gou et al-2015-groundwater (1)

Mapping Potential Groundwater-DependentEcosystems for Sustainable Managementby Si Gou1, Susana Gonzales1, and Gretchen R. Miller2

AbstractEcosystems which rely on either the surface expression or subsurface presence of groundwater are known as groundwater-

dependent ecosystems (GDEs). A comprehensive inventory of GDE locations at an appropriate management scale is a necessaryfirst-step for sustainable management of supporting aquifers; however, this information is unavailable for most areas of concern.To address this gap, this study created a two-step algorithm which analyzed existing geospatial and remote sensing data to identifypotential GDEs at both state/province and aquifer/basin scales. At the state/province scale, a geospatial information system (GIS)database was constructed for Texas, including climate, topography, hydrology, and ecology data. From these data, a GDE index wascalculated, which combined vegetative and hydrological indicators. The results indicated that central Texas, particularly the EdwardsAquifer region, had highest potential to host GDEs. Next, an aquifer/basin scale remote sensing-based algorithm was created toprovide more detailed maps of GDEs in the Edwards Aquifer region. This algorithm used Landsat ETM+ and MODIS images to trackthe changes of NDVI for each vegetation pixel. The NDVI dynamics were used to identify the vegetation with high potential touse groundwatersuch plants remain high NDVI during extended dry periods and also exhibit low seasonal and inter-annual NDVIchanges between dry and wet seasons/years. The results indicated that 8% of natural vegetation was very likely using groundwater.Of the potential GDEs identified, 75% were located on shallow soil averaging 45 cm in depth. The dominant GDE species were liveoak, ashe juniper, and mesquite.

IntroductionMultiple ecosystems in semi-arid regions are likely

to be stressed by the increasing pressures of climate, landuse, and population change (Baldwin et al. 2003; Smithet al. 2003). Groundwater-dependent ecosystems (GDEs),typically vegetative communities that rely on the surfaceor subsurface expression of groundwater, are especiallysensitive to these changes. Greater understanding of GDEswill enable more informed management strategies aschanges are observed. While previous studies have beenable to identify and monitor individual GDEs, muchremains to be done to document their collective spatialdistribution, influence on the water balance, and responseto changing water availability. The purpose of this studyis to develop a method to map GDEs using existinggeospatial and remote sensing datasets and apply thismethod to create state and aquifer scale maps in Texas.

1Zachry Department of Civil Engineering, Texas A&MUniversity, College Station, TX 77843-3136.

2Corresponding author: Zachry Department of Civil Engineer-ing, Texas A&M University, College Station, TX 77843-3136; (979)862-2581; fax: (979) 862-1542; [email protected]

Received June 2013, accepted January 2014.2014, National GroundWater Association.doi: 10.1111/gwat.12169

GDEs in the United States occur in a number ofpotentially stressed ecoregions, particularly the GreatBasin in Nevada (Naumburg et al. 2005; Steinwand et al.2006), the Edwards Plateau in Texas (Jackson et al. 1999;McElrone et al. 2004), the Sonoran Desert in Arizona(Scott et al. 2008), and in California, the Owens Valley(Elmore et al. 2003; Goedhart and Pataki 2010) and thefoothills (Miller et al. 2010) and riparian meadows of theSierra Nevadas (Loheide et al. 2009; Loheide and Gore-lick 2005, 2007; Lowry et al. 2011). Two distinct types ofGDEs are significant for sustainable groundwater devel-opment (Eamus et al. 2006): (1) biota living in and aroundsprings, groundwater-fed wetlands, and riparian zones, allof which rely on the surface expression of groundwater;and (2) vegetation with root access to deeper (more than2 m) stores of water which require the subsurface pres-ence of groundwater. A third class of GDEs, subsurfacemicrobial communities, is also recognized. While thesepopulations and the processes they facilitate are environ-mentally significant, they are substantially different intheir character and thus will not be included in this study.We considered the vegetation belongs to the first two typesof GDEs in this study, and have referred to them as low-land GDEs and upland GDEs, correspondingly. Eamuset al. (2006) also suggested that the vegetation may relyon groundwater if it meets one or more of the followingcriteria: (1) the groundwater or the capillary fringe is

NGWA.org Vol. 53, No. 1GroundwaterJanuary-February 2015 (pages 99110) 99

Table 1Selected Studies of Known Phreatophytes in the Southwestern United States

Source Location Species

McElrone et al. (2003) Texas, karst uplands Juniper (Juniperus ashei ) and live oak (Quercus fusiformis)Wilcox et al. (2006) Texas, riparian zones Saltcedar (Tamarix chinensis, Tamarix ramosissima)

Texas, karst uplands Ashe juniper (Juniperus ashei Buchholz )Texas, deep soils Mesquite (Prosopis glandulosa Torr. var. glandulosa)

Schaeffer et al. (2000) Arizona, stream channels Willow (Salix goodingii Ball ) and cottonwood (Populusfremontii Wats .)

Scott et al. (2008) Arizona, savannas Velvet mesquite (Prosopis velutina Woot .)Miller et al. (2010) California, savannas Blue Oak (Quercus douglasii )Steinwand et al. (2006) California, scrubland Rabbitbrush (Chysothamnus nauseosus), Nevada saltbush

(Atriplex lentiformis ssp. torreyi ), and greasewood (Sarcobatusvermiculatus)

Loheide and Gorelick(2005, 2007)

California, riparian zones Wet-meadow vegetation (sedges, rushes, and some otherherbaceous species)

Martinet et al. (2009) New Mexico, riparian zones Cottonwood (Populous deltoids spp. wislizeni), salt cedar(Tamarix chinesis), Russian olive (Elaeagnus angustifolia),mesquite (Prosopis pubescens), saltbush (Atriplex L. spp.)

within the vegetation rooting depth; (2) significant surfaceexpressions of groundwater are present (e.g., springs),and the vegetation associated with these expressions isdifferent from other nearby vegetation; (3) the vegetation,or a portion of it, remains green and physiologically activeduring extended dry periods; (4) the vegetation showsslow seasonal changes in leaf area index while others donot; (5) the vegetation exhibits lower water stress than thenearby vegetation without accessing groundwater; (6) theannual transpiration is significantly larger than the annualrainfall and run-on rate; and (7) daily or seasonal changesin groundwater depths are observed, not due to lateral flowor percolation.

Table 1 shows the wide variety of known phreato-phyte species in the southwestern United States. WithinTexas, at least six phreatophyte species have been identi-fied; two in karst upland areas (juniper and live oak), onein deep upland soils (mesquite), and three lining riparianzones (willow, salt cedar, and giant reed, Arundo donax ).

Two prior approaches have been used to predict thepresence or absence of GDEs in a given region. The mostcommon is the creation of an index based on key factorslinking GDEs and abiotic factors, such as pedological,morphological, hydrological, and climate characteristics(Bertrand et al. 2012). On the basis of these existingdatasets, some studies created index values for smallwatersheds; for example, studies had been conducted atthe hydrologic unit code-12 (HUC-12, subwatershedsswith the average area of 40 mile2) scale (Howard andMerrifield 2010) and the HUC-6 (basins with the averagearea of 10,000 mile2) scale (Brown et al. 2010). Thismethod can highlight areas with high index values whichindicate the areas host large numbers of GDEs. The resultscan provide useful information to incorporate GDEsinto groundwater management at large scale. However,the previous GDE index systems had only consideredthe factors related to the lowland GDE types (gainingstreams, springs, riparian zones, potential groundwater-fed

wetlands, and perennial lakes) and did not include thefactors linked with the deep-rooted, upland phreatophytes.In addition, the index approach considered watershedscale areas of interest (e.g., HUC-12) as the smallestestimation units. Thus, the detailed information on GDEdistributions within a watershed was unavailable.

Alternate approaches directly identified potentialGDEs based on their own specific characteristics or behav-iors, such as relatively slower changes in their physiolog-ical activity than that of nearby, non-GDE plants. Variousremote sensing based indices, such as Normalized Dif-ference Vegetation Index (NDVI), Normalized DifferenceWater Index (NDWI), and MODIS Enhanced VegetationIndex (EVI), have been used to detect such GDE char-acteristics. On the basis of the changes in NDVI andNDWI response to water-limiting conditions, Barron et al.(2012) identified potential GDEs in Western Australiausing Landsat imagery. Dresel et al. (2010) combinedthree remote sensing measures, including the NDVI dataderived from Landsat images, the changes of MODIS EVI,and the remote sensing based classification, to identifyGDEs in Australia. GIS modeling can further improveremote sensing based GDE identification methods. Com-bining remote sensing derived NDVI, vegetation green-ness, and soil moisture, and the GIS modeled groundwaterand landscape wetness information, Munch and Conrad(2007) classified the potential GDEs for a 2400 km2 regionin South Africa.

These remote sensing based approaches can providemore detailed GDE distribution than the index systemapproach. However, the previous studies did not considerthe impacts of vegetation density and plant phenologyon NDVI dynamics. The potential biases need to beaddressed when NDVI dynamics was used to identifyplant groundwater use. The combination of various remotesensing measures may ameliorate these biases. In addition,the remote sensing based approaches usually needed highquality images combined with additional calculations and

100 S. Gou et al. Groundwater 53, no. 1: 99110 NGWA.org

careful user interpretation. Therefore, they were typicallyused only for GDE identification at small scales.

Our study aims to combine the best features of bothapproaches to address the different management needsfound at various scales and to produce a more holisticassessment. At the large state/province scale, a GIS-basedGDE index approach was used to generally identify whichareas contained considerable numbers of both lowlandand upland GDEs. The GDE index approach highlightedthe critical areas where water and ecosystem managersshould consider GDEs in their planning. For these criticalareas, a remote sensing-based approach was developedto provide more thorough information about the spatiallocation of GDEs. Thus, the objectives of this study areto: (1) develop a GIS-based method to estimate a GDEindex value for each subregion in Texas; (2) propose aremote sensing-based method to delineate detailed GDEdistributions for the area highlighted in objective (1);and (3) analyze the impacts of various factors on GDEdistribution, including vegetation types, soil depth, andlandforms. These mapping efforts represent a key steptowards providing groundwater managers and modelerswith the information they need to assess GDEs at differentscales.

MethodsWe mapped GDEs based on the criteria proposed

by Eamus et al. (2006) using a two-step approach: atthe state scale, a GIS-based method was first used tocalculate a GDE index for each state subdivision, thatis, groundwater management area (GMA) or hydrologicunit code-6 (HUC-6) sized watershed in Texas. We choseGMAs and HUCs for GDE index estimation because theseareas are the spatial scales used for groundwater andwatershed management in Texas (TWDB 2012). Usingpublically available data, the GIS-based method served asa screening tool to identify critical regions with a highpotential to host a significant number of GDEs. Next,at the aquifer/basin scale, a remote sensing method wasapplied to a critical region in order to identify ecosystemsthat exhibit the physiological hallmarks of groundwaterdependence. This method provided detailed, smaller scaleinformation on GDE distributions.

GDE Index Method for State/Province ScalesThe criteria proposed by Eamus et al. (2006), espe-

cially the first two criteria (see Introduction), mainly focuson two aspectswhether the groundwater is accessibleby vegetation, and whether the vegetations dynamicsare associated with the available groundwater. There-fore, we created a new GDE index system, which com-bined two categories of GDE indicatorsvegetative andhydrological. The vegetative indicators denoted the veg-etation with high potential to be GDEs based on ecosys-tem type, while the hydrological indicators identified theareas where groundwater is most likely to be accessedby ecosystems. To derive these indicators, a GIS databasewas established with a variety of geospatial information

on Texas topography, hydrology, and ecology, includingpreviously generated data on springs (Brune 1975; USGS2012a), wetlands (USFWS 2012; USGS 2012b), lan-duse/landcover (USGS 2012b), vegetation types (TPWD2012), base flow index (Wolock 2003), gaining/losingstreams (Slade et al. 2000), HUCs (USGS 2012a), andGMAs (TWDB 2012).

The vegetative indicators included representationof groundwater-fed wetlands and phreatophytes, whichrepresented dominant ecosystem types of lowland andupland GDEs. The lowland GDEs in the riparian zonesand around the springs were excluded in the estimationof the vegetative indicators, since our analysis suggestedthat their areas were insignificant when compared to thetotal area of wetlands and phreatophytes (see Results andDiscussions). Groundwater-fed wetlands were specifiedbased on wetland types in National Wetland Inventory ofU.S. Fish and Wildlife Service (2012) and included allnoncoastal wetland types: freshwater emergent wetlandand freshwater forested/shrub wetland. In some inlandareas not covered by the National Wetland Inventory,wetland locations were derived from the USGS NationalLand Cover Database (2012b), and the emergent herba-ceous wetlands and woody wetlands in these areas wereconsidered to be groundwater-fed wetlands. To identifypotential upland GDEs, the vegetation belonging tophreatophytic species was identified from vegetationcover data based on the list of species known to occur inthe southwestern United States (Table 1). The vegetativeindex was calculated for each state subdivision (GMA orHUC-6) (Equation 1).

Vegetative Index = Phreatophyte area + Wetland areaTotal subdivision area

(1)

A higher vegetative index value denoted that the areahad more of the ecosystem types that are likely to usegroundwater. However, in some areas, groundwater istoo deep to be accessed by these ecosystems. In thatcase, even though the dominant vegetation belongs tothe phreatophytic species, they may not be groundwaterdependent. Therefore, a hydrological indicator was intro-duced to show the areas where groundwater is accessibleby ecosystems. Ideally, the hydrological indicators shouldinclude the information on the location and depth of nearsurface water tables. However, this information is unavail-able in many regions. Instead, we used the USGS Base-flow Index (BFI) was used as a surrogate hydrologicalindicator of regional groundwater-surface water interac-tions. BFI is a measure of the contribution of baseflowto a streams overall flow and was produced by USGSbased on their streamgauging data (Wolock 2003; Wahland Wahl 2007). It also included some spring flows con-tributing to the streams (Wahl and Wahl 1995). A high BFIshows a high proportion of total flow coming from morereliable groundwater sources, which implies a high poten-tial that groundwater presented to land surface or watertable rised near land surface to contribute to streamflow.

NGWA.org S. Gou et al. Groundwater 53, no. 1: 99110 101

The USGS BFI data used in this study were point-based estimates created from and assigned to individualUSGS streamgauges, rather than the spatially interpolatedgrids that were also available (Wolock 2003). In somecases, a HUC-6 watershed contained multiple USGSstreamgauges with different BFI values, while in others,a HUC-6 watershed did not have any streamgauges orhad streamgauges not reporting BFI values. For thewatersheds with multiple USGS streamgauges, BFI valueswere averaged. For the watersheds without data, a BFIvalue was assigned based on the average BFI value ofthe larger HUC-4 watershed where the HUC-6 watershedresides (e.g., Upper Beaver, HUC 111001, was assignedthe average BFI value for North Canadian, HUC 1110).For each GMA, a BFI value was determined to be thearea weighted average of corresponding HUC-6s BFIvalues (Equation 2). The area of a specific HUC-6 withina certain GMA was calculated in ArcGIS.

GMAj _BFI

=

ni=1

HUCi_BFI Area of HUCi within GMAj

Total Area of GMAj(2)

Finally, a GDE Index was developed to integrateboth vegetative and hydrological indices. For each statesubdivision (GMA or HUC-6), the vegetative index wasmultiplied by the hydrological index (regional BFI ) tocalculate the GDE Index of each specific area:

GDE Index = Vegetative Index Hydrological Index(3)

The two indices were combined multiplicatively,such that if one index was not satisfied then anotherindex could not compensate for it. Both hydrological andvegetative indices need to be above zero in order foran area to be identified as potentially hosting a GDE.Multiplying the two indices yielded zero if one of theindices was not satisfied, which eliminated some directly.When both indices were above zero, those areas withhigher multiplication results implied a higher potential tohost GDEs. Efforts to sustainably manage groundwaterin areas with high GDE indices should focus attentionon these vulnerable ecosystems as potential groundwaterconsumers.

Remote Sensing-Based Method at Aquifer ScaleFor regions identified as highly likely to contain

GDEs, more accurate information about the spatial distri-bution of GDE was needed to support sustainable ground-water management. The exercise was not straightforward;many factors combined together to impact GDE distri-bution, such as plant characteristics, climate, soil, andgeology. Not all the plants belonging to the phreato-phytic species depend on groundwater. For example, ifa mesquite was on deep soil and the local precipitation

was adequate to support its water use, this mesquite mayonly rely on water stored in deep soil from rainfall events,rather than using groundwater. However, even in the sameclimate regime, another mesquite may be located on shal-low soil with a lower available water content. Undersuch conditions, a mesquite may need to access deepergroundwater to support its water use. Even though thetwo mesquites were the same species under the sameclimate, they may be different in groundwater depen-dency. Additionally, different, yet co-occurring, speciesmay have different levels of groundwater dependence. Forexample, if a live oak is located in upland area withshallow water table, this live oak may access ground-water using its deep roots, while shallow-rooted grassesaround it may not access groundwater directly. Therefore,we needed to develop a more complex method, based onremote sensing data, to detect the physiological signaturesof groundwater-dependent vegetation.

The criteria of Eamus et al. (2006) were also appliedto guide the GDE identification using remote sensing.Two of these criteria can be assessed by analyzing remotesensing data: (1) A proportion of the vegetation that usesgroundwater remains green and physiologically activeduring extended dry periods, and (2) the vegetation thataccesses groundwater exhibits lower seasonal changes inleaf area index than the other nearby vegetation does. Inaddition, Tweed et al. (2007) highlighted a third criterionfor GDE identification: (3) Vegetation with low inter-annual variability of vegetation photosynthetic activity islikely to access groundwater.

To assess these criteria remotely, the NDVI waschosen as an indicator. NDVI is widely used to monitorvegetation cover and biomass production. It is sensitiveto leaf area index change until a full vegetation coverhas been reached (Carlson and Ripley 1997) and providesuseful information about vegetation physiological functionunder clear weather conditions (Wang et al. 2004; Tweedet al. 2007). Two different remote sensing products fromLandsat 7 Enhanced Thematic Mapper (ETM+) andMODIS were used to relate vegetation NDVI variability togroundwater use (Figure 1). Landsat ETM+ has relativelyhigh spatial resolution, which helps to discern the finescale distribution of GDEs, while MODIS has relativelyhigh temporal resolution, enabling it to capture NDVIchanges of vegetation within short time periods.

We selected the Edwards Aquifer region, belong-ing to GMA 10, as case study area for the remotesensing-based methods. This area hosts three knownphreatophytic species: live oak (Quercus fusiformis), ashejuniper (Juniperus ashei ), and mesquite (Prosopis glandu-losa) (McElrone et al. 2004; Wilcox et al. 2006). Numer-ous springs appear along the Balcones fault zone. Thedata from National Land Cover Database (USGS 2012b)showed that the dominant plant functional types wereshrublands (48% of total natural vegetative areas), grass-lands (22%), evergreen forests (20%), deciduous forests(8%), and woody wetlands (1%). Data from Web SoilSurvey (USDA 2013) showed that 57% of the EdwardsAquifer region has shallow soils(average depth of 45 cm),


Criterion 1. Vegetationremains green andactive in dry season

Criterion 2. Vegetationexhibits low seasonalchanges in LAI

Criterion 3. Vegetationhas low inter-annualvariability

Landsat ETM+ imageof mid-summer ofYear 2002 (3030 m)

MODIS NDVI imagesevery 16 days for Year2011 (250250 m)

MODIS NDVI images inJuly from Year 2002 to2011 (250250 m)

Calculate NDVI valueRemove urban areas, water, farmlands andpasturesCluster into 5 groups using K-means fromlowest NDVI (1) to highest NDVI (5) (3030 m)

Calculate NDVI standard deviationRemove urban areas, water, farmlands andpasturesCluster into 5 groups using K-means fromlowest SD (5) to highest SD (1) (250250 m)

Calculate NDVI standard deviationRemove urban areas, water, farmlands andpasturesCluster into 5 groups using K-means fromlowest SD (5) to highest SD (1) (250250 m)

Add the threedatasets togetherCluster the resultsinto five GDElikelihood groupsusing K-means(3030 m)

Criteria Data Source

Resamplethe results

Data Processing Results

Figure 1. Flow chart for remote sensing method: Three criteria were used for the remote sensing based method, which wasused to identify GDEs at the aquifer/basin scale. For criterion 1, one Landsat ETM+ image from mid-summer in 2002 wasprocessed using the steps shown above; it indicated the vegetation with high NDVI values in dry season. For criteria 2 and 3,multiple MODIS NDVI images collected throughout 2011 and during July of 2002 to 2011were selected, in order to identifyvegetation with low seasonal and inter-annual changes in NDVI. The results from the three criteria were integrated togetherto classify each 30 30 m pixel into one of five GDE likelihood groups.

and the remaining 43% has deep soils (more than 200 cm).The region also has a subtropical to semi-arid climate.Precipitation is highly variable in time, but is generallyhighest in May and September. In July and August, theprecipitation is usually low, while the potential evapo-transpiration is high. The precipitation is out of phasewith potential evapotranspiration during this period. Thisimplies that GDEs are most likely to rely on groundwaterduring July and August. Therefore, satellite imagery fromJuly was used in the analyses relating to Criteria One andThree.

During the extended summer dry season, vegetationwith high NDVI values was considered to be physiologi-cally active, indicating that there was a high likelihood itwas using groundwater (Criterion One). To verify the firstcriterion remotely, Landsat ETM+ images (30 30 m)from July 2002 were used; these images were high qualityand relatively cloud-free. The NDVI value of each pixelwas calculated (Equation 4). Since we only consideredthe natural vegetation in this study, the pixels representingurban areas, water, farmlands, and pastures were removedfrom the NDVI results based on land cover data fromUSGS (2012b).

NDVIj =(NIRj Rj

)(NIRj + Rj

) (4)

where j is the j th vegetation pixel, NIRj and Rj refers tothe spectral reflectance measurements in the near-infraredand red regions, respectively.

We chose an unsupervised classification technique,K-means, to cluster the NDVI results into five groups.K-means is a widely used algorithm to automaticallyclassify the data into K clusters according to their sim-ilarity (MacQueen 1967). Unlike supervised classification

methods, this unsupervised classification technique doesnot need prior knowledge to define training sets. Instead,it attempts to find the underlying cluster structure auto-matically (Canty 2007), thus it was suitable for our regionof interest, which lacked previous studies of GDEs. Weconducted the K-means classification in ENVI software(version 4.8) (Canty 2007), which we used to classifythe pixels into five groups according to the similarity oftheir NDVI values. We then calculated the average NDVIvalue of the pixels in each group and assigned each groupa value from 1 (the group containing the lowest averageNDVI value) to 5 (highest NDVI group).

Vegetation exhibiting low seasonal changes in leafarea index over a whole year may also access groundwater(Criterion Two). For each vegetation pixel, the standarddeviation in NDVI across a year-long time series wascalculated by Equation 5:

SDj =1

n

nt=1

(NDVIt,j NDVImean,j

)2 (5)

where n is the number of time-series satellite images, jis the j th vegetation pixel, NDVIt,j is the NDVI valueof the j th pixel at time t , NDVImean is the mean NDVIvalue of the j th pixel for the n images. A low NDVIstandard deviation implies that the vegetation pixel hadslow changes in leaf area during the study period.

MODIS NDVI products (MOD13Q1, 250 250 m),collected every 16 d in Year 2011, were used to analyzethe seasonal NDVI changes over the whole year. Thedry year 2011 was chosen for the analysis because Year2009 and Year 2010 were also dry years, minimizing thepotential impacts of antecedent soil moisture on vegetationdynamics. As in the previous analyses, urban areas, water,


farmlands, and pastures were also removed. The K-meanstechnique was then applied to cluster the seasonal NDVIstandard deviation into five groups. Five groups wereassigned the values from 1 (the group with highest averageseasonal NDVI standard deviation) to 5 (lowest NDVISD). Higher group values indicated that the vegetationlocated within a pixel had relatively higher potential to beusing groundwater.

A similar method was used to identify the vegetationwith low inter-annual changes in leaf area index (CriterionThree). The MOD13Q1 data (250 250 m), from imagestaken in July for each year from 2002 to 2011, wereused to calculate the inter-annual NDVI standard deviationvalue for each natural vegetative pixel. The resultswere also clustered into five groups using the K-meansalgorithm and assigned from 1 (the group with highestaverage inter-annual NDVI standard deviation) to 5(lowest NDVI SD). Both results from the MODIS-basedanalyses in Criteria Two and Three were further resampledin ArcGIS to change the cell size from 250 250 m to30 30 m resolution to correlate to the spatial resolutionof Landsat ETM+.

Each criterion yielded a dataset containing potentiallyunique information to identify GDEs. However, eachcriterion still had its own disadvantages, which centeredon its biases in regard to certain plant functional types (seeResults and Discussion). To overcome these, we mergedall three datasets from Criteria One, Two, and Three usingthe raster calculator in GIS. The assigned values (1 to5) were summed for each pixel and the resulting sumhad the values ranging from 3 to 15. Using the K-meansalgorithm, these values were further classified into fiveGDE likelihood groups of the final resultsvery likelyto be GDEs (the group with highest average values), likelyto be GDEs, about as likely as not to be GDEs, unlikelyto be GDEs and very unlikely to be GDEs (the group withlowest average values).

Results and Discussions

GDE Index at the State/Province ScaleWe estimated the GDE Index for each GMA and

HUC-6 in Texas. Phreatophytes clustered in the middleregions of Texas, from the High Plains through the CentralGreat Plains and Edwards Plateau to the Southern TexasPlains. Live oak and mesquite were the two dominantphreatophyte species (Figure 2a). Other phreatophytes,including cottonwood, saltceder, and willow oak, werefound in riparian areas. The woody wetlands and theemergent herbaceous wetlands were mainly found ineastern Texas (Figure 2b). A large number of thesewetlands were located in riparian areas, where they maybe fed by shallow groundwater. The regions with thehighest vegetative index were located in central Texas.BFI values indicated that the streams in central andeastern Texas had high baseflow ratios (Figure 2c).Correspondingly, the highest hydrological index valueswere also found in the central Texas.

GMAs 7, 9, and 10 had the highest GDE index values,indicating they had the highest potential to contain GDEs(Figure 2d). The HUC-6 basins with the highest GDEindex values were the Colorado River and the NeucesRiver (Figure 2e). The BFI values in these regions rangedfrom 0.3 to 0.45, and they are underlain by a number ofmajor aquifers, including the Edwards, Edwards-Trinity,Trinity, Ogallala, Pecos Valley, and Seymour. Karstedcarbonate rocks and other permeable formations in theseareas are known to produce numerous springs, includingthe two largest: Comal and San Marcos. These areaswere almost fully covered by phreatophytic plant species,with wetlands scattered in the riparian areas and aroundthe large springs, making upland GDEs the dominanttype. Plans for sustainable groundwater management needto address the groundwater use of potential GDEs andthe risks of disturbances on GDEs, such as the landuse changes, groundwater over-extraction, and climatechange. In addition, managers in some specific areasneed to consider the influence of GDEs on public watersupplies, including the potential changes to groundwaterrecharge and baseflow that may result from their presenceor expansion (Wilcox 2002).

Remote Sensing-Based Results in the Edwards AquiferRegion

Results Using the Three Groundwater-Dependence CriteriaEach criterion captured the groundwater use potential

of different plant functional types. Table 2 shows thepercentages of each plant functional type, as classifiedinto likelihood groups (with 5 being the highest). If oneplant functional type had the largest portion in the highestlikelihood group and the smallest portion in the lowestlikelihood group, it was considered as the type has thehighest likelihood to use groundwater, such as the wetlandin Criterion One.

Criterion One identified the areas with high NDVIvalues in the dry summer (Figure 3a). The pixels in thehighest likelihood group (group 5) had an NDVI valuegreater than 0.5, similar to the 0.35 to 0.5 range suggestedby Barron et al. (2012). Wetlands were identified as theplant functional type most likely to use groundwater.In contrast, a large portion of the grasslands (78%) wasclassified into the lowest likelihood group. These grass-lands likely depend only on soil water. There were 10%of deciduous forests classified into the highest likelihoodgroup, while only 4% of evergreen forests were includedinto this group. The different results between deciduousand evergreen forests may because of vegetation densityrather than differences in their groundwater dependency;NDVI is tightly related to the vegetation density (Carlsonand Ripley 1997; Purevdorj et al. 1998). Van Auken et al.(1981) found that in central Texas, the evergreen forestshad significantly lower density and species richnessthan the deciduous forests. Therefore, Criterion One canexclude some non-GDEs with low NDVI in the dryseason because they did not access groundwater, butit may also ignore some GDEs with low NDVI due to


(a)

(d) (e)

(b) (c)

Figure 2. GDE index developed by integrating the vegetative and hydrological indicators (a, b, c) for each GMA (d) andHUC-6 (e) watershed. Darker areas indicate a higher likelihood of supporting significant numbers of GDEs.

Table 2Plant Functional Types Captured by Each Criterion

Percentage of Each Type (%)Criterion 1 Criterion 2 Criterion 3

Plant Functional TypesHighest

Likelihood (5)Lowest

Likelihood (1)Highest


Likelihood (1)Highest


Likelihood (1)

Deciduous forest 101 42 13 4 27 9Evergreen forest 4 43 23 2 36 6Shrubland 2 67 14 8 11 18Grasslands/herbaceous 2 78 9 16 10 26Wetlands 36 22 28 11 28 9Total vegetation covered areas 3 62 15 8 17 16

1Percentage was calculated as 10% = Deciduous Forest in Group 5/Total Deciduous Forest 100%. Other percentages were calculated in the same way.


(a)

(b)

(c) (f)

(e)

(d)

Figure 3. Remote sensing results using three GDE criteria:The left three figures show the individual results from thethree criteria, shown in blue. The areas with higher criterionvalues imply the higher probability of containing GDEs.These three results were synthesized together to generate thefive likelihood groups in Figure 4. The right three figuresshow the relationship between each two criteria, shown ingreen. The areas with the highest agreement imply the twocriteria had the same results in these areas.

their low density, such as the evergreen forests. Thesimilar problem occurred for the shrublands due to theirlow canopy coverage. Only 2% of total shrublands wereincluded in the highest likelihood group.

Criterion Two indicated areas with slow seasonalchanges in NDVI for a dry year (Figure 3b). As comparedto Criterion One, more areas (15% of the total vegetationcovered areas) were classified into the highest likelihoodgroup and much fewer areas (8%) were in the lowestlikelihood group (Table 2); obvious increases occurred inthe number of both evergreen forests and shrublands inthe highest likelihood group. Wetlands were still the typewith the highest potential to use groundwater, while grass-lands were still the type with the lowest groundwater usepotential. Criterion Two focused on the rate of change inNDVI rather than the NDVI value itself, which eliminatedthe impacts of vegetation density on NDVI in CriterionOne. The percentage of the deciduous forests in thehighest likelihood group (13%) only increased slightly.Due to their growth pattern and phenological stages, thedeciduous forests essentially exhibited faster seasonal

NDVI changes compared to the evergreen forests. There-fore, Criterion Two may ignore some GDEs with fasterseasonal NDVI changes due to their essential seasonalgrowth pattern rather than their dependence on groundwa-ter. However, Criterion Two still had its own advantageto capture the species which may access groundwater attimes outside of the dry season, while Criteria One andThree only analyzed the NDVI variability in dry periodsto distinguish vegetation with different water use patterns.

The third criterion analysis indicated areas with lowinter-annual changes in NDVI in dry seasons for multipleyears (Figure 3c). By using only the satellite imagescollected in July, each plant functional type in the imageswas at a consistent phenological stage for every year. Iteliminated the impacts of plant growth pattern on NDVIstandard deviation in Criterion Two. Focusing on theresponses of vegetation in various precipitation conditionsduring the dry season, Criterion Three segregated theeffects of annual variations in precipitation from theimpact of vegetation growth patterns and phenologicalstages. Compared to the results of Criterion Two, moredeciduous forests (27%) were classified into the highestlikelihood group. Grasslands still had the lowest potentialto use groundwater based on Criterion Three. Evergreenforests had the highest groundwater use potential inCriterion Three, rather than wetlands as identified inCriteria One and Two.

We analyzed the results to determine if the threecriteria each contained distinct information on GDEs.For one vegetation pixel, if two criteria yielded thesame results, we considered that these criteria had thehighest agreement. For example, if both Criteria One andTwo classified a pixel into Group 5, they had highestagreement in this pixel; if they classified the pixel intoGroups 1 and 5, they had the lowest agreement. Similaritymaps for each pair of criteria are shown in Figure 3 (partsd, e, and f). The results indicated that Criteria One andTwo had the highest agreement over wetlands, and thelowest agreement in shrublands. Therefore, if the resultsof Criteria One and Two were summed, their combinationwould strengthen the final results in wetlands and wouldhelp ameliorate the disadvantage of Criterion One inshrublands. Criteria One and Three also had the highestagreement in wetlands, but the lowest agreement inevergreen forests; Criteria Two and Three had the highestagreement in deciduous forests, but the lowest agreementin the wetlands. In general, Criteria One and Two hadthe most distinct results (Figure 3d): 14% of the totalvegetation cover had the highest agreement and 9% hadthe lowest. Criteria Two and Three had the most similarresults (Figure 3f), with 27% in this highest agreementcategory and 2% in the lowest agreement category.

Combination of Three CriteriaAcross the Edwards Aquifer, the sum of the three

criteria ranged from 3 to 15. The K-means algorithmwas used to find classification thresholds for the fivegroupsvery likely to be GDEs (values ranging from12 to 15), likely to be GDEs (10 to 11), about as likely


(d)

(a)

(c)(b)

Figure 4. GDE mapping in the Edwards Aquifer using the remote sensing method: Three areas were highlighted to show thatGDEs were most likely to be found around springs, along the streams, and in the upland where groundwater is accessibleby phreatophytes. The figure on the left corner showed the soil depth of the Edwards Aquifer region; deep soil refers to thatwith depths over 200 cm. A large number of pixels classified as Very Likely to Contain GDEs (shown in dark green color)were found on shallow soils over carbonate rocks, while the remaining were associated with deep alluvial soils.

as not to be GDEs (8 to 9), unlikely to be GDEs (6 to 7)and very unlikely to be GDEs (3 to 5). The five groups,from Very Likely to Very Unlikely, were 8%, 19%,32%, 26%, and 15% of total natural vegetation coveredareas, respectively (Figure 4). The group Very Likely tobe GDEs was further divided into lowland and uplandGDEs. Potential GDEs within 200 m from a stream orwithin 500 m around a spring or a wetland were classifiedas lowland GDEs. The analysis showed that 11% of thepotential GDEs belonged to lowland category, with 8%occurring in riparian zones and around springs and 3%in other groundwater-fed wetlands. The remaining 89%of the potential GDEs were located in uplands, indicatingthat they were the dominant GDE category in the EdwardsAquifer region.

For the areas very likely to contain GDEs, weexamined vegetation types, the soil depth, and thelandforms to determine whether or not the remotesensing results coincided with our understanding ofimportant GDE characteristics. While water table depthsfor surficial, unconfined aquifers were not available forthis area, the soil data from Web Soil Survey (USDA

2013) indicated that more than 99% of the study areahad a water table deeper than 2 m. Therefore, in orderfor vegetation to access groundwater in this area, it mustpossess a deep root system;. Previous studies have foundthat live oak, ashe juniper, and mesquite are able todevelop such rooting patterns (Wilcox 2002; McElroneet al. 2004).

In the Edwards Aquifer region, live oaks andashe juniper dominated 45% of total natural vegetation,and mesquite dominated 47% (TPWD 2012). However,81% of the potential GDEs were live oak-ashe juniperparks/woods, while only 14% were mesquite dominatedforests and shrublands. We further determined that a largefraction of potential GDEs (75%) were located on shallowsoils, where live oaks and ashe junipers are chiefly found(Mesquite may be located on both shallow and deepsoils.) Why is this significant? Shallow soil areas hadthe average soil depth of 45 cm, which places significantlimits on soil water storage. As a result, vegetation mayneed to access groundwater to complement its wateruse, especially during the dry season. However, deepsoil areas had depths greater than 200 cm, creating with


large storage reservoirs which may be used to buffer theimpacts of droughts and low rainfall periods. Therefore,plants on shallow soils exhibited a higher potential to usegroundwater than those on deep soils.

Landform type was also correlated with areas highlylikely to support GDEs; 66% of potential GDEs werelocated on ridges with shallow soils weathered fromlimestone, and 9% were located on plains covered byshallow soils, also weathered from limestone (Figure 4aand 4c). The remaining GDEs were found on deep soils,mainly near streams (Figure 4b) with alluvial deposits:10% on flood plains 12% on stream terraces and theirerosion remnants, and 3% on paleoterraces.

In the Edwards Aquifer region, the total area poten-tially hosting GDEs was 840 km2. Assuming 90 mm/yearof groundwater is consumed by GDEs, based on litera-ture estimates (Orellana et al. 2012), a total of 2.0 1010gallons of groundwater is used by the potential GDEsevery year in the Edwards aquifer region. For compari-son, this rate of water consumption is nearly 30% of theannual net groundwater use of the City of San Antonio(TWDB 2013b), indicating the significance of GDEs togroundwater management.

ConclusionsWe proposed a methodological framework to identify

potential GDES and applied it to map GDEs in Texas.To address the different management requirements atvarious scales, we developed a two-step approach for thestate/province scale using GIS and the aquifer/basin scaleusing remote sensing-based techniques. We produced statescale GDE index maps for GMAs and HUC-6s in Texasand aquifer/basin scale, 30 30 m resolution maps ofpotential GDEs distributions in the Edwards Aquiferregion. The GDE index maps aimed to identify criticalregions with vulnerable GDEs. These GDE index mapsindicated that areas in central Texas, which host streamswith high baseflow ratios, numerous springs, large areas ofphreatophyte species, and groundwater-fed wetlands, hada high potential to contain a significant amount of GDEs.

The remote sensing-based analysis aimed to identifyGDEs for more specific management and study; inthis case, the Edwards Aquifer region was used asa demonstration of the method. Three criteria weredeveloped, and these captured the physiologic signature ofgroundwater use associated with different plant functionaltypes. Analysis of the criteria showed that each hadidentifiable biases when assessing plant groundwateruse. Criterion One captured the potential groundwateruse of wetlands, but failed to capture it in shrublandsand evergreen forests, due to the impact of their lowvegetation density on NDVI. This disadvantage waseliminated by Criterion Two, but Criterion Two failed tocapture the deciduous forests due to their relatively fastseasonal changes in their leaf areas. These impacts weremitigated by Criterion Three, but Criterion Three failedto capture the groundwater use potential of wetlands.Three criteria were combined together to ameliorate their

disadvantages and yielded a final detailed map of thelocations of potential GDEs. The results indicated thatnot all plants belonging to phreatophyte species or withinwetlands were groundwater dependent. Only 9% of thetotal phreatophytes and 31% of woody and herbaceouswetlands were classified as having the highest potential touse groundwater. Soil depth and landforms were found tobe the critical factors impacting vegetation groundwateruse. Of potential GDEs, 75% were found on ridges andplains with shallow soils. The remaining 25% of potentialGDEs were located on soils deeper than 200 cm, and thesewere mainly associated with streams.

The proposed methods had several limitations. Inthe GDE index method, phreatophytes, woody wetlands,and emergent herbaceous wetlands were considered asthe vegetative indicators . However, this assumption ledto overestimation of the potential GDEs, as compared tothe independent remote sensing-based results. In someinstances, the overestimated vegetative index values mayhave overwhelmed the effect of the hydrological indexon the overall GDE index . In the remote sensing-basedmethod, due to the relatively coarse spatial resolution ofthe satellite images from Landsat ETM+ and MODIS,vegetation pixels with mixed vegetation coverage (e.g.,phreatophytes mixed with bare soil or grasses) may haveNDVI changes that do not accurately reflect the actualvegetation water use pattern. Also, the Criteria Twoand Three results from MODIS data were resampled to30 30 m resolution, which produced some loss of infor-mation. Finally, due to the lack of previous GDE studiesin our study area, future field studies are needed to fullyverify the results produced by the remote sensing method.

In summary, this two-step approach can provide use-ful GDE information for decision makers. The generalunderstanding of the occurrence of GDEs gained fromGDE index maps can help groundwater managers screenareas and integrate the consideration of GDEs into man-agement practices. Detailed GDE distributions obtainedfrom remote sensing provides researchers with a guid-ing tool for the study of GDEs, indicating priority areasfor field-based assessment and monitoring. The resultscan also be used in numerical models intended to sim-ulate the groundwater use of GDEs and their potentialimpacts on water supply, including the tools developedby the Texas Groundwater Availability Modeling pro-gram (TWDB 2013a). In addition, the remote sensing-based method highlights the potential to use satellites toremotely monitor GDE dynamics and health under chang-ing hydrological and climatological conditions.

AcknowledgmentsThe authors would like to acknowledge funding from

the Texas Water Resources Institute in the form of a MillsScholarship and the Schlumberger Foundation in the formof a Faculty for the Future Fellowship, both awarded toS. Gou. We would like to thank the Louis Stokes Alliancefor Minority Participation at Texas A&M Universityand its Undergraduate Research Program for support of


S. Gonzalez. We would also like to acknowledge Dr.Huilin Gao and two anonymous peer reviewers for theirhelpful feedback.

ReferencesBaldwin, C.K., F.H. Wagner, and U. Lall. 2003. Water resources.

In Rocky Mountain/Great Basin Regional Climate-ChangeAssessment. Report for the U.S. Global Change ResearchProgram , ed. F.H. Wagner, 79112. Logan, Utah: UtahState University.

Barron, O.V., I. Emelyanova, T.G. Van Niel, D. Pollock, and G.Hodgson. 2012. Mapping groundwater-dependent ecosys-tems using remote sensing measures of vegetation andmoisture dynamics. Hydrological Process 28: 372385.DOI:10.1002/hyp.9609.

Bertrand, G., N. Goldscheider, J.M. Gobat, and D. Hunkeler.2012. Review: From multi-scale conceptualization to aclassification system for inland groundwater-dependentecosystems. Hydrogeology Journal 20, no. 1: 121.

Brown, J., L. Bach, A. Aldous, A. Wyers, and J. DeGagne.2010. Groundwater-dependent ecosystems in Oregon: Anassessment of their distribution and associated threats.Frontiers in Ecology and the Environment 9, no. 2: 97102.

Brune, G. 1975. Major and historical springs of Texas. TexasWater Development Board Report 189 . http://www.twdb.texas.gov/publications/reports/numbered_reports/doc/R189/Report189.asp (accessed January 2012).

Canty, M.J. 2007. Image Analysis, Classification and ChangeDetection in Remote Sensing: With Algorithms forENVI/IDL. Boca Raton, Florida: Taylor & Francis, CRCPress.

Carlson, T.N., and D.A. Ripley. 1997. On the relation betweenNDVI, fractional vegetation cover, and leaf area index.Remote Sensing of Environment 62, no. 3: 241252.

Dresel, P.E., R. Clark, X. Cheng, M. Reid, J. Fawcett, andD. Cochraine. 2010. Mapping Terrestrial GroundwaterDependent Ecosystems: Method Development and ExampleOutput . Melbourne, Australia: Victoria Department ofPrimary Industries.

Eamus, D., R. Froend, R. Loomes, G. Hose, and B. Murray.2006. A functional methodology for determining thegroundwater regime needed to maintain the health ofgroundwater-dependent vegetation. Australian Journal ofBotany 54, no. 2: 97114. DOI:10.1071/bt05031

Elmore, A.J., J.F. Mustard, and S.J. Manning. 2003.Regional patterns of plant community response tochanges in water: Owens Valley, California. EcologicalApplications 13, no. 2: 443460. DOI:10.1890/1051-0761(2003)013[0443:RPOPCR]2.0.CO;2

Goedhart, C.M., and D.E. Pataki. 2010. Ecosystem effects ofgroundwater depth in Owens Valley, California. Ecohydrol-ogy 4, no. 3: 458468. DOI:10.1002/eco.154

Howard, J., and M. Merrifield. 2010. Mapping groundwaterdependent ecosystems in California. PloS ONE 5, no. 6:e11249. DOI:10.1371/journal.pone.0011249

Jackson, R.B., L.A. Moore, W.A. Hoffmann, W.T. Pockman,and C.R. Linder. 1999. Ecosystem rooting depth determinedwith caves and DNA. Proceedings of the National Academyof Sciences of the United States of America 96, no. 20:1138711392.

Loheide, S.P., R.S. Deitchman, D.J. Cooper, E.C. Wolf, C.T.Hammersmark, and J.D. Lundquist. 2009. A framework forunderstanding the hydroecology of impacted wet meadowsin the Sierra Nevada and Cascade Ranges, California, USA.Hydrogeology Journal 17, no. 1: 229246.

Loheide, S.P., and S.M. Gorelick. 2007. Riparian hydroe-cology: A coupled model of the observed interactions

between groundwater flow and meadow vegetation pat-terning. Water Resources Research 43, W07414. DOI:10.1029/2006WR005233

Loheide, S.P., and S.M. Gorelick. 2005. A local-scale, high-resolution evapotranspiration mapping algorithm (ETMA)with hydroecological applications at riparian meadowrestoration sites. Remote Sensing of Environment 98, no.2: 182200.

Lowry, C.S., S.P. Loheide II., C.E. Moore, and J.D. Lundquist.2011. Groundwater controls on vegetation compositionand patterning in mountain meadows. Water ResourcesResearch 47, no. 10: W00J11.

MacQueen, J. 1967. Some methods for classification and analy-sis of multivariate observations. Paper read at Proceedingsof the Fifth Berkeley Symposium on Mathematical Statis-tics and Probability.

Martinet, M.C., E.R. Vivoni, J.R. Cleverly, J.R. Thibault,J.F. Schuetz, and C.N. Dahm. 2009. On groundwaterfluctuations, evapotranspiration, and understory removal inriparian corridors. Water Resources Research 45, no. 5:W05425. DOI:10.1029/2008wr007152

McElrone, A.J., W.T. Pockman, J. Martinez-Vilalta, and R.B.Jackson. 2004. Variation in xylem structure and functionin stems and roots of trees to 20 m depth. New Phytologist163: 507517.

McElrone, A.J., W.T. Pockman, and R.B. Jackson. 2003. Thecontribution of deep roots to the water cycle: Seasonalpatterns of water uptake and hydraulic lift from below 7m on the Edwards Plateau, Texas. Eos Trans. AGU, FallMeet. Suppl . 84, no. 46: Abstract H32C-0581.

Miller, G.R., X. Chen, D. Baldocchi, and Y. Rubin. 2010.Groundwater uptake by woody vegetation in a Mediter-ranean oak savanna. Water Resources Research 46:W10503. DOI:10.1029/2009WR008902

Munch, Z., and J. Conrad. 2007. Remote sensing and GIS baseddetermination of groundwater dependent ecosystems in theWestern Cape, South Africa. Hydrogeology Journal 15, no.1: 1928.

Naumburg, E., R. Mata-Gonzalez, R.G. Hunter, T. McLendon,and D.W. Martin. 2005. Phreatophytic vegetation andgroundwater fluctuations: A review of current researchand application of ecosystem response modeling withan emphasis on Great Basin vegetation. EnvironmentalManagement 35, no. 6: 726740.

Orellana, F., P. Verma, S.P. Loheide, and E. Daly. 2012.Monitoring and modeling water-vegetation interactions ingroundwater-dependent ecosystems. Reviews of Geophysics50: RG3003. DOI: 10.1029/2011RG000383

Purevdorj, T., R. Tateishi, T. Ishiyama, and Y. Honda.1998. Relationships between percent vegetation cover andvegetation indices. International Journal of Remote Sensing19, no. 18: 35193535.

Schaeffer, S., D.G. Williams, and D.C. Goodrich. 2000. Tran-spiration of cottonwood/willow forest estimated from sapflux. Agricultural and Forest Meteorology 105: 257270.

Scott, R.L., W.L. Cable, T.E. Huxman, P.L. Nagler, M.Hernandez, and D.C. Goodrich. 2008. Multiyear riparianevapotranspiration and groundwater use for a semiaridwatershed. Journal of Arid Environments 72, no. 7:12321246. DOI:10.1016/j.jaridenv.2008.01.001

Slade, R.M., J.T. Bentley, and D. Michaud. 2000. Results ofstreamflow gain-loss studies in Texas, with emphasis ongains from and losses to major and minor aquifers. U.S.Geological Survey Open-File Report 02-068 . Reston,Virginia: USGS.

Smith, J.B., R. Richels, and B. Miller. 2003. The potentialconsequences of climate variability and change for theWestern United States. In Climate Change Impacts onthe United States . Washington, DC: U.S. Climate ChangeScience Program.


Steinwand, A.L., R.F. Harrington, and D. Or. 2006. Waterbalance for Great Basin phreatophytes derived from eddycovariance, soil water, and water table measurements.Journal of Hydrology 329, no. 34: 595605.

TPWD. 2012. The Vegetation Types of Texas . Austin, Texas:Texas Parks and Wildlife Department. http://www.tpwd.state.tx.us/publications/pwdpubs/pwd_bn_w7000_0120/(accessed January 2012).

TWDB. 2013a. Groundwater Availability Models. Austin,Texas: Texas Water Development Boards. http://www.twdb.state.tx.us/groundwater/models/gam/index.asp (accessedJune 2013).

TWDB. 2013b. Historical Water Use Estimates. Austin, Texas:Texas Water Development Board. http://www.twdb.state.tx.us/waterplanning/waterusesurvey/estimates/ (accessedJune 2013).

TWDB. 2012. Groundwater Management Areas in Texas (GMA).http://www.twdb.state.tx.us/mapping/gisdata.asp (accessedJanuary 2012).

Tweed, S.O., M. Leblanc, J.A. Webb, and M.W. Lubczynski.2007. Remote sensing and GIS for mapping groundwaterrecharge and discharge areas in salinity prone catchments,southeastern Australia. Hydrogeology Journal 15, no. 1:7596.

USDA. 2013. Soil Survey Staff, Natural Resources ConservationService, United States Department of Agriculture. WebSoil Survey . http://websoilsurvey.nrcs.usda.gov/ (accessedOctober 18, 2013).

USFWS. 2012. National Wetlands Inventory website. Washing-ton, DC: U.S. Department of the Interior, Fish and WildlifeService. http://www.fws.gov/wetlands/ (accessed January2012).

USGS. 2012a. National Hydrography Dataset (NHD). Reston,Virginia: U.S. Department of the Interior, U.S. GeologicalSurvey. http://nhd.usgs.gov/ (accessed January 2012).

USGS. 2012b. National Land Cover Database (NLCD 2006).Reston, Virginia: U.S. Department of the Interior, U.S.Geological Survey. http://nationalmap.gov/landcover.html(accessed January 2012).

Van Auken, O., A. Ford, and J. Allen. 1981. An ecologicalcomparison of upland deciduous and evergreen forests ofcentral Texas. American Journal of Botany : 12491256.

Wahl, T.L., and K.L. Wahl. 2007. BFI: A Computer Program forDetermining an Index to Base Flow . Lakewood, Colorado:U.S. Department of the Interior, Bureau of Reclama-tion. http://www.usbr.gov/pmts/hydraulics_lab/twahl/bfi/(accessed January 2012).

Wahl, K.L., and T.L. Wahl. 1995. Determining the flow ofcomal springs at New Braunfels, Texas. In Texas Water95 , American Society of Civil Engineers, August 1617,1995, San Antonio, Texas, 7786.

Wang, Q., J. Tenhunen, N.Q. Dinh, M. Reichstein, T. Vesala, andP. Keronen. 2004. Similarities in ground-and satellite-basedNDVI time series and their relationship to physiologicalactivity of a Scots pine forest in Finland. Remote Sensingof Environment 93, no. 1: 225237.

Wilcox, B.P. 2002. Shrub control and streamflow on rangelands:A process based viewpoint. Journal of Range Management55: 318326.

Wilcox, B.P., M.K. Owens, W.A. Dugas, D.N. Ueckert, and C.R.Hart. 2006. Shrubs, streamflow, and the paradox of scale.Hydrological Processes 20, no. 15: 32453259.

Wolock, D.M. 2003. Base-flow index grid for the contermi-nous United States: U.S. Geological Survey Open-FileReport 03-263 . Lawrence, Kansas: U.S. Geological Survey.http://water.usgs.gov/lookup/getspatial?bfi48grd (accessedJanuary 2012).


gou et al-2015-groundwater (1)

Documents

identifypotential gdes

individual gdes

detailed maps of gdes

national groundwater

texas amuniversity

central texas

stateprovince scale

aquifer scale maps