow/ice% 0% No/Yes%15% Possible%shadow% 0% No%

!


MOD13Q1/A1 VI Quality detailed QA SDS

MOD13&QA&


SUMMARY&QA&SDS&


Parameter'Name' Condition' Score'Aerosol'Quantity' If'aerosol'climatology'was'used'for'atmospheric'

correction'(00)'2'

'' If'aerosol'quantity'was'high'(11)' 3'Atmosphere'Adjacency'Correction'

If'no'adjacency'correction'was'performed'(0)' 1'

Atmosphere'BRDF'Correction' If'no'atmosphereEsurface'BRDF'coupled'correction'was'performed'(0)'

2'

Mixed'Clouds' If'there'possibly'existed'mixed'clouds'(1)' 3'Shadow' If'there'possibly'existed'shadow'(1)' 2'View'zenith'angle'(vz)' If''vz'>'40' 1'Sun'zenith'angle'(sz)' If''sz'>'60' 1'

!


MOD13 Q1/A1 QA Detailed SDS > VI Usefulness Field (bits 2-5)

MOD13&QA&


SUMMARY&QA&SDS&


MOD13 MONTHLY ALGORITHM

• This algorithm uses all 16-day VI products which overlap within a calendar month.

• Once all 16-day composites are collected, a weigh- ing factor based on the degree of temporal overlap is applied to each input.

• In assigning the pixel QA, a worst case scenario is used, whereby the pixel with the lowest quality determines the final pixel QA.

Figure 4: Monthly MODIS VI flow diagram.


The monthly 1-km MOD13A3 VI product has 11 SDS’s, as listed in Table 12. Comparedwith MOD13A2, the only difference (besides the temporal aggregation) is the lack of thecomposite day of the year SDS.

Table 12: Product MOD13A3: monthly 1-km VI.

Science Data Set Units Data type Valid Range Scale factor

1km monthly NDVI NDVI int16 -2000, 10000 0.00011km monthly EVI EVI int16 -2000, 10000 0.00011km monthly VI Quality Bits uint16 0, 65534 NA1km monthly red reflectance (Band 1) Reflectance int16 0, 10000 0.00011km monthly NIR reflectance (Band 2) Reflectance int16 0, 10000 0.00011km monthly blue reflectance (Band 3) Reflectance int16 0, 10000 0.00011km monthly MIR reflectance (Band 7) Reflectance int16 0, 10000 0.00011km monthly view zenith angle Degree int16 -9000, 9000 0.011km monthly sun zenith angle Degree int16 -9000, 9000 0.011km monthly relative azimuth angle Degree int16 -3600, 3600 0.11km monthly pixel reliability Rank int8 0, 3 NA

22

MOD13 CMG (0.05 DEG) ALGORITHM

• The VI CMG series is a seamless global 3600x7200 pixel data product with 13 SDS

• It incorporates a QA filter scheme that removes lower quality pixels in aggregating the 1-km pixels into the 0.05-deg CMG product

• Uses only clean pixels to compute final value

• If all pixels are cloudy, it uses historical time series values

This work was supported by NASA-MODIS Contract no. NNG04HZ20C

Climatology Fill StrategyThe MOD13C1 uses the entire MODIS data record to calculate a reliable VI fill value in case input data is missingor deemed cloudy. The fill value is calculated from the average of good data from all previous years’ CMGs of that composite period. It is mainly used for replacing completely cloudy data, but is powerful enough to reliably fill in whole missing tiles (images below had tiles missing in processing).

EOS Coresite - JiParana - Climatology Fill(Note: Boxes on a line indicates fill values)

00.20.40.60.8

1

2003

001

2003

033

2003

065

2003

097

2003

129

2003

161

2003

193

2003

225

2003

257

2003

289

2003

321

2003

353

2004

017

2004

049

2004

081

Composite Period

VI V

alue

Filtered & FilledNDVI (1x1)

Unfiltered NDVI(1x1)

Filtered & FilledEVI (1x1)

Unfiltered EVI(1x1)

All fill strategies have their fallacies and pitfalls. In the Climatology Fill case, certain highly dynamic regions may show discrepancies where fill values were used. This is most obvious when missing input tiles are replaced – edges may be visible. For pixels filled due to high cloud contamination, thisfill strategy will perform well.

The fill completes the two VI layers with data. Other layers will contain their respective fill values, except data layer 11,which is set to 0 – i.e., no good input data. Below is a time series analysis of the EOS Core site “JiParana” which shows clear improvements from the climatology based fill method.

Examples of Outputs

The above histograms show how seasonal changes between years. The two areas covered are North and South America. Comparing the histograms for February, May, August, and November, clear differences between years become visible. These figures also show the typical dynamic range and “VI-structure” across wide ranges in vegetation types.

NorthAmerica Seasonal NDVI Histograms(2000-049 to 2004-097)

0.00E+00

5.00E-03

1.00E-02

1.50E-02

2.00E-02

2.50E-02

3.00E-02

-0.2

00

-0.1

25-0

.050

0.02

5

0.10

00.

175

0.25

0

0.32

5

0.40

00.

475

0.55

0

0.62

50.

700

0.77

5

0.85

00.

925

1.00

0

NDVI Value

Freq

uenc

y (%

are

a co

vera

ge)

2000-0492000-1452000-2412000-3212001-0492001-1452001-2412001-3212002-0492002-1452002-2412002-3212003-0492003-1452003-2412003-3212004-049

NorthAmerica Seasonal EVI Histograms(2000-049 to 2004-097)

0.00E+00

5.00E-03

1.00E-02

1.50E-02

2.00E-02

2.50E-02

3.00E-02

-0.2

00

-0.1

25

-0.0

500.

025

0.10

0

0.17

5

0.25

0

0.32

5

0.40

0

0.47

5

0.55

0

0.62

5

0.70

0

0.77

50.

850

0.92

5

1.00

0

EVI Value

Freq

uenc

y (%

are

a co

vera

ge)

2000-0492000-1452000-2412000-3212001-0492001-1452001-2412001-3212002-0492002-1452002-2412002-3212003-0492003-1452003-2412003-3212004-049

SouthAmerica Seasonal NDVI Histograms(2000-049 to 2004-097)

0.00E+00

2.00E-03

4.00E-03

6.00E-03

8.00E-03

1.00E-02

1.20E-02

1.40E-02

-0.200 -0.060 0.080 0.220 0.360 0.500 0.640 0.780 0.920

NDVI Value

Freq

uenc

y (%

are

a co

vera

ge)

2000-0492000-1452000-2412000-3212001-0492001-1452001-2412001-3212002-0492002-1452002-2412002-3212003-0492003-1452003-2412003-3212004-049

SouthAmerica Seasonal EVI Histograms(2000-049 to 2004-097)

0.00E+00

1.00E-03

2.00E-03

3.00E-03

4.00E-03

5.00E-03

6.00E-03

7.00E-03

8.00E-03

9.00E-03

-0.200 -0.060 0.080 0.220 0.360 0.500 0.640 0.780 0.920

EVI Value

Freq

uenc

y (%

are

a co

vera

ge)

2000-0492000-1452000-2412000-3212001-0492001-1452001-2412001-3212002-0492002-1452002-2412002-3212003-0492003-1452003-2412003-3212004-049

1. Regional Histograms

The measure of standard deviation can be used to detect variability in vegetation over time. The above standard deviation images, created by combining all composite periods into an “average year”, are a depiction of the complete 4-year MODIS VI record. The standard deviation is a measure of seasonal variations about the annual mean. (Northern latitude snow cover is likely to cause some inaccuracies.) Areas with high variation in the NDVI and EVI are highlighted in the images. Brighter areas in the images indicate a higher standard deviation (seasonality).

3. Vegetation Variability through Temporal Average Standard Deviation

2. Total Veg. Biomass, as depicted by the annual cumulative VI average

EVI NDVI

The images above are single image representations of the complete MODIS data record over 4 years. Values used to create these images were the 4-year mean annual VI-response profiles.

Andree Jacobson, Kamel Didan, and Alfredo Andree Jacobson, Kamel Didan, and Alfredo HueteHueteTerrestrial Biophysics and Remote Sensing Laboratory

The University of Arizona, Tucson{andree,kamel,ahuete}@AG.Arizona.EDU

The Vegetation Index Climate Modeling Grid (CMG) is a new product ready for incorporation in MODIS Data Collection 5, scheduled for processing in June 2005. The VI CMG is a seamless 3600x7200 pixel data product with 12 layers, at approximately 544MB per composite period. This is a higher quality climate product useful in time series analyses of earth surface processes. It incorporates a QA (quality analyses) filter scheme that removes lower quality, cloud contaminated pixels in aggregating the 1 km pixels into the 0.05 degree CMG product. It also incorporates a data fill strategy, based on historic data records, to produce a continuous and reliable product for readyentry into biogeochemical, carbon, and growth models. With its very manageable size, the VI CMG can be used for many purposes, some of which are presented here.

Abstract

MODIS VI CMG Data Layers (1) NDVI

Recomputed NDVI(2) EVI

Recomputed EVI(3) VI Quality

Output QA set to dominant input QA (4) red, (5) NIR, (6) blue, and (7) MIR reflectances

Spatially averaged and cloud filtered.(8) Average sun zenith angle(9) NDVI & (10) EVI standard deviation

Standard deviation of cloud filtered input VIs(11) Number of 1km pixels used

Number of 1km pixels used in calculation of new layers.If all pixels were deemed cloudy, this number is 0.

(12) Number of 1km pixels within +/-30 degreesNumber of the cloud filtered 1km pixels that are within30 degrees view angle.

CMG Production 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35

00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17

(286 Tiles x 31MB ! 9GB) (12 layers, 3600x7200 pixels, ! 544MB)

MOD13C1

MOD13A2

To offer the best quality data for climate studies, the VI CMG algorithm uses QA filters to remove contaminated pixels from the input, 1km layers. The figure below highlights the presence of cloud contamination that persists in the 1km data layer.

Cloud Filtered vs Unfiltered VIs (3x3 Extracts) for EOS Coresite 'Tapajos'

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

2003

081

2003

097

2003

113

2003

129

2003

145

2003

161

2003

177

2003

193

2003

209

2003

225

2003

241

2003

257

2003

273

2003

289

2003

305

2003

321

2003

337

2003

353

2004

001

2004

017

Composite Period

VI V

alue

Unfiltered Avg NDVI (3x3)

Filtered NDVI (3x3)

Unfiltered Avg EVI (3x3)

Filtered EVI (3x3)

The plot below shows an example of what data may look like before and after the QA cloud filter has been applied. The QA filter reduces noise due to the CMG sub-pixel cloud contamination, particularly in the NDVI layer.

Cloud Filtering

EVI w/o Fill EVI w Fill

EVI Standard Deviation NDVI Standard Deviation

The MODIS CMG-VI is a high quality, reliable, and seamless product ready for use in vegetation-climate studies and biogoechemical growth models. The QA filtering and data fill methods further enable this product to be available in near-real-time.

Conclusions

http://tbrs.arizona.edu/

Processing FlowAt most 36, 1km pixelsdepending on latitude.

Inv Map/Projection

QA Filter

Complete CMGPixel at 0.05 deg

Filters pixels that are cloudy, mixed clouds, fill, or missing in input.

Inverse mapping / projection of input data to geographical coordinates.

Spatial Calc.

Reflectances averagedVIs recomputed Dominant QAStandard deviations “Climatology Record”

23 Avg. Composites One clean average year>1 high quality pixel

retained from QA filter

Figure 5: MOD13 CMG Processing flow.

The algorithm attempts to minimize clouds in the output product. To do so, it employsthree different averaging schemes. All input 1-km pixels (nominally 6x6) will either be allclear, all cloudy, or mixed. These averaging schemes work as follows: If all input pixels areclear, they will be all averaged to produce one output value; If all input pixels are cloudy,the pixel will be computed from the historical database; and, If the input pixels are mixed,only the clear pixels are averaged to produce one output value

The MOD13C1 uses the entire MODIS data record to calculate a reliable VI fill value incase input data is missing or deemed cloudy. The fill value is calculated from the averageof good data from all previous years CMGs of that composite period. It is mainly used forreplacing completely cloudy data, but is powerful enough to reliably fill in whole missingtiles (Fig. 6).

All fill strategies have their fallacies and pitfalls. In the Climatology Fill case, certainhighly dynamic regions may show discrepancies where fill values were used. This is mostobvious when missing input tiles are replaced, where edges may be visible. For pixels filleddue to high cloud contamination, this fill strategy will perform well.

The fill completes the two VI layers with data. Other layers will contain their respectivefill values, except data layer 11 (#1km pix used), which is set to 0, i.e., no good inputdata.


The 16-day 0.05-deg MOD13C1 VI product has 13 SDSs, as listed in Table 15.

25

MODIS VI PRODUCTION SCHEME

• To ensure the best quality, the MODIS program is periodically updated to integrate the latests advances and user needs.

• Upgrades are released as “Collections”. Current is Collection 5 (C5), based on a 16-day composite period (1 image each 16 days)

• Terra and Aqua are interleaved, then a combined 8-day product is possible. Considerations required.

MODIS VI C5 MAIN CHANGES

• Improvement of the Constrained View angle - Maximum Value Composite (CV - MVC) compositing method

• CV-MVC was modified to favor smaller view angles.

• MVC is used when all input days are cloudy

• Update of the EVI backup algorithm from SAVI to a 2–band EVI

• Added Composite day of the year and Pixel reliability output parameters

• Added the 0.05 deg CMG product series

MODIS VI C5 APPLICATIONS

Left: NDVI anomaly (Z units) for a given three-month period, computed from the aggregate1-km 16-day MOD13 product (9 yr); Right: corresponding aerosol load

extracted from the companion QA metadata

MOD13 PERSPECTIVES: TOWARDS C6

MODIS VI C6 CONSIDERATIONS

• Upgrades for next C6 include increasing time resolution to 8 days

• Also desired is using a standard SR parameter (derived from the Atmosphere section) for VI

• More studies on BRDF effects and the CV-MVC aggregating approach

• Impacts of these changes on the performance of the MODIS VI product for C6 are unknown.

MODIS VI C6 STUDIES

• We developed test data for studying proposed changes impacts

• A revised aggregating algorithm to reduce BRDF effects was proposed (reducing large VZ angles)

MODIS VI C6 STUDIES - WHY BRDF REVISITING?

VIs computed from the angle-corrected (zenith-normalized) ASRVN subset. Data correspond to the

GSFC validation site, dates 2002 241-256

123

!"!!#

!"$!#

!"%!#

!"&!#

!"'!#

!"(!#

!")!#

!"*!#

!"+!#

!",!#

-*!# -(!# -&!# -$!# $!# &!# (!# *!#

!"#

!$%&#'()*%#+,%)-#

./01#

201#

!"!!#

!"$!#

!"%!#

!"&!#

!"'!#

!"(!#

!")!#

!"*!#

!"+!#

!",!#

-*!# -(!# -&!# -$!# $!# &!# (!# *!#

!"#

!$%&#'()*%#+,%)-#

./01#2.3#

401#2.3#

FIGURE B.5: VIs computed from the MODIS ASRVN data set. Left panel shows the VIsderived from the actual MODIS geometry. Right panel shows the VIs computed from theangle-corrected (zenith-normalized) ASRVN subset. Data correspond to the GSFC site,dates 2002 241-256

similar performance of both algorithms for EVI (Figures B.6 and B.7) and NDVI (Figures

B.8 and B.9). BRDF biases were visible in both VI products and both algorithms, with EVI

showing decreasing values towards positive scan angles (forward scattering) while NDVI

slightly towards negative angles (backscattering). These findings were in agreement with

results obtained using single-site ASRVN data (Section B.5.1).

BRDF effects were more pronounced on the EVI than on the NDVI and more easily

appreciated on the Amazonian tile h12v09 (Figure B.6, right panels), covered basically by

undisturbed forests. Main VIs appeared concentrated in a narrow dynamic range (at about

0.52 EVI and 0.89 NDVI units), depicting a linear region of high density values in the VI vs.

VZ space. Prominent VI classes from the more heterogeneous h10v05 tile (Figure B.6, left

panels) had a larger dynamic range mainly starting from about 0.15 (0.20) VI units up to 0.7

(0.9) VI units for EVI (and NDVI respectively).

A more detailed observation evidenced slightly reduced angle span on C6 results for tile

MODIS VI C6 STUDIES - WHY BRDF REVISITING?

116

!"!!#

!"$!#

!"%!#

!"&!#

!"'!#

!"(!#

!")!#

!"*!#

!"+!#

!",!#

-*!# -(!# -&!# -$!# $!# &!# (!# *!#

!"#

!$%&#'()*%#+,%)-#

./01#

201#

!"!!#

!"$!#

!"%!#

!"&!#

!"'!#

!"(!#

!")!#

!"*!#

!"+!#

!",!#

-*!# -(!# -&!# -$!# $!# &!# (!# *!#

!"#

!$%&#'()*%#+,%)-#

./01#2.3#

401#2.3#

FIGURE B.5. VIs computed from the MODIS ASRVN data set. Left panel shows the VIsderived from the actual MODIS geometry. Right panel shows the VIs computed from theangle-corrected (zenith-normalized) ASRVN subset. Data correspond to the GSFC site,dates 2002 241-256

similar performance of both algorithms for EVI (Figures B.6 and B.7) and NDVI (Figures

B.8 and B.9). BRDF biases were visible in both VI products and both algorithms, with EVI

showing decreasing values towards positive scan angles (forward scattering) while NDVI

slightly towards negative angles (backscattering). These findings were in agreement with

results obtained using single-site ASRVN data (Section B.5.1).

BRDF effects were more pronounced on the EVI than on the NDVI and more easily

appreciated on the Amazonian tile h12v09 (Figure B.6, right panels), covered basically by

undisturbed forests. Main VIs appeared concentrated on a narrow dynamic range (at about

0.52 EVI and 0.89 NDVI units), depicting a linear region of high density values on the VI

vs. VZ space. Prominent VI classes from the more heterogeneous h10v05 tile (Figure B.6,

left panels) had a larger dynamic range mainly starting from about 0.15 (0.20) VI units up

to 0.7 (0.9) VI units for EVI (and NDVI respectively).

A more detailed observation evidenced slightly reduced angle span on C6 results for tile

BRDF effects on daily MODIS. Differential effects on VI bands. Large impact on VIs. Data correspond to the GSFC validation site, dates 2002 241-256

11

Unchanged angles were defined as |dVZ|< 1⇥ to avoid rounding differences.

2.5 Results and discussions

2.5.1 BRDF effects from ASRVN data

ASRVN reflectance data corresponding to the instantaneous MODIS acquisition geometry

were mapped against the VZ angle for the study site GSFC. All three VI-related bands

showed significant bias as a function of the acquisition angle (Figure 2.4). Reflectance

values were larger towards negative angles (backscattering direction) and decreased towards

positive angles (forward scattering direction). The NIR band exhibited the largest bias,

with values ranging from 0.2511 to 0.3599, followed by the blue band (0 to 0.0374) and

lastly the red band (0.0232 to 0.0418). In relative units, the NIR varied ±19% from the

estimated unbiased (zenith) value, while the blue and red bands diverged ±77% and ±29%

respectively.

!"!!#

!"!$#

!"%!#

!"%$#

!"&!#

!"&$#

!"'!#

!"'$#

!"(!#

)*!# )$!# )'!# )%!# %!# '!# $!# *!#

!"#"

$%&'

$"()*

+(

,-".(&'/0"()1"/+(

+,-#

./+#

012,#

!"

#"

$!"

$#"

%!"

%#"

&!"

&#"

'!"

'#"

()!" (#!" (&!" ($!" $!" &!" #!" )!"

!"#$%&#'()$*#+,&$-.

&+/$

0'&1$*#+,&$-.&+/$

FIGURE 2.4. Effective BRDF effects on individual MODIS reflectance bands (left figure)and corresponding sun zenith angles (right figure), as derived from the ASRVN for actualMODIS geometry. Data correspond to the GSFC site, dates 2002 241-256.

Should be constant independently of the view angle!

MODIS VI C6 STUDIES - BRDF RESULTS

127

-60 -40 -20 0 20 40 60

02000

4000

6000

8000

vz (deg)

C5

ND

VI (

x10k

)

��

��

��

��

��

��

� ��

��

��

��

��

-60 -40 -20 0 20 40 60

02000

4000

6000

8000

vz (deg)

C6

ND

VI (

x10k

)

FIGURE B.8: Distribution of NDVI vs. acquisition angle for two different MODIS tiles.Upper panels correspond to NDVI C5 and bottom panels to proxy NDVI C6. Left columncorrespond to tile h10v05 and right column correspond to tile h12v09. Both tiles correspondto compositing date 2002-177.

127

-60 -40 -20 0 20 40 600

2000

4000

6000

8000

vz (deg)

C5

ND

VI (

x10k

)

-60 -40 -20 0 20 40 60

02000

4000

6000

8000

vz (deg)

C6

ND

VI (

x10k

)

��

��

��

��

��

��

� ��

��

FIGURE B.8: Distribution of NDVI vs. acquisition angle for two different MODIS tiles.Upper panels correspond to NDVI C5 and bottom panels to proxy NDVI C6. Left columncorrespond to tile h10v05 and right column correspond to tile h12v09. Both tiles correspondto compositing date 2002-177.

Distribution of NDVI vs. acquisition angle for an Amazonian MODIS tile (h12v09). Left: NDVI C5; Right: proxy NDVI C6. Both tiles correspond to date 2002-177.

MODIS VI C6 STUDIES - BRDF RESULTS

Distribution of EVI vs. acquisition angle for MODIS tile h10v05. Left: EVI C5; Right: proxy EVI C6. Compositing date is 2002-177.

125

-60 -40 -20 0 20 40 60

02000

4000

6000

8000

vz (deg)

C5

EV

I (x1

0k)

-60 -40 -20 0 20 40 60

02000

4000

6000

8000

vz (deg)

C6

EV

I (x1

0k)

-60 -40 -20 0 20 40 60

02000

4000

6000

8000

vz (deg)

C6

EV

I (x1

0k)

FIGURE B.6: Distribution of EVI vs. acquisition angle for two different MODIS tiles.Upper panels show EVI C5 and bottom panels show proxy EVI C6. Left column showstile h10v05, located in SE US, and right column shows tile h12v09, located in EasternAmazonia. Both tiles correspond to compositing date 2002-177.

125

-60 -40 -20 0 20 40 60

02000

4000

6000

8000

vz (deg)

C5

EV

I (x1

0k)

-60 -40 -20 0 20 40 600

2000

4000

6000

8000

vz (deg)

C6

EV

I (x1

0k)

-60 -40 -20 0 20 40 60

02000

4000

6000

8000

vz (deg)

C6

EV

I (x1

0k)

FIGURE B.6: Distribution of EVI vs. acquisition angle for two different MODIS tiles.Upper panels show EVI C5 and bottom panels show proxy EVI C6. Left column showstile h10v05, located in SE US, and right column shows tile h12v09, located in EasternAmazonia. Both tiles correspond to compositing date 2002-177.

MODIS VI C6 STUDIES - FLUX GPP CORRELATION

62

1500 2000 2500 3000 3500 4000 4500

02

46

8

EVI C5 (x10k)

GPP

(gC

m−2

d−1 )

0 10 20 30 40 50

0.55

0.60

0.65

0.70

0.75

0.80

0.85

R2 GPP

vs

EVI

MODIS footprint (km2)

GPP vs EVI C5fit lineR2 EVI C5 (16d)R2 EVI C6 (8d)R2 EVI C6 (16d)

FIGURE A.15. As in Figure A.13, but showing a better correlation for proxy 16-day C6relative 16-day C5. Data from Skukuzq - Kruger National Park, ZA (ZA-Kru) site.Correlation recovers for 16-day “C6” re-

composites

16-day “C6”

16-day C5

8-day “C6”

76

In general, mean differences were small. In this dynamic range, there was a general

bias towards negative C6 - C5 differences, with the start of season fairly around zero and

the peak time marginally negative. The end of the season had a small bias toward negative

values, meaning that the growing season was detected a little sooner in C6 than in C5. The

length of the growing season consequently showed reduced values for the majority of sites,

indicating slightly smaller season lengths for C6 than for C5. The EVI values at the peak of

the season were smaller for C6 than for C5 for the majority of sites (Figure A.26)

●

●

●

●

●

●●

●

●

Start End Length Peak time Peak val

−1.5

−1.0

−0.5

0.0

0.5

1.0

1.5

16−d

ay p

erio

d

−600

−400

−200

020

040

060

0

EVI (

10k−

scal

ed)

FIGURE A.26. Distribution of mean difference values in seasonality parameters as derivedfrom MOD13 C5 and proxy MOD13 C6 (differences were computed as C6 - C5). Parametersincluded are: date of start of green season (Start), end of season (End), length of season(Length), peak of season (Peak time), and EVI value at the peak of season (Peak val). Datescorrespond to 16-day units as defined by the MOD13 compositing scheme. Peak valuescorrespond to difference in 10k-scaled EVI units.

A.6.3.2 Sensitivity analysis to the MOD09 aggregation approach

The observed differences between collections are a result of the intrinsically different

MOD13 compositing algorithms, but also a function of the approach followed to aggregate

MODIS VI C6 STUDIES - SEASONALITY EFFECTS

Small increase in SOSSmall reduction on EOS and LOS and on peak EVI

Green season

SOS = Start of SeasonEOS = End of SeasonLOS = Length of Season

MODIS VI C6 STUDIES - SOME CONCLUDING REMARKS

• 8-day “C6” EVI showed reduced EVI-GPP correlation. Most likely cause is the reduction in the compositing period (i.e. less probability of clear days).

• 16-day “C6” EVI algorithm restored EVI-GPP correlation to C5 levels

• Some impact on seasonality parameters: slight to moderate reduction on EOS, LOS and peak EVI value


• BRDF effects were evidenced on MODIS data using daily ASRVN reference data

• BRDF effects act different on individual B, R and NIR bands, hence affecting VIs

• EVI and NDVI are affected differently: EVI is more biased and in the opposite direction than NDVI

• BRDF effects are also present on the standard MODIS VI product

• Proxy C6 results showed reduced VZ angles than current C5, but it is also affected. Lack of good data?


IMPORTANT

• These studies were conducted using a proposed new aggregating algorithm for further reducing BRDF effects

• Actual C6 algorithm to be implemented may be different for a number of reasons

THANK YOU

Questions?

MODIS Vegetation Indices (MOD13)

Education

Transcript of MODIS Vegetation Indices (MOD13)