Using Remote Sensing and GIS to Compute Evapotranspiration in the Rio Grande Bosque

Julie Coonrod and Dianne McDonnell

Department of Civil Engineering

University of New Mexico

Albuquerque, NM


          Increasing demand for water and the scarce supply of water is a growing concern in the semiarid regions of the United States and the world. Semiarid regions lose a tremendous amount of water to evapotranspiration. The middle Rio Grande bosque has native cottonwood trees and invasive salt cedar trees.Four towers extend above the canopy to measure evapotranspiration from the two tree species.Landsat data and AVHRR data are used in ArcView Image Analysis and ArcView Spatial Analyst to derive a method to compute evapotranspiration purely from satellite imagery.






          Ground Measurement Technique

          Remote Sensing Technique








Evapotranspiration is a major source of water depletion in arid and semiarid environments.  Approximately one-third of the worldwide land surface occurs in regions with arid or semiarid climates, and about 20% of the world’s population presently lives in these regions.  The scarcity of water and growing human demand for water often result in conflicts between human needs and those of biota within these river corridors (Warren, Sud & Tozanov, 1996).

The Rio Grande, located in the southwestern United States, is an example of a river that has many competing demands in a semiarid/arid climate.  More specifically, the Middle Rio Grande runs through central New Mexico, U.S.A. and is typically defined by the reach of river between Cochiti Dam and Elephant Butte Reservoir.  The contributing watershed to this reach of river is shown in Figure 1.  Water demands include those of the state’s largest city (Albuquerque), irrigation districts, and several endangered species. Water budgeting is critical due to these demands and the legal compacts between adjacent states requiring the delivery of mandated quantities of water. The inputs and outputs of water from the upper and lower end of the river corridor are reasonably well quantified by permanent gauges with long-term records. Additional sources and depletions of water along the Middle Rio Grande are more poorly quantified.  Evapotranspiration is believed to account for about one quarter of the total water depletion along the semi-arid Middle Rio Grande Valley.  An accurate estimate of evapotranspiration (ET) is an important component of the water budget.  Understanding the water budget will assist in developing effective riparian restoration strategies for this area.

            An interdisciplinary and interagency effort to assess the annual water budget for the Middle Rio Grande resulted in a recent report summarizing estimated averages for 1972-1997 (Middle Rio Grande Water Assembly, 1999). Five major sources of surface water depletion are identified along this approximately 320 km reach of river. Major depletions include 1) open-water evaporation, 2) riparian zone ET, 3) irrigated agriculture, 4) groundwater recharge, and 5) urban consumption. Annual estimates are provided along with data on interannual variation in supplies and depletions where adequate data are available. The consensus opinion of the scientific team who participated in the development of this water budget is that open-water evaporation, riparian zone ET, and irrigated agriculture represent the three largest depletions along the corridor. It also is understood that major uncertainties presently exist in the quantification of all major depletions and that interannual variability will be large.  For example, annual estimates of the extent of riparian zone ET along the Middle Rio Grande corridor range from 150 x 106 m3 to 375 x 106 m3. Similar levels of uncertainty and/or interannual variability exist in the estimates of other major water depletions. 


The Middle Rio Grande and the adjacent riparian forest (or bosque) have historically been the center of development within the region.  This development and subsequent river management have led to increased stress on the associated aquatic and riparian ecosystem.  The resulting changes in riparian habitat are believed to have increased ET depletion rates.

The Middle Rio Grande riparian forests are dominated by Populus deltoides commonly known as cottonwood.  In fact this riparian corridor is home to the most extensive cottonwood forest remaining in the southwestern United States (Molles, 1999).  Cottonwood forests are the preferred habitat for many animals from insects to birds (Rowland, 1998).  Before river management, runoff from spring snowmelt in the mountains of Colorado and New Mexico created river flows with enough magnitude to flood the river’s banks.  Cottonwood seeds depend on open, scoured sandy bars and this periodic flooding in order to germinate and become established (Thibault et al., 2001). 

The changes in the Middle Rio Grande have led to the influx of  various exotic species including Russian olive (Elaeagnus angustifoliea), salt cedar (Tamarix ramosissima), and Siberian elm (Ulmus pumila) (Crawford et al., 1993).  Although all three species increase competition for resources, especially groundwater, the most infamous of the three is the salt cedar.  Salt cedar can tolerate salt concentrations up to 15,000 ppm, and has the ability to concentrate salts in its leaves.  As the leaf litter accumulates beneath the trees, salinity also increases and thus prevents less tolerant species from colonizing the area (Barrow, 1996).   Salt cedar can consume large volumes of water, perhaps more than the native species (Carpenter, 1998).  According to Tickner et al., 2001, the development of dense salt cedar thickets along riparian corridors should be avoided to promote water conservation. Figure 2 illustrates a typical cottownwood stand and a typical salt cedar stand.

Figure 3. Land Use Changed along the Rio Grande near Albuqerque

Figure 3 illustrates some of the changes to Middle Rio Grande, which have resulted in the loss of cottonwood habitats.  In the 1935 view, the river is more braided and weaves in and out of the sandbars.  Today, flows are regulated and the river is constrained within levees that greatly limit the range of overbank flooding.  The sandbars cannot be seen in the 1999 image.  The 1935 coverage was developed by the Albuquerque office of the Corps of Engineers from aerial photography.  The 1999 coverage was developed from a Landsat 7 image using ArcView Image Analyst.



            Determination of the difference in evapotranspiration rates between native and non-native trees, as well as areas that are flooded and non-flooded will provide data for decision makers in regards to river restoration and river management.

            Eight study sites have been established (as shown in Figure 1.)  Each site has five wells to study the relationship of groundwater levels to river discharge and to evapotranspiration.  Four of the eight sites are equipped with towers that extend above the canopy to measure ET.  The northern most sites (Rio Grande Nature Center and Shirk) have cottonwood forests with no flooding.  The next two sites (Los Lunas and Belen) are also cottonwood but these sites typically experience flooding.  The four most southern sites are dominated by salt cedar with the Bernardo and Sevilleta site typically remaining dry and the two Bosque del Apache sites experiencing flooding.

            The four sites equipped with towers allow for a ground measurement of ET.  However, it is necessary to estimate the ET for the entire corridor.  Remote sensing techniques are considered for the entire corridor estimation of ET.

Ground Measurement Technique

Eddy covariance instruments (Campbell Scientific, Logan, Utah, U.S.A.) are affixed to towers, positioned approximately 1.5 to 2 meters above the cottonwood or saltcedar canopy, depending on the site. Four tower sites, two in cottonwoods stands (~ 25 m) and two in saltcedar stands (~ 10 m), are instrumented to measure ET.  Photos of two towers are shown in Figure 4..  Latent energy and sensible heat are calculated at 30-minute intervals. A three-dimensional sonic anemometer in which ultra-sonic signals are pulsed between three pairs of transducers is used to determine vertical wind velocity. Supporting instruments are deployed to measure temperature, humidity, and atmospheric pressure, which are used in the eddy covariance equations. In addition, it is also necessary to monitor wind direction to identify periods when wind strikes the back of the sensors, compromising flux data. Orienting the instruments into the direction of southerly prevailing winds minimizes the problem.

Average annual growing season ET was measured with the eddy covariance method at the two tower sites (Sevilleta and Bosque del Apache - South) in predominantly saltcedar vegetation in 1999 and 2000. Eddy covariance measurements at the predominantly cottonwood stands (South Valley and Belen) were made in the growing season of 2000. Daily measurements of ET (mm d-1) at all four sites are continuing during the 2001 growing season. An example of the daily ET measurements made at the South Valley (Shirk) site with a cottonwood-dominated overstory with an extensive understory of exotic species (Russian olive and saltcedar) are shown in Figure 5. A few days with missing data are seen due to instrument failures commonly associated with precipitation events. Daily measurements of ET range from less than 1 mm d-1 early and late in the growing season to a maximum of 9 mm d-1 during the peak of the summer. Variations in rates of plant transpiration throughout the growing season reflect the onset of the growing season, development of full canopy, daily variability in weather conditions, and plant senescence. Integrating daily measurements of ET with extrapolation between adjacent measurements for dates with missing data yield an annual estimate of growing season ET for the plot. The average growing season estimate of ET for the South Valley site in 2000 is 93 cm.

Estimates of annual ET from the four sites with tower-based eddy covariance measurements are shown in Table 1.

The lowest rates of ET (57 cm yr-1 in 1999 and 60 cm yr-1 in 2000) were at the Sevilleta site with a moderate-density saltcedar stand that rarely floods (Cleverly et al., in review). The other saltcedar site (Bosque del Apache – South) floods occasionally during spring snowmelt and summer monsoonal thunderstorms. It is a dense monotypic stand of saltcedar. The onset of plant transpiration at this site actually began earlier in 2000, but drought conditions reduced summer ET rates in 2000 and overall ET was lower. The South Albuquerque (Shirk) site had the highest amount of ET during 2000.  This cottonwood site rarely floods and contains many understory Russian olive and saltcedar trees.  The cottonwood site in Belen had significantly less ET than Albuquerque.  This site contains few understory trees.  Review of the year 2000 ET rates at the four sites shows a large variability due to type of vegetation and flooding regime.  To integrate ET rates for the entire corridor requires establishing relationships between ET, vegetation, and flooding regime.  The use of satellite imagery is investigated to establish such relationships.

Remote Sensing Technique

High resolution aerial photography has been used for years to determine land use and land cover.  However the cost of the aerial photography has prevented frequent flights that can document changes over time as well as changes throughout the year.  Satellite imagery is increasingly more available at various temporal and spatial scales.  Furthermore, satellite sensors measure reflected radiation within the electromagnetic spectrum at various bandwidths producing a number of indices that can be used for analysis. 

Carlson, et. al. (1995) proposed determining ET exclusively from satellite imagery.  Carlson’s proposed equation uses surface temperature.  The most common satellite used to measure surface temperature is the National Oceanic and Atmospheric Administration (NOAA), Polar Orbiting Environmental Satellite, which carries the Advanced Very High Resolution Radiometer (AVHRR).  The AVHRR sensors scan the United States twice daily and record images with a one-kilometer spatial resolution.  The U.S. Geological Survey’s EROS Data Center converts the data into biweekly composites.  AVHRR has two bands that fall into the thermal infrared portion of the electromagnetic spectrum.  Determination of surface temperatures using AVHRR uses both thermal infrared bands and has generally been over large homogenous areas including sea surfaces and agricultural fields.  As shown in Figure 6,

Figure 6. AVHRR grid cell near Sevilleta Tower Site

the one-kilometer spatial resolution provided by AVHRR images simply cannot accurately measure the surface temperature of the narrow riparian corridor along the Rio Grande.  Largely because of the limited spatial resolution, the results of radiant temperatures and estimated kinetic surface temperatures do not correlate well with the measured tower surface temperatures (McDonnell, 2000).

Landsat 7, maintained by the Landsat 7 Project Science Office at the NASA Goddard Space Flight has considerably better resolution than AVHRR; however, there is only one band in the thermal infrared range.  Landsat 7 carries the Enhanced Thematic Mapper plus.  The sensors measure reflected radiation within the electromagnetic spectrum at eight different bandwidth ranges.  The results are used to characterize or infer properties of the landscape.   Typically Bands 4, 3, and 2 are combined to make false-color composite images where band 4 is shown in red, band 3 is shown in green, and band 2 is shown in blue. Each of these bands has a 30-meter grid cell size.  The images in Figure 7

Figure 7. Landsat 7 scenes near study sites

are examples of false-color composite images for the six most northern sites.  The Bosque del Apache sites are not available in the same Landsat 7 scene.  Scenes containing the Bosque del Apache sites have not yet been acquired for this project.

To date, the Landsat 7 scenes are being used for vegetation classification of the corridor and for determining other vegetation indices that might be appropriate for scaling ET at the plot scale to the corridor scale.  Because Landsat 7 passes a location every 16 days, the imagery can also be used to document the change in vegetation over the growing season.  The Landsat 7 image taken May 9, 2000 is used to determine vegetation classification and indices that can be compared to data collected during a June 2000 field campaign.  During this field campaign, James Cleverly and Jim Thibault took leaf area index (LAI) measurements using a plant canopy analyzer (Li-Cor Environmental Division).  The LAI measurements compare well with ET measurement as shown in Figure 8.

Figure 8. Field based ET and LAI

Leaf area index is a key biophysical variable influencing land surface photosynthesis and transpiration and can be related to remotely sensed data.  A recent study shows a strong general relationship between spectral vegetation indices derived from the Landsat 5 Thematic Mapper and LAI (Turner et al., 1999).  Using similar techniques described in Turner et al., 1999 and Schultz and Engman, 2000, this paper attempts to determine LAI using the May 9, 2000 Landsat 7 image for the six project sites which fall within the boundaries of the image.  A comparison of the resulting LAI to field based LAI measurement and tower based ET estimates are made.  Analysis and image correction are done in ArcInfo and ArcView. 

            Greg Shore from the University of New Mexico Sevilleta program provided a rectified image using established control points from Tiger maps.  Using instructions provided with the Landsat 7 image (, corrections are made and values are converted to unitless reflectance values.  Reflectance can be correlated to a variety of values.  For example, band 6 represents temperature reflectance and can be corrected to degrees Kelvin when known canopy temperatures are available.  Once the corrections to reflectance are made, the images are ready for analysis.  Each of the 8 bands is extracted from the main image.  Values for reflectance at each band are determined using Arc View Spatial Analyst. 


            Although ArcView’s Image Analyst can be used for vegetation classification, and specifically for identifying the riparian forest, it is difficult to differentiate between different tree species.  To assist in this differentiation, individual bands can be analyzed. The results of plotting the midpoints of the Landsat 7 image bandwidths (as shown in Figure 9) show distinguishable patterns between the cottonwood sites and salt cedar sites.

Figure 9. Multispectral Signatures

Band 4 (near-infrared) in relationship with band 5 (mid-infrared) provides the clearest differentiation.  Since band 4 reflectance values are the result of plant cell wall structure, it is likely that differentiation is largely due to very different physiological leaf structures between salt cedar and cottonwood.  

Temperature grids are shown in Figure 10.

Figure 10. Temperature grids

Temperature is an important component in ET and the resulting temperature gradients show interesting patterns worthy of comment (Carlson et al., 1995).  The Sevilleta site is significantly warmer than any of the other sites while South Valley and Belen are much cooler than the other sites.  This does not follow any leaf area or ET pattern.  However, based on Figure 10, it is easy to see that the Sevilleta site is far from the river while the two cooler sites are closest to the river.  The images are taken at 10:30 AM local time.  Climatic conditions around a river valley, especially in the early morning time can be quite cool.  This is primarily true in high desert climates where heat escapes during the night and cool air sinks to the lowest elevation, typically the river valley.  The opposite occurs during the day and the river valley can actually heat up considerably.  (Barry and Chorley, 1998)   Temperature values may not reflect the maximum daily ET.  Leaf area is gradual indicator of transpiration and does not respond as quickly as temperature to diurnal changes, therefore, it may be a more appropriate indicator of ET.

            There have been strong correlations between a red and near-infrared transmittance ratio and LAI.  Chlorophyll absorbs high red energy while plant foliage has relatively low transmittance (reflectance) of red energy.  Plant cell walls, in particular lignin, cause scattering of near-infrared energy resulting in high near-infrared transmittance and reflectance (Turner et al., 1999).   The relationship between near-infrared and red reflectance by plants has led to the development of a variety of remote sensing based vegetation based indices.  The two indices determined in this study are the Normalized Difference Vegetation Index (NDVI) and the Soil Adjusted Vegetation Index (SAVI).

NDVI is a well-known vegetation index defined as:

NDVI = (NIR – R)/ NIR + R where NIR = near-infrared reflectance (Band 4) and R = red reflectance (Band 3)

The range is theoretically between –1 and 1 but red is rarely larger than infrared (Schultz and Engman, 2000).Figure 11 displays the NDVI grids in the areas of the six sites. Note that lower values for NDVI for the salt cedar than the cottonwood sites relating to their overall lower values of near-infrared reflectance in salt cedar.

Figure 11. NDVI grids

SAVI uses the same principals as NDVI but adjusts for the effect of soil conditions.   SAVI is defined as:

SAVI = (1+ L) (NIR – R) / (NIR + R + L)where L = 0.5

The SAVI grids are shown in Figure 12.  Similar to the NDVI grids, lower values are seen in the salt cedar sites.

Figure 12. SAVI grids

Since there is only one image, six project sites, and three tower sites, there can be no real statistical determination of LAI using NDVI or SAVI.There have been other studies that have determined a non-linear relationship between LAI and SAVI, that relationship will be applied here.  According to Schultz and Engman, 2000, LAI is related to SAVI. 

SAVI = c1 – c2 e –c3LAI where c2

                                                c3  =  0.91                   

Therefore:            LAI   =  - ln (SAVI + .371)/.48        

The LAI grid developed with this relationship is shown in Figure 13.

Figure 13. Derived LAI grids


The vegetation indices NDVI and SAVI show lower values for the salt cedar than the cottonwood sites relating to their overall lower values of near-infrared reflectance in salt cedar (figures 11 and 12).  The grids showing NDVI, SAVI, and LAI do not appear to show any correlation with either field measured LAI or tower estimates of ET.  In fact, the site with the lowest LAI has the highest ET.  The results mean little without considering site characteristics.

Table 2 summarizes the data with the site characteristics.Figure 14 displays the discrepancies between the remotely sensed LAI and the field based LAI.

          According to Table 2, the three sites with the largest understory have the highest DLAI (the difference between remote sensing estimates of LAI and field measured LAI).  This makes sense because the satellite image cannot see the understory and is likely to underestimate LAI at these sites.  The D LAI at the two flooded cottonwood sites with native understory is about the same.   However, the non-flooded cottonwood site with salt cedar understory shows nearly two times the D LAI.  The Rio Grande Nature Conservancy site, with a limited understory, has similar values for LAI determined using remote sensing and field measured LAI.   The Bernardo site has cottonwoods associated with the site, this may cause a slight undervaluing of LAI.  The Sevilleta site, the only homogenous site, is also the only site where the satellite derived LAI overestimates field measured LAI.  This is likely due to the grasses in the understory which the LAI meter did not measure.  The canopy is sparse here and the remote sensing measurements probably include the grass leaf area in its reflectance values.


Salt cedar and cottonwoods both have unique spectral signatures and the two species can be differentiated using Landsat 7 satellite imagery.  The study also shows the possible affect of river climate on temperature gradients relative to river proximity.   It is impractical to speculate the cause of the variations in LAI measurements with only one satellite image.  As more data and imagery are collected, the feasibility of using Landsat 7 imagery to delineate understory type based on the underestimates of LAI will be investigated.  Furthermore, the combination of understory and topstory knowledge may lead to a better estimation of ET.  In other words, can we correct for understory underestimates of LAI and then determine ET depletion rates along the corridor?

In order to answer the above question, correlations between LAI measurements (field verses remote sensing) should first be done in homogenous systems without an understory.   If correlations can be made, then comparisons with canopies with varying understory types can be made.  If Landsat 7 images underestimate LAI in similar understory types by approximately the same amount, then it may be possible to distinguish understory types using remote sensing.  If we know the canopy understory, it may be possible to derive a correction for LAI for each understory type.   This finally brings us back to the original purpose of the study, determining whether or not Landsat 7 imagery can be used to estimate LAI and whether the LAI estimates can be related to tower based estimates of ET.  We are conducting an intensive field campaign to collect more LAI data during the summer, 2001.  We will continue to investigate the relationships between field measured LAI, remotely sensed LAI, and field measured ET with acquisition of additional Landsat images during the summer, 2001.


            The authors thank the many members of the Hydrogeoecology Group at the University of New Mexico, past and present, for their help with field and laboratory work. The U.S. National Aeronautics and Space Administration (NASA) funded this research through contract NAG5-6999. In addition, the U.S. National Science Foundation through award DEB 0080529 supported part of the research. The authors specifically thank Dr. Cliff Dahm, Dr. James Clevery, Jim Thibault, and Greg Shore for their assistance in many aspects of the design and implementation of this study.


Barrow, B.W., 1996, Tamarisk Control and Common Sense; California Exotic pest Plant Council, 1996 Symposium. 

Barry, R.G., Chorely, J.R., 1998, Atmosphere, Weather & Climate 7th edition. Routledge, New York.

Carlson, T.N., Capehart,W.J., Gillies, R.R., 1995, A new look at the simplified method for remote sensing of daily evapotranspiration:  Remote Sensing Environment, 54: 161-167.

Carpenter, A.T., 1998.  Element Stewardship Abstract for Tamarix ramossissima Ledebour, Tamarix pentandra Pallas, Tamarix chinensis Loureiro, Tamarix parviflora De Cadolle, Saltcedar, Salt cedar, Tamarisk.  The Nature Conservancy, Arlington, VA.

Cleverly, J.R., Dahm, Thibault J.R., C.N., Gilroy, D.J., Coonrod, J.A.,  (not yet published), Seasonal estimates of actual evapotranspiration from Tamarix ramosissima stands using 3-dimensional eddy covariance.

Crawford, C.S., Cully, A.C., Leutheuser, R., Sifuentes, M.S., White, L.H., Wilber, J.P., October 1993.  Middle Rio Grande Ecosystem:  Bosque Biological Management Plan.  The Middle Rio Grande Biological Interagency Team, Albuquerque, NM.

Landsat 7 handbook for May 2000 (this handbook is provided digitally with the image)

(Landsat 7 website -

Middle Rio Grande Water Assembly. (1999) Middle Rio Grande Water Budget (Where Water Comes From, and Goes, and How Much): Averages for 1972-1997. Middle Rio Grande Council of Governments of New Mexico, Albuquerque, 10 pp.

McDonnell, D., 2000, Using Remote Sensing To Estimate Surface Temperature At The Top Of A Forested Canopy Along The Middle Rio Grande, Course Paper.

Molles, M.C., 1999,  Ecology Concepts and Applications. WCB/McGraw-Hill Companies, New York.

Rowland, D.L., 1998, Variation in Physiological, Morphological, and Demographic Traits Among Cottonwood Populations in New Mexico:  Evidence for an Environmental and Genetic Basis of Variation.  Ph. D. Dissertation University of New Mexico Department of Biology.

Schultz, G. A., Engman, E.T., 2000, Remote Sensing in Hydrology and Water Management.  Springer, New York.

Thibault J., Dahm, C., Cleverly, J., 2001, Quantifying Riparian Evapotranspiration:  Dialogue, Water Dialogue, Gallup New Mexico.

Tickner, D.P., Angold, P.G., Gurnell, A.M., Mountford, J.O., 2001, Riparian plant invasions:  hydrogeomorphological control and ecological impacts.  Progress in Physical Geography. 25, 1 pp22-52.

Turner, D.P., Cohen, W.B., Kennedy, R.E., Fassnacht, K.S., Briggs, J.M., 1999, Relationships between Leaf Area Index and Landsat TM Spectral Vegetation Indices across Three Temperate Zone Sites:  Remote Sensing Environment, 70: 52-68.

Warren A., Sud Y.C. & Rozanov B. (1996) The future of deserts. Journal of Arid Environments, 32, 75-89.


Julie Coonrod

Assistant Professor

Department of Civil Engineering

University of New Mexico

Albuquerque, NM  87131

(505) 277-3233

Dianne McDonnell

Ph.D. Candidate

Department of Biology

University of New Mexico

Albuquerque, NM  87131

(505) 277-4556