Journal of Wind Engineering & Industrial Aerodynamics 168 (2017) 247–259

Contents lists available at ScienceDirect

Journal of Wind Engineering & Industrial Aerodynamics

journal homepage: www.elsevier .com/locate/ jweia

Effectiveness of an array of porous fences to reduce sand flux: OceanoDunes, Oceano CA

J.A. Gillies a,*, V. Etyemezian b, G. Nikolich b, R. Glick c, P. Rowland d, T. Pesce d, M. Skinner d

a Division of Atmospheric Sciences, Desert Research Institute, 2215 Raggio Parkway, Reno, NV 89512, USAb Division of Atmospheric Sciences, Desert Research Institute, 755 E. Flamingo Rd., Las Vegas, NV 89119, USAc Oceano Dunes State Vehicular Recreation Area, California State Parks, 340 James Way Suite 270, Pismo Beach, CA 93449, USAd Coastal San Luis Resource Conservation District, 1203 Main St. Suite B., Morro Bay, CA 93442, USA

A R T I C L E I N F O

Keywords:Sand fencesSand transport affected by porous fencesLocal PM10 affected by porous fences

* Corresponding author.E-mail addresses: jack.gillies@dri.edu (J.A. Gillies),

prowland@coastalcrd.org (P. Rowland), tpesce@coastalcr

http://dx.doi.org/10.1016/j.jweia.2017.06.015Received 11 April 2016; Received in revised form 1 May

0167-6105/© 2017 Elsevier Ltd. All rights reserved.

A B S T R A C T

Arrays of porous fences, ≈12 ha in 2014 and ≈ 15 ha in 2015, were constructed in the Oceano Dunes StateVehicular Recreation Area to evaluate their effectiveness to reduce sand flux and affect local PM10 dust con-centrations. The 1.22 m high, �244 m long, ≈50% porous fences were placed 10 fence heights apart in 2014 and 7in 2015. Measurements of sand flux through the arrays indicated that it diminished exponentially with increasingdistance, reaching equilibrium at ≈93 fence heights for the 10 h spacing and ≈27 fence heights for the 7 h spacing.Fences spaced 7 h apart reduced sand flux for the entire area by 78%, and 86% for the area �27 h. Fences spacedat 10 h reduced sand flux for the entire area by 40%, and 56% for the area �93 h. PM10 monitoring upwind anddownwind of the array and in the absence of the array in 2015, indicated that the downwind PM10 concentrationwas less than the upwind for the fence array, whereas in the absence of fences PM10 increased in the downwinddirection over the same fetch distance, suggesting the presence of the fences was reducing the flux of PM10 fromwithin the fence array.

1. Introduction

The Oceano Dunes, part of a quaternary age coastal dune complex(Orme and Tchakerian, 1986) in California (Fig. 1), contains the OceanoDunes State Vehicular Recreation Area (ODSVRA) California State Parkconsisting of ≈500 ha of dune environment that allows off-road recrea-tional vehicle activity as well as ≈280 ha of dune preserve that only al-lows access by pedestrians. Under conditions of elevated wind speed,typically >5 m s�1 with a dominant westerly component as measured10 m above ground level (AGL), the threshold for sand transport isexceeded and once this occurs it is accompanied by dust emissions(Gillies and Etyemezian, 2014). For periods of wind erosion within thedune system that last for�6 h, air quality measurements made by the SanLuis Obispo County Air Pollution Control District (SLOCAPCD) down-wind of the eastern boundary of the park have been observed to exceedthe 24 h mean standard for the mass concentration of particulate matter�10 μm aerodynamic diameter (i.e., PM10) for both US EPA and Cali-fornia State air quality regulations (i.e., 150 μg m�3 US EPA, 50 μg m�3

California Air Resource Board). As part of an on-going effort to reduce

vic.etyemezian@dri.edu (V. Etyemd.org (T. Pesce), mskinner@coastalcr

2017; Accepted 19 June 2017

PM10 dust emissions that contribute to the violation of the standards andthat are associated with the saltating sand in the dune areas, controlmeasures are being evaluated.

Fences of various construction materials and design are used to con-trol the location and rate of erosion and deposition of sand and snow. Byextension, the control of saltation of sand-sized particles moving acrosssediments containing silt and clay-sized particles will also affect the dustemission process driven by the ballistic impact of these grains on to thesurface (Gomes et al., 1990; Shao, 2001; Shao et al., 1993). In 2014, 20parallel rows of sand fences 1.22 m high and ≈244 m long separated by adistance of 10 fence heights (≈12.2 m) and oriented perpendicular to theexpected dominant sand transporting wind direction (WNW) encom-passing ≈12 ha were established and left in place for three months. In2015, 35 parallel rows of the same type of fence, with the same orien-tation having a separation distance of seven fence heights (≈8.5 m) wereemplaced in a different area of the ODSVRA encompassing ≈15 ha andleft in place for four months. We report here on the effectiveness of thesearrays of sand fences to modulate sand flux over the area they coveredand how the array in 2015 influenced PM10 concentration from the

ezian), george.nikolich@dri.edu (G. Nikolich), Ronnie.Glick@parks.ca.gov (R. Glick),d.org (M. Skinner).

Fig. 1. Location of the Oceano Dunes in California, USA. The ODSVRA and dune preserve areas are within the marked rectangle.

J.A. Gillies et al. Journal of Wind Engineering & Industrial Aerodynamics 168 (2017) 247–259

upwind to downwind extent of the array. The expectation is that con-trolling sand flux will also reduce the flux of PM10 that contributes to theobserved levels that breach the air quality standards.

There is a rich scientific literature describing the effect that twodimensional barriers, solid and porous, have on airflow in front and inthe lee of these structures. Sand fences deployed for their intended pur-pose of modifying wind erosion processes are always porous. Li andSherman (2015) trace the formal research for sand fences to Bates(1911), with more thorough scientific-based investigations occurring inthe 1930s. Most of this research, even up to the present time, has focusedon single lengths of fence placed perpendicular to the flow (e.g., Lee andKim, 1999; Dong et al., 2011; Ping et al., 2013; Tsukahara et al., 2012;Zhang et al., 2015). Much less research has been published on air flowover multiple barriers (e.g., Woodruff and Zingg, 1955; Iqbal et al., 1977;McAneney and Judd, 1991; Wilson and Yee, 2003; Bitog et al., 2009). Liand Sherman (2015) note in their extensive review of the literaturerelated to sand fences that there have been few controlled fence experi-ments with sand transport.

Li and Sherman (2015) present a thorough review of the aero-dynamics and morphodynamics of sand fences. Here we provide a briefreview of the key aspects of what is known of flow and sand transport asaffected by fences with the reader referred to Li and Sherman (2015) fordetailed information.

The aerodynamics and morphodynamics of a fence depend on thegeometry of the fence design, with the elements of height, length, width,porosity, opening size/distribution/geometry, and orientation relative tothe wind being most important. Fence porosity (ε), the ratio of a fence'sopen area to its total area is considered the most important single

248

parameter controlling its performance and usually reported as a per-centage of open area (Bean et al., 1975). Permeability is a related char-acteristic that relates the velocity, u, of the air through the material(under the action of a pressure gradient) to the pressure gradient acrossthe material as defined by Darcy's Eq:

u ¼�K=μ

�∂p∂x

(1)

where K is the permeability with units of length squared, μ is the dynamicviscosity of the fluid, p the pressure and x the direction of flow. K rep-resents the ability of the material to transmit fluid through itself anddepends on the material porosity (ε) and the pore's characteristics.

The flow behind a single porous fence can be generalized as follows: anew boundary layer is created at the edge of the fence with large eddiesdeveloping just downwind and that boundary layer returning to equi-librium a certain distance in the lee of the fence generally at distances≫10 h, where h is fence height (Fig. 2). Immediately behind the fence isthe bleeding flow zone where individual jets are formed when the airpasses through the fence (Wilson et al., 1990), followed by the recircu-lating or standing eddy zone (Plate, 1971) wherein flow at the surface istowards the fence (opposite the above-fence flow). The blending zonesare found where the flow from the middle layer blends with the regionwhere the flow begins to establish the new boundary layer and betweenthe middle layer and the outer layer (Fig. 2).

Woodruff and Zingg (1955) provide a comparative analyses of airflow patterns about a multiple porous fence array (four fences, spaced at15 h) for both a field setting and a scaled wind-tunnel test. They observed

Fig. 2. Idealized flow zones for a single porous fence (adapted from Li and Sher-man, 2015).

J.A. Gillies et al. Journal of Wind Engineering & Industrial Aerodynamics 168 (2017) 247–259

that three or four successive barriers are insufficient to reach a maximalcumulative effect on the air flow. They also noted that successive fencescause the zone of maximum near-surface wind speed reduction to occurnearer to the second and third fences than the first. Their wind speedmeasurements also indicated that the maximum velocity fluctuationsbetween two successive fences occurred at 6 h, with the greatestdisturbance of the air flow at 0.125 h AGL at 2 h and 6 h behind thefirst barrier.

McAneney and Judd (1991) present a schematic of the wind fieldbehind a fence deep in an array (Fig. 3) on a relatively flat surface. Theyobserved from field measurements that most of the boundary-layeradjustment to the increased surface drag created by the presence of thefences occurred in a distance of eight fence rows, with an equilibriumcondition thereafter. Their definition of equilibrium is that the flow be-tween successive windbreaks is dynamically similar and that inputs ofturbulent energy are approximately balanced by losses between adjacent

Fig. 3. Schematic of the flow zones between adjacent fences in a multiple fence array wellremoved from the initial fence (adapted from McAneney and Judd, 1991).

249

fences. Wilson and Yee (2003) used McAneney and Judd (1991) resultsto test the performance of simple Reynolds-averaged Navier-Stokes(RANS) models to reproduce the turbulent kinetic energy (TKE) in awindbreak network. They found that all closure methods they consideredgave mediocre predictions and suggest this is due to the inability of themodels to effectively treat the parameterizing of the flow-fence interac-tion not only for the TKE and dissipation rate equations, but also themomentum equations.

Sand fences are deployed to reduce wind speed over erodible sur-faces, which should lower the probability of entrainment and reducetransport rates of sand in their lee. The degree of wind speed reduction inthe stream-wise direction defines their shelter distance (d), which can bereferenced to the distance downwind to reach a specified proportion ofthe undisturbed wind speed (e.g., 50%, 100% or the wind speed (or shearstress) required to achieve the threshold for transport). For fences thathave holes to create porosity (as opposed to vertical slats) shelter dis-tance is maximized for fence porosities between ε ¼ 0.30–0.40 (Cornelisand Gabriels, 2005; Dong et al., 2006; Jensen, 1954). Bitog et al. (2009),report from wind tunnel testing that ε¼ 0.2 created the greatest decreasein wind speed from the surface to the height of the fence. For slat-typefences Dong et al. (2011) found the maximum wind speed reductionoccurred at a ε ¼ 0.1.

Porosity and permeability of the fence also affect the transfer of sandthrough the fence. Fences of the same porosity can be constructed withholes of different size and shape (e.g., square, rectangular, round).Manohar and Bruun (1970) carried out wind tunnel tests that demon-strated for the fences they tested, the smallest slat gap had the greatesttrapping efficiency and that trapping efficiency decreases as gap sizeincreases. Field studies by Savage (1963) and Savage and Woodhouse(1968) also observed that fabric fences with smaller openings trappedmore sand than slat fences with the same porosity and height. To preventblocking through a pore or between slats, Hotta et al. (1987) suggest thatthe opening size needs to be � 10 times the diameter of the sand. Zhanget al. (2015) closely examined the effect that a porous fence had on thespeed and trajectory of sand particles after they had passed through thepores and reported that particle speed was slowed between 50% and 36%for particles of diameter 250 μm and 100 μm, respectively. They alsoobserved that this reduction in particle speed scaled as a function of graindiameter with smaller particles decreasing their speed to greater extentthan larger ones. Thus, the fence effectively reduces the impact velocityof sand grains in the region in its lee. This suggests that further manip-ulation of fence permeability characteristics affecting the passage ofsaltating sand particles could be used to improve fence performance.

Information on how the flux of sand is affected as it encountersmultiple fences of various separation distances is very limited in theliterature. Gill et al. (2003) deployed a staggered array of 22, 76 m longsand fence panels at Owens Lake, CA, spaced 55 fence heights apart andreported that sand transport was 88% lower than the flux upwind anddownwind of the array and for measurements of flux made adjacent tothe array in an area without sand fences. White et al. (2003) carried out awind tunnel experiment to evaluate how sand fences ε¼ 0.50 could affectsand transport. They measured the ratio of wind shear velocity tothreshold wind shear velocity as a function of downwind distance formultiple fences of varying spacing and developed sand transport-controlrelationships site-specific to Owens Lake, CA. Their study did not involvesand interacting with model fences.

Designing a wind tunnel study that involves multiple model fencesand saltating sand would be exceedingly difficult as there will be con-straints on scaling the size of the model fences to the size of the sandgrains and the length scales of the saltation process as well as thediameter of the sand with respect to the size of the pore openings in themodel fence. A further complication is that it has recently been suggestedthat saltation in wind tunnels is altered in its length scale characteristicscompared to unconstrained wind driven saltation (Sherman and Farrell,2008; Li and McKenna Neuman, 2012; Martin and Kok, 2017). Availablesaltation models such as COMSALT (Kok and Renno, 2009) are not

J.A. Gillies et al. Journal of Wind Engineering & Industrial Aerodynamics 168 (2017) 247–259

currently configured to account for the presence of macro-scale rough-ness elements, which limits their use to evaluate how an array of fenceswill modulate the saltation process.

Testing at the full scale in the atmospheric boundary level can bedifficult due to the vagaries of weather, wind conditions, and the nativespatial (Gares et al., 1996; Jackson et al., 2006) and temporal variability(Lee, 1987; Stout and Zobeck, 1997; Bauer et al., 1998) of sand transport,which can introduce high variability in measurements of mean quantities(e.g., saltation flux). Nevertheless, it offers the best opportunity toobserve the interaction of saltating sand with porous fences in theabsence of boundary-layer and scaling effects that are present in windtunnel testing.

2. Material and methods

2.1. Sand fence array design

As we were unaware at the planning stage of any theoretical orempirical relationships to guide the engineering of a multiple sand fencearray (e.g., White et al., 2003) to meet a design objective for sand fluxreduction, the following rationale was used. For construction purposesand cost considerations the choice of fencing material was a flexibleplastic mesh 1.22 high (h), with ovoid holes ≈8.2 cm by ≈ 3.7 cm holesand 209 holes per 1 m2 of mesh, ε ≈ 0.56 (Fig. 4). In order to evaluatehow this fencing could influence sand transport we made use of therelationship of Bradley and Mulhearn (1983) that shows the reduction inshear velocity, u* (Prandtl, 1935), in the lee of a porous (ε ¼ 0.50) fencecan be defined by:

u*leeu*

¼ 0:201 lnðNDÞ þ 0:429 (2)

Fig. 4. A close up of pores in the sand fence (top panel) and a ≈ 4.9 m

250

where u*lee is the shear velocity at positions downwind of the fence, u* isthe shear velocity without the fence present, and ND is normalized dis-tance (distance downwind of fence/fence height). For a threshold shearvelocity, u*t, we assumed the typical value for fine sand of 0.3 m s�1

(u10m ≈ 8 m s�1). For a 1.22 m high sand fence when the regionalu* ¼ 0.3 m s�1, the distance downwind of the fence where sand motioncommences will be at ≈15 h. This distance decreases as regional u* in-creases. The effectiveness of fences to reduce sand flux for fences placedat 6 h, 8 h, 10 h, and 15 h can be evaluated based on how far past thefence u*lee reaches u*t and the mean shear stress in the zone that extendsto the point of full recovery (i.e., u*lee=u* ¼ 1). Based on the fraction ofsurface that would be below threshold at a regional u* ¼ 0.4 m s�1

(u10m ≈ 10 m s�1), and hence provide a 50% reduction in sand transportbetween two consecutive fences the spacing of 10 h was chosen as thefence spacing design criteria for 2014.

The fence arrays were constructed by ODSVRA personnel. To anchorthe fences 2.4 m long, 0.13 diameter wooden posts were driven into thesand approximately 12.2 m apart. A screen made of 3.175 mm diameterwire interlaced to make holes 0.203 m by 0.152 m, with 36 holes per1 m2 was strung continuously between the wooden posts. The plasticfence material was subsequently affixed to both sides of the wire mesh,offset slightly to achieve ε ≈ 0.5.

In 2014, 20 fence rows ≈244 m long and spaced ≈10 h (12.2 m) apartwere deployed in a roughly rectangular shape within the ODSVRA(Fig. 5) creating an area of ≈12 ha defined by the perimeter. For 2015,using the same fencing method and materials, 35 fence rows wereestablished in a different location creating an area of ≈15 ha (Fig. 5).Spacing between the fences was changed to ≈7 h to evaluate how thiscould further affect the sand flux reduction characteristics of the array.The irregular shape was due to the topography and the need to not

length of fence placed on a dune at the ODSVRA (bottom panel).

Fig. 5. The position of the sand fence array 2014 in the ODSVRA (orange square, large panel) and 2015 array (gray polygon, large panel). Inset panel shows details of the 2015 sand fencearray with fences demarcated with parallel lines, the transects of BSNEs as red lines and ID labels A, B, and C and the position of the E-BAMS: E1, E2, E3, E4, and the anemometer/windvane measurement locations, T1 and T2. The blue line marks the park boundary. (For interpretation of the references to colour in this figure legend, the reader is referred to the webversion of this article.)

J.A. Gillies et al. Journal of Wind Engineering & Industrial Aerodynamics 168 (2017) 247–259

251

Fig. 6. The position of the sand traps in 2014 and 2015. Dashed lines indicate fences andnote that the figure is not drawn to scale and not all the fences are shown for 2015.

J.A. Gillies et al. Journal of Wind Engineering & Industrial Aerodynamics 168 (2017) 247–259

impinge upon a key travel route for vehicles operating within theODSVRA, especially emergency access vehicles. Due to the constraints ofoperating large machinery within the undulating dune topography (e.g.,steep slip faces and sides), some of the fence rows (in both years) werediscontinuous. In 2014, 2015, the fraction of area where fences could notbe emplaced due to terrain constraints represented ≈10% and ≈20%,respectively, of the total area defined by the perimeter of the arrays.

2.2. Instrumentation

In 2014, two types of sand traps were used to measure sand fluxexterior and interior to the array. Thirty-six Big Spring #8 (BSNE) traps(Fryrear, 1986) and 12 Cox Sand Catchers (CSC) (Gillies et al., 2015)were placed into the fence array as illustrated in Fig. 6. In 2015, 54 BSNEstyle traps were placed into the array as illustrated in Fig. 6. For bothyears of measurement, the bottom of the trap orifice was set at 0.15 m.

The placement of the traps in the array in 2014 created five lineartransects through the array. The array of samplers in the north and southpositions were ≈40 m from the edge of the array with a separation of≈40 m between the three instrument array transects interior to the northand south lines. A BSNE and CSC trap were placed 1.22 m in front of thewestern-most (upwind) sand fence for each sampling transect. In be-tween successive fences one to three traps were placed at 1 h (≈1.22 m),5 h (6.10 m), and 9 h (10.98 m). The middle transect had the highestnumber of traps (19) and the north, south and middle transects had trapsemplaced beyond the last fence.

CSC style traps were not used in 2015 as they proved to be toodifficult to maintain a consistent orifice height under the sand fluxconditions encountered, especially upwind of the array. A differentpattern of traps was used in 2015. Three linear transects one with 18and two with 20 BSNE traps, placed within eight consecutive fenceswere established within the array and designated with the letters A, B,and C. Transect A was the furthest north and C the furthest south(Figs. 5 and 6). Transect A consisted of two BSNE traps ≈1.22 m upwindof the first fence in the array, separated in the north-south directionby ≈2 m, followed by 16 BSNEs placed perpendicular to the fencealignment with one or three BSNEs placed between consecutive fences.Transect B consisted of two BSNE traps ≈1.22 m upwind of the firstfence in the array, separated in the north-south direction by ≈2 m,followed by 18 BSNEs placed perpendicular to the fence alignment withone or three BSNEs placed between consecutive fences. Transect C wasestablished at the 13th fence row from the leading western fence, toprovide characterization of sand flux patterns deep into the array. TwoBSNE traps were placed ≈1.22 m west of the 13th fence, separated inthe north-south direction by ≈ 2 m, followed by 18 BSNEs placedperpendicular to the fence alignment with one or three BSNEs placedbetween consecutive fences. For 2015, the general pattern for posi-tioning the BSNEs between two consecutive fences was to place them at1 h (≈1.22 m), 3 h (≈3.66 m), and 6 h (≈7.32 m) measured from thewestern fence. In the case where only one BSNE was placed betweenconsecutive fences it was located at the 6 h position. Typically sandtransport events occur at the dunes between 10 a.m. and 6 p.m. Thesand in the traps was emptied into labelled Ziploc® bags the morningafter a transport event and retained for weighing. Samples were driedand weighed to 0.01 g precision.

In 2015 the effect of the sand fence array to modulate the associateddust emissions was examined by placing PM10 monitors (E-BAM, MetOne Instruments, Inc., Grants Pass OR) at four locations in the vicinity ofthe sand fence array (Fig. 5) for May through July. Two samplers werelocated at the fence array. The first (E1) was located just upwind of theleading (western) edge of the array and the second (E2) at the trailing(eastern) edge of the array and aligned with each other along the lineperpendicular to the fence direction, separated by a distance of ≈375 m.A wind vane and anemometer downwind of the array was also positionedwith the E-BAM located at the downwind edge of the array from May toJuly 31, 2015 (T1, Fig. 5). The second pair of PM10 monitors (E3 and E4)

252

was located approximately ≈490 m to the south separated by ≈ 300 maligned along the same azimuth, but with only open sand between them(Fig. 5). A wind vane and anemometer were positioned with the E-BAMlocated furthest to the east from June 9 to July 31, 2015 (T2, Fig. 5).

3. Results

3.1. Sand transport event conditions

In 2014 between 04-09-2014 and 06-30-2014, 29 separate sandtransport events were sampled with the sand traps. Of note is that for thedates 4-18-2014, 4-19-2014, 4-21-2014, 6-21-2014, and 6-22-2014, themass of sand collected in the traps was low (total trap mass [5 BSNE traps

J.A. Gillies et al. Journal of Wind Engineering & Industrial Aerodynamics 168 (2017) 247–259

combined] upwind and exterior to the fences <20 g). In 2015 between04-14-2015 and 06-25-2015, 22 separate sand transport events weresampled with the sand traps.

With the inclusion of the PM10 monitoring in 2015 a record of thewind speed and direction at the fence array was obtained for the daysrepresented by sand trap samples. In 2015, for hourly mean wind speedand direction at 10 m for mean speed �5 m s�1 (assumed threshold forsand transport/dust emissions) the wind direction range was282.5�–312.5�, which was �9.5� to 20.5� from perpendicular to theorientation of the fence array. The percentage occurrence of hourly meanwind direction for mean hourly wind speed �5 m s�1 is shown in Fig. 7.The winds were more frequently shifted to the north of the perpendiculardirection of 292�. The hourly mean wind speed range was 5 m s�1 to9.6 m s�1 for the sand collection periods.

3.2. Change in sand flux as a function of distance into the array

Although sand flux changes with elevation above the surface (e.g.,Shao and Raupach, 1992; Gillies et al., 2013), relative flux among mea-surement positions can be evaluated using a one-height measurement ifthe surface sediment is homogeneous sand (Gillies et al., 2015), which isthe case at all trap locations. The accumulated mass in the sand trapsfollowing a sand transport event was used to calculate the NormalizedSand Flux (NSF). NSF is defined as the ratio of sand flux at a measurementlocation within the roughness array divided by a measurement externalto the roughness on the upwind side.

To examine how the sand flux was affected by the presence of thesand fences, mean NSF (the mean of all BSNE sand traps between twoadjacent fence rows/mean of all sand traps exterior and upwind of thearray) was plotted as a function of ND for each transport event in 2014and 2015. For 2014, these data show a high degree of scatter especiallyfor ND � 29 (Fig. 8). The best fit regression line for NSF as a function ofND shown in Fig. 8 is for ND � 99. Mean NSF appears to generally sta-bilize at ND � 93. In 2015 the two transects A and B (Fig. 9) exhibit highvariability in NSF from the leading edge of the array to ND ≈20. There is,however, for the 2015 data a more defined pattern of decreasing NSFwith increasing ND at similar rates of exponential decay (Fig. 9) fortransects A and B. The best fit regression lines for NSF as a function of NDshown in Fig. 9 for ND � 27, as in both transects ND appears to havereached an equilibrium value at this position. For transect A, the meanNSF at ND ¼ 6 was also removed before fitting the exponential rela-tionship due to its extremely high mean and standard deviation. Fortransect C (Fig. 9), there was no obvious change in NSF (referenced to themean flux of all traps upwind of transects A and B) with increasing ND asthe air flow and saltation flux transitioned from the large gap area to thewest into the next eight fences.

Fig. 7. The percent frequency of wind direction (5� bins) for mean hourly wind speed�5 m s�1 during the 2015 monitoring period when sand trap samples were collected.

253

3.3. Mean sand flux in the array

For both years the sand flux reduction is characterized by two zones: ahighly variable zone that extends to a certain ND followed by a zone oflower variability through to the end of the array. For 2014, the zone ofhigh flux variability was between ND 6–93 and the zone of decreased fluxvariability was ND � 93. The mean NSF values and their respectivestandard deviation values are provided in Table 1. The events with verylow sand catches with very low sand catches (04-18-14, 04-19-14, 04-21-14, 06-21-14, and 06-22-14) were not included in the calculation of themean NSF values.

In 2015 the mean NSF values for the zones defined by ND 6–20 (highvariability of NSF), and ND � 27 (low variability of NSF) for the threetransects are presented in Table 1. In 2015, in the zone of higher fluxvariability and where NSF decreases exponentially with increasing ND(i.e., transects A and B) the mean NSF was lower by approximately onehalf compared to 2014, although with a very high standard deviation(±2.06). For the zone of lower flux variability in 2015, the mean NSF waslower by a factor of 3.4 (just BSNE data) compared to 2014.

3.4. PM10 concentration changes at the array and exterior to the array

To examine how the presence of the fence array may be affecting theairborne concentration of PM10 the hourly mean data from the E-BAMswere filtered by wind speed (�3 m s�1) and wind directions approachingthe array through 22.5� either side of normal (i.e., 270�–315�) using themeteorological data collected at position T1 for E1 and E2 (Fig. 5), andfor E3 and E4 for dates prior to 06-09-2015, and at position T2 from 07 to09-2015 through to 07-31-2015. The relationship between mean PM10and mean hourly wind speed for 1 m s�1 bins from 3.5 to 11.5 m s�1 areplotted for May, June, and July 2015 in Fig. 10. Fig. 10 shows that PM10responds to wind speed in a non-linear fashion and the threshold windspeed for dust emission lies between 5 m s�1 and 6 m s�1. Along with themean PM10 concentration for each 1 m s�1 wind speed bin, the best fitpower function equation is shown for the wind speed bins �5.5 m s�1.

During May 2015, a decrease in PM10 concentration was observedbetween E1 and E2 when the winds were above threshold with the dif-ference dependent on wind speed. For the wind speed bin 5.5 m s�1 E2was 26% lower than E1. That difference decreased to 23% for the10.5 m s�1 bin, based on the best fit regression relationships (Fig. 10). ForJune, E2 was lower than E1 for equivalent wind speed bins for the meanhourly PM10 data pairs, but the regression-derived data values suggestequivalence. For July there appears to be no appreciable difference inPM10 between E1 and E2.

For the pair of E-BAM samplers E3 and E4, which are south of thearray and have no fences between them the differences in the meanhourly PM10 as a function of mean hourly wind speed for May, June, andJuly are also shown in Fig. 10. For May, this pair of samplers showshigher PM10 values for the sampler further to the east with the E4sampler ≈26% higher than the E3 sampler for the 5.5 m s�1 mean hourlywind speed bin increasing to 45% higher for the 10.5 m s�1 wind speedbin. For June and July, for equivalent hourly wind speed bins, E4 ishigher than E3, but the difference is ≈ 5% at lower wind speedsincreasing to ≈13% at the highest wind speed bin of 10.5 m s�1.

4. Discussion

4.1. Sand fence effectiveness to reduce sand flux

The sand fence arrays constructed in the ODSVRA represent thelargest test areas to date for evaluating the effectiveness of multi-fencearrays to modulate the flux of saltating sand and the associated dustemissions in a coastal dune environment. The mean NSF weighted by therelative areas of the zones of high flux variability (ND � 93, NSF ¼ 0.86)and lower variability (ND > 93, NSF ¼ 0.44) gave a sand flux reductionfor 2014, of NSF ¼ 0.60 (±0.20), which is 10% higher than the design

Fig. 8. Relationship between mean Normalized Sand Flux (all events with total BSNE sand catch >20 g and all traps at the same ND position) and Normalized Distance for 2014. Error barsrepresent the standard deviation of the mean NSF value.

J.A. Gillies et al. Journal of Wind Engineering & Industrial Aerodynamics 168 (2017) 247–259

target of 0.5, although the target was exceeded for 62% of the area. In2015, the sand fence array was not rectangular and contained areaswhere fences could not be emplaced due to topographical constraints.Based on its irregular shape and the places where fences could not beemplaced, a better approximation of the sand flux reduction by the entirearray needs to take this into account. The mean NSF weighted by therelative areas of the zones of high flux variability (ND� 20, NSF¼ 0.49),lower variability (ND � 27, NSF ¼ 0.12), and for the gap areas withoutcontinuous fences (NSF ¼ 0.62), gives an overall effectiveness of sandflux reduction for 2015, of NSF ¼ 0.27 (±0.22). The value of NSF for thegap areas was calculated by dividing the mean mass of the two traps tothe west of the fence marking the beginning of transect C by the mean ofthe four mass measurements upwind of transects A and B for each of the23 events and then taking the mean of these 23 values. By moving thedistance between consecutive sand fences from 10 h to 7 h, the effec-tiveness of the sand fence array to reduce sand flux was increased by afactor of 2.2.

As noted by previous researchers (e.g., Gares et al., 1996; Jacksonet al., 2006; Stout and Zobeck, 1997) the saltation process, even onrelatively simple (i.e., flat) sand surfaces possesses significant variabilityacross space and through time. Gares et al. (1996) showed span-wisetransport variability on the order of 25% over 15-min intervals. Jack-son et al. (2006) found transport variability commonly exceeded 150%over a span-wise distance of 5 m. It is this inherent variability in wind-driven sediment transport, and the effect of the fences to enhance tur-bulence (Wilson and Yee, 2003) and increase maximum velocity fluctu-ations (Woodruff and Zingg, 1955) as the flow encounters the fences thatgives rise to the high levels of variability of sand flux reported here.

The 2015 NSF data, binned by their relative position betweenconsecutive fences (i.e., 1 h, 3 h, and 6 h), show that the 3 h position has adiscernible change in the distribution of NSF magnitude compared to thedistributions for the traps placed at 1 h and 6 h (Fig. 11). The 3 h positionshows a distribution of NSF shifted towards higher values and is also thelocation where the highest NSF values are observed regardless of whichsequential fences are examined. The higher sand flux at 3 h coincideswith the transition at the surface from a zone of low turbulence to higher

254

turbulence in McAneney and Judd, (1991) equilibrium flow model(Fig. 3). Higher flux between fences is not associated with the 6 h posi-tion, where Woodruff and Zingg (1955) observed a zone of maximumvelocity fluctuations between successive fences.

The variability in sand flux is further enhanced by wind directionchanges that occur during a transport event. These directional shifts inwind alters the surface shear across the surface as the flow interacts withthe fences at oblique angles. The patterns of sand flux reduction as afunction of ND and the decrease in its variability with increasing NDpresented here provide, for the first time, a clearer picture of the effect ofmultiple porous barriers on sand transport and demonstrate a relationwith the flow conditions as described by McAneney and Judd (1991).

The mean NSF results can be compared with the predicted effec-tiveness of multiple sand fences to reduce sand transport as presented byWhite et al. (2003). Based on their derived sand transport controlequation for multiple fences:

Sand Transport Controlð%Þ ¼ A�Lh

�2

þ B�Lh

�þ 1 (3)

where L is distance between fences, and A and B are functions of the ratioof the shear velocity to the threshold shear velocity (i.e., u*=u*t), and withthe wind shear velocity a factor of 1.5 above threshold, for the 10 h arrayof 2014 and 7 h array for 2015, White et al., (2003) wind tunnel-derivedmodel predicts NSF ¼ 0.1 and NSF ¼ 0.07, respectively. For the 10 h and7 h arrays the estimated NSF in the zone where the sand flux was inequilibrium were both higher than the model predicted values (i.e., 0.44(±0.18) and 0.12 (±0.32)), which suggest the White et al. (2003) modelover-predicted effectiveness as a function of fence separation distance forthese fence arrays. White et al. (2003) based their model on predicteddownwind changes in shear stress behind fences and did not actuallymeasure sand flux as part of their experiments, which may in part explainsome of the discrepancy between the field measurements and the windtunnel model. White et al., (2003) model was also developed for windsperpendicular to the array, whereas this was not the case for our fieldsite. Another distinct difference between the wind tunnel and the field

Fig. 9. Relationship between mean Normalized Sand Flux (all events) and all BSNE traps at the same ND position between fences for transects A (white triangle) and B (black diamond)(top panel) and separately for transect C (bottom panel) for 2015. Error bars represent the standard deviation of the mean NSF value. For Transect A (top panel), the data for ND ¼ 6 werenot included in the least squares regression. For transect C (bottom panel) numbers in parentheses on the x-axis represent ND as measured from the first fence in the array.

Table 1NSF characteristics for 2014 and 2015.

Transect Mean NSF Std. Dev NSF Mean NSF Std. Dev NSF

2014 ND < 93 ND < 93 ND > 93 ND > 93A, B, C combined [BSNEa] 0.86 0.93 0.44 0.18A, B, C combined [CSCb] 0.73 0.49 0.56 0.34

2015 ND ≤ 20 ND ≤ 20 ND > 20 ND > 20A 0.76 2.84 0.09 0.21B 0.20 0.35 0.17 0.44C 0.09 0.09 0.12 0.33

A, B 0.48 2.06 0.13 0.36A, B, C combinedc 0.12 0.32

a Data are from the BSNE-style traps.b Data are from the CSC-style traps.c All C data used for the ND > 20 case.

J.A. Gillies et al. Journal of Wind Engineering & Industrial Aerodynamics 168 (2017) 247–259

255

Fig. 10. Relationships between mean hourly PM10 and mean hourly wind speed (binned to 1 m s�1 increments) for instrument pairs E1 (gray diamond), E2 (gray triangle) [left panels],and E3 (gray square) and E4 (gray circle) [right panels]. Error bars represent the standard deviation of the mean for the mean hourly PM10 associated with each wind speed bin. Data werefiltered for wind speed �3 m s�1 and wind directions approaching the array through 45� from normal (i.e., 270�

–315�). The best fit regression lines are for wind speed �5.5 m s�1 (i.e.,above threshold wind conditions). The white square and white circle in the E3 and E4 May data signify only a single datum for the wind speed bin.

J.A. Gillies et al. Journal of Wind Engineering & Industrial Aerodynamics 168 (2017) 247–259

situation is the topographic relief associated with the dunes may beplaying a role not accounted for in the model. White et al., (2003) modelis based on the developing internal boundary layer that contains thefences on a flat surface, whereas on the dunes the fences are placed wherethere is considerable topographic relief, which may be creating alongwith the sand fences, multiple internal boundary layer conditions that themodel does not take into account.

The results of Gill et al. (2003) are more difficult to reconcile with theresults of this study. Gill et al. (2003) report, for fences 1.22 m and 1.6 mhigh, spaced ND ¼ 55 apart, a reduction in sand transport that is 88%lower (i.e., NSF ¼ 0.12) within the sand fences than sand transportexterior to their array. According to the White et al. (2003) model, theexpectation would be a 60% reduction. Based on the results presented

256

here and observations during transport events at the ODSVRA, it isdifficult to imagine that a between fence spacing of ND ¼ 55 would havehad any measureable effect on sand flux under the conditions present atthe ODSVRA, except within close proximity to a fence. The test site of Gillet al. (2003), which was on the Owens (dry) Lake CA, was described asbeing a thin sand sheet overlying a crusted playa surface, which mayaccount in part for the difference as compared to the ODSVRA conditions.The sand supply at Owens Lake was much more restricted than theessentially limitless supply at the ODSVRA. Gill et al. (2003) note thatthere are periods when the surface at Owens Lake was affected by crustdevelopment, which would also potentially interfere with sand transportprocesses and the magnitude of the flux for different wind conditions.

Fig. 11. The distribution of NSF at the three measurement positions between fences for all transects, for 2015.

J.A. Gillies et al. Journal of Wind Engineering & Industrial Aerodynamics 168 (2017) 247–259

4.2. Changes in sand flux as it enters the fence array

The decrease in sand flux as it encounters a step change in roughnesshas been reported by Gillies et al. (2006, 2015) and Gillies and Lancaster(2013) for field-scale studies using arrays of large (>0.3 m high) spatiallydistributed roughness elements. These papers reported that NSF de-creases exponentially with increasing ND, reaching an equilibrium valuesome distance into the array. The rate of change of the reduction in NSFwith ND, as observed in those studies, was a function of the roughnessdensity, λ (λ¼ [element frontal area�#elements]/surface area occupiedby the elements]) of the array. In this experiment the 2-dimesional fencesalso exhibit an exponential reduction in NSF as a function of ND. Basedon the results of McAneney and Judd (1991) it could be expected that thesand flux would approach equilibrium conditions at eight fence heights(ND¼ 48 for their experiment), as this is the distance at which they claimthe flow conditions will be in equilibrium with the fences. The mea-surements from this study suggest that a greater distance is required forthe sand flux to come into equilibrium for the fences spaced at 10 h(ND ≈ 97), and a lesser distance for the fences spaced at 7 h (ND ≈ 27).

For fences approximately perpendicular to the direction of the sandbearing winds, the rate of change of sand flux appears to be controlled bythe inter-fence spacing. A 30% decrease in the spacing distance betweenfences resulted in NSF decreasing by a factor ≈2.2, and the distance toreach equilibrium NSF decreased by a factor of ≈4.7. In terms of hori-zontal distance the change is from ≈114 m to ≈31 m. The two-dimensional fences cannot be characterized by λ, but comparing thetwo roughness types in terms of an equal NSF of 0.22 demonstrates theirrelative effects on the rate of change of sand flux with distance. Toachieve NSF ¼ 0.22 would require λ ≈0.031 for an array of 0.3 m highroughness elements (Gillies et al., 2015) with the distance to reach anequilibrium flux of ≈51 m (ND ≈170). At least for this level of controleffectiveness it appears that the 2-dimesional fences modulate the sandflux at a faster rate than the 3-dimensional roughness array.

The decrease in the distance to reach equilibrium flux for the fencesmay be due to several factors. Plausible mechanisms that could explainthis result include the interaction of the sand with the porous fence andan effect related to element height. A porous barrier, as described byZhang et al., (2015), effectively slows the sand as it passes through by upto 50%, thus reducing its kinetic energy as well. For the solid roughnesselements used by Gillies et al. (2015) some sand will impact the elementand rebound thus offering a reduction in sand speed as well, but sand will

257

also be swept around the edges of the solid elements where local shear isincreased (McKenna Neuman and B�edard, 2015), which may slow theloss of kinetic energy of the systemwith downwind distance compared tothe fence array. There is also some probability that sand particles can besaltating at heights greater than 0.3 m, but the probability of sand sal-tating higher than the 1.22 m high fences is much lower. Another dif-ference is that a fence slows the flow along its entire length forcing itupwards. In the case of solid roughness elements in a staggered array, theflow can re-route between elements for several rows into the array beforeit starts to acquire near-equilibrium characteristics.

4.3. Sand fence array effect on PM10 concentrations

The delta changes in PM10 concentration between pairs of measure-ments at the fence array and south of the fence array absent any fences(Fig. 11) demonstrate most clearly for May 2015 that the fence arrayaffected the PM10 in the atmosphere at this localized scale. In the absenceof the sand fence array over approximately equal fetch distances, theconcentration of dust at the reference sampling height above the surfaceincreased with downwind distance, although the observed relative dif-ference decreased for similar wind speeds between consecutive months.

An increase in PM10 concentration between E3 and E4 is consistentwith the physics of wind erosion, where a fetch effect has been theorized(Owen presented in Gillette et al., 1996) and observed (Gillette et al.,1996; Gillette, 1999). The fetch effect is characterized as the increase ofsediment mass flux with increasing downwind distance. The causes ofthis effect are multiple and include an avalanching mechanism, aero-dynamic feedback, and a soil resistance mechanism (Gillette et al., 1996).At the ODSVRA sites there is an additional process that can cause anincrease of sand flux in the west to east direction which is the increase inelevation from the shoreline through the dune complex. This rise inelevation compresses the streamlines of the air flow increasing its speed,shear stress, and hence the sand flux (Wiggs et al., 1996). Gillies andEtyemezian (2014) observed this increase of wind speed (10 m AGL)measurements along linear transects. In the absence of physical ob-structions (such as fence arrays) PM10 should increase with downwinddistance as the dust emissions scale with the sand flux (Shao, 2000).

Our results suggest that the dust emissions associated with the sal-tating sand were modulated substantially by the fences due to reducedsand flux. Gillette et al., (1997) observed that for sandy soils the ratio ofdust flux to saltation flux was relatively invariant. This linear coupling

J.A. Gillies et al. Journal of Wind Engineering & Industrial Aerodynamics 168 (2017) 247–259

between sand flux and dust emissions suggests that the amount of dustflux reduction in the fence array is equivalent to the sand flux reduction.The reason that the change in PM10 concentration does not match theobserved sand flux reductions is that the site is embedded within a verylarge area of active dust emissions when the wind speed is abovethreshold for sand transport. The dispersion of dust from the surroundingdune environment is likely impacting the downwind monitor E2. Thatthe difference is still observable in this restricted area suggests that theflux of PM10 from within the fences was reduced substantially.

5. Conclusions

The effectiveness of two arrays of sand fences of the same height, butwith different inter-fence spacing, to reduce sand flux was examined attest sites in the ODSVRA, Oceano, CA. Measurements of sand fluxthrough the arrays indicated that fences with a spacing of 7 h reducedsand flux by 78% considering the entire area within the perimeter, and by86% for the areas that were �27 fence heights from the leading edge ofthe array. For fences spaced at 10 h sand flux was controlled for the entirearea by 40%. At distances �93 fence heights, which was 62% of the totalarea, sand flux was reduced by 56%. There was a substantial increase incontrol effectiveness achieved by moving the fences closer together. Thedegree of control achieved by the 7 h and 10 h fence spacings was lessthan predicted by the model developed by White et al., (2003), but thismay be due, in part, to the variable wind direction and complex topog-raphy of the dunes compared to the flat surface the model was developedfor. If the model of White et al., (2003) were to be used, caution is sug-gested for application in a dune environment and the assumptions usedshould be very conservative.

As previously observed for arrays of three dimensional roughnesselements and similarly for sand fences, sand flux decreases exponentiallywith increasing distance into the array until an equilibrium flux isreached. The data from this study of multiple fences suggests that the rateof change of the decrease in sand flux is controlled by the spacing be-tween the fences. The 1.22 m high fences, for an equivalent amount ofsand flux reduction that could be achieved using the roughness elementarray approach, reduce the sand flux as it enters the fences at a faster ratethan 0.3 m high roughness elements. Scaled by the height of theobstruction the distance to the equilibrium flux condition occurs atND ≈ 25 for the fences and ND ≈ 170 for the roughness elements. Thissuggests that the porosity and height of the obstructions may be animportant control on the distance into the roughness where the sand fluxcomes into equilibrium and may figure into design considerations forusing roughness, 2-dimensional or 3-dimensional, to achieve sand fluxreduction targets.

Although we did not directly measure the reduction in dust fluxassociated with the presence of the sand fences, the PM10 concentrationdata measured upwind and downwind of the fence array in 2015indicate that the downwind concentration was up to 26% less than theupwind for May 2015 and equal or slightly lower than the upwindconcentration for equivalent wind speeds in June and July 2015. In theabsence of fences for an approximately equivalent separation distance,PM10 was observed to increase in the downwind direction. This in-crease is due to the fetch effect described by Gillette (1999) and alsodriven by the increasing wind speed between the two measurementpositions due to the increasing elevation of the dune complex, whichcompresses and accelerates the airflow. An increase in wind speed andwind shear will increase the sand flux and the associated dust emis-sions. The actual effectiveness of the fences to reduce dust flux over thearea they occupied may have been obscured by its location within alarge uncontrolled area of dust emissions. A conservative estimate ofthe reduction in dust emissions attributable to the fence arrays is that isequivalent to the reduction achieved in the sand flux, as for sandy soilsit has been observed that the ratio of dust flux to sand flux is relativelystable and independent of wind speed.

258

Acknowledgements

We would like to acknowledge the efforts of ODSVRA personnelinvolved in the sand fence array construction. Funding from CaliforniaState Parks (Contract #C1553026) to J.A. Gillies, V. Etyemezian and G.Nikolich is gratefully acknowledged.

References

Bates, C.G., 1911. Windbreaks: Their Influence and Value. Report. US Dept. ofAgriculture, Forest Service.

Bauer, B.O., Yi, J., Namikas, S.L., Sherman, D.L., 1998. Event detection and conditionalaveraging in unsteady aeolian systems. J. Arid Environ. 39, 345–375.

Bean, A., Alperi, R.W., Federer, C.A., 1975. A method for categorizing shelterbeltporosity. Agric. Meteorol. 14, 417–429.

Bitog, J.P., Lee, I.-B., Shin, M.-H., Hong, H.-S., Hwang, H.-S., Seo, I.-H., Yoo, J.-I.,Kwon, K.-S., Kim, Y.-H., Han, J.-W., 2009. Numerical simulation of an array of fencesin Saemangeum reclaimed land. Atmos. Environ. 43, 4612–4621. http://dx.doi.org/10.1016/j.atmosenv.2009.05.050.

Bradley, E.F., Mulhearn, P.J., 1983. Development of velocity and shear stress distributionsin the wake of a porous shelter fence. J. Wind Eng. Ind. Aerod. 15, 145–156.

Cornelis, W.M., Gabriels, D., 2005. Optimal windbreak design for wind erosion control.J. Arid Environ. 61, 315–332.

Dong, Z., Qian, G., Luo, W., Wang, H., 2006. Threshold velocity for wind erosion: theeffects of porous fences. Environ. Geol. 51 (3), 471–475. http://dx.doi.org/10.1007/s00254-006-0343-9.

Dong, Z., Luo, W., Qian, G.Q., Wang, H.T., 2011. Evaluating the optimal porosity of fencesfor reducing wind erosion. Sci. Cold Arid Regions 3, 1–2.

Fryrear, D.W., 1986. A field dust sampler. J. Soil Water Conservation 41, 117–120.Gares, P.A., Davidson-Arnott, R.G.D., Bauer, B.O., Sherman, D.J., Carter, R.W.G.,

Jackson, D.W.T., Nordstrom, K.F., 1996. Alongshore variations in aeolian sedimenttransport, Carrick Finn Strand, Ireland. J. Coast. Res. 12, 673–682.

Gill, T.E., Cahill, T.A., Copeland, S.A., White, B.R., 2003. Sand fences for control of winderosion and dust emission at Owens Lake, CA; Full scale testing, field deployment,and evaluation of effectiveness. In: 11th International Conference on WindEngineering, Lubbock, TX, pp. 2773–2780.

Gillette, D.A., 1999. A qualitative geophysical explanation for “hot spot” dust emittingsource regions. Contrib. Atmos. Phys. 72 (1), 67–77.

Gillette, D.A., Herbert, G., Stockton, P.H., Owen, P.R., 1996. Causes of the fetch effect inwind erosion. Earth Surf. Process. Landforms 21, 641–659.

Gillette, D.A., Fryrear, D.W., Gill, T.E., Ley, T., Cahill, T.A., Gearhart, E.A., 1997. Relationof vertical flux of particles smaller than 10 μm to total aeolian horizontal mass flux atOwens Lake. J. Geophys. Res. 102 (D22), 26,009–026,015.

Gillies, J.A., Etyemezian, V., 2014. Wind and PM10 Characteristics at the ODSVRA fromthe 2013 Assessment Monitoring Network. Report prepared for California StateParks. ODSVRA, Oceano, CA and OHMVR Division, Sacramento, CA, 10-07-2014.

Gillies, J.A., Green, H., McCarley-Holder, G., Grimm, S., Howard, C., Barbieri, N., Ono, D.,Schade, T., 2015. Using solid element roughness to control sand movement: KeelerDunes, Keeler, California. Aeolian Res. 18, 35–46. http://dx.doi.org/10.1016/j.aeolia.2015.05.004.

Gillies, J.A., Lancaster, N., 2013. Large roughness element effects on sand transport,Oceano Dunes, California. Earth Surf. Process. Landforms 38 (8), 785–792. http://dx.doi.org/10.1002/esp.3317.

Gillies, J.A., Nickling, W.G., King, J., 2006. Aeolian sediment transport through largepatches of roughness in the atmospheric inertial sublayer. J. Geophys. Res. - EarthSurf. 111, F02006 http://dx.doi.org/10.1029/2005JF000434.

Gillies, J.A., Nickling, W.G., Tilson, M., 2013. Frequency, magnitude, and characteristicsof aeolian sediment transport: McMurdo Dry Valleys, Antarctica. J. Geophys. Res.Earth Surf. 118, 1–19. http://dx.doi.org/10.1029/2012JF002473.

Gomes, L., Bergametti, G., Coude-Gaussen, G., Rognon, P., 1990. Submicron desert dusts:a sandblasting process. J. Geophys. Res. 95, 13927–13935.

Hotta, S., Kraus, N.C., Horikawa, K., 1987. Function of sand fences in controlling wind-blown sand. Coast. Sediments 87, 772–787.

Iqbal, M., Khatry, A.K., Seguin, B., 1977. A study of the roughness effects of multiplewindbreaks. Boundary-Layer Meteorol. 11 (2), 187–203.

Jackson, N.L., Sherman, D.J., Hesp, P.A., Klein, A.H.F., Ballasteros, F., Nordstrom, K.F.,2006. Small-scale spatial variations in aeolian sediment transport on a fine-sandbeach. J. Coast. Res. 379–383 (Special Issue 39).

Jensen, M., 1954. Shelter Effect: Investigations into the Aerodynamics of Shelter and itsEffects on Climate and Crops. Danish Technical Press, Copenhagen.

Kok, J.F., Renno, N.O., 2009. A comprehensive numerical model of steady state saltation(COMSALT). J. Geophys. Res.-Atmos. 114, D17204 http://dx.doi.org/10.1029/2009JD011702.

Lee, J.A., 1987. A field experiment on the role of small scale wind gustiness in aeoliansand transport. Earth Surf. Process. Landf. 12 (3), 331–335. http://dx.doi.org/10.1002/esp.3290120311.

Lee, S.-J., Kim, H.-B., 1999. Laboratory measurements of velocity and turbulence fieldbehind porous fences. J. Wind Eng. Ind. Aerod. 80, 311–326.

Li, B., McKenna Neuman, C., 2012. Boundary-layer turbulence characteristics duringaeolian saltation. Geophys. Res. Lett. 39, L11402 http://dx.doi.org/10.1029/2012GL052234.

Li, B., Sherman, D.J., 2015. Aerodynamics and morphodynamics of sand fences: a review.Aeolian Res. 17, 33–48. http://dx.doi.org/10.1016/j.aeolia.2014.11.005.

J.A. Gillies et al. Journal of Wind Engineering & Industrial Aerodynamics 168 (2017) 247–259

Manohar, M., Bruun, P., 1970. Mechanics of dune growth by sand fences. Dock Harb.Auth. 51, 243–252.

Martin, R., Kok, J., 2017. Linear Scaling of Wind-driven Sand Flux with Shear StressarXiv:1609.09458 [physics.geo-ph].

McAneney, K.J., Judd, M.J., 1991. Multiple windbreaks: an aeolean ensemble. Boundary-Layer Meteorol. 54 (1–2), 129–146.

McKenna Neuman, C., B�edard, O., 2015. A wind tunnel study of flow structureadjustment on deformable sand beds containing a surface-mounted obstacle.J. Geophys. Res. Earth Surf. 120 (9), 1824–1840. http://dx.doi.org/10.1002/2015jf003475.

Orme, A.R., Tchakerian, V.P., 1986. Quaternary dunes of the Pacific coast of California.In: Nickling, W.G. (Ed.), Aeolian Geomorphology. Allen and Unwin, London,pp. 149–175.

Ping, L.V., Dong, Z.B., Luo, W.Y., Qian, G.Q., 2013. The pressure-field characteristicsaround porous wind fences: results of a wind tunnel study. Environ. Earth Sci. 68 (4),947–953. http://dx.doi.org/10.1007/s12665-012-1797-6.

Plate, E.J., 1971. The aerodynamics of shelter belts. Agric. Meteorol. 8, 203–222.Prandtl, L., 1935. The mechanics of viscous fluids. In: Durand, F. (Ed.), Aerodynamic

Theory. Springer-Verlag, Berlin, pp. 57–100.Savage, R.P., 1963. Experimental study of dune building with sand fences. In: Proceedings

of the 8th Conference on Coastal Engineering, Mexico City, pp. 380–396.Savage, R.P., Woodhouse, W.W., 1968. Creation and stabilization of coastal barrier dunes.

In: Proceedings of the 11th Conference on Coastal Engineering. American Society ofCivil Engineers, New York, pp. 617–685.

Shao, Y., 2000. Physics and Modelling of Wind Erosion. Kluwer Academic Publishers,Dordrecht, p. 393.

Shao, Y., 2001. A model for mineral dust emission. J. Geophys. Res. 106 (D17),20239–20254.

259

Shao, Y., Raupach, M.R., 1992. The overshoot and equilibration of saltation. J. Geophys.Res. - Atmos. 97 (D18), 20559–20564.

Shao, Y., Raupach, M.R., Findlater, P.A., 1993. The effect of saltation bombardment onthe entrainment of dust by wind. J. Geophys. Res. 98D, 12719–12726.

Sherman, D., Farrell, E., 2008. Aerodynamic roughness lengths over movable beds:Comparison of wind tunnel and field data. J. Geophys. Res. Atmos. 113 (F2) http://dx.doi.org/10.1029/2007JF000784.

Stout, J.E., Zobeck, T.M., 1997. Intermittent saltation. Sedimentology 44 (5), 959–970.Tsukahara, T., Sakamoto, Y., Aoshima, D., Yamamoto, M., Kawaguchi, Y., 2012.

Visualization and laser measurements on the flow field and sand movement on sanddunes with porous fences. Exp. Fluids 52 (4), 877–890. http://dx.doi.org/10.1007/s00348-011-1157-4.

White, B.R., Cho, H.M., Gill, T.E., 2003. Sand fences for control of wind erosion and dustemission at Owens Lake, CA: design and wind tunnel simulation. In: 11thInternational Conference on Wind Engineering, Lubbock TX.

Wiggs, G.F.S., Livingstone, I., Warren, A., 1996. The role of streamline curvature in sanddune dynamics: evidence from field and wind tunnel measurements. Geomorphology17 (1–3), 29–46.

Wilson, J.D., Yee, E., 2003. Calculation of winds disturbed by an array of fences. Agric.For. Meteorol. 115, 31–50.

Wilson, J.D., Swaters, G.E., Ustina, F., 1990. A perturbation analysis of turbulent flowthrough a porous barrier. Q. J. R. Meteorol. Soc. 116, 989–1004.

Woodruff, N.P., Zingg, A.W., 1955. A comparative analysis of wind-tunnel andatmospheric airflow patterns about single and successive barriers. Trans. Am.Geophys. Union 36 (2), 203–208.

Zhang, N., Lee, S.J., Chen, T.-G., 2015. Trajectories of saltating sand particles behind aporous fence. Geomorphology 228, 608–616. http://dx.doi.org/10.1016/j.geomorph.2014.10.028.