ESTIMATION OF EVAPORTRANSPIRATION OVER THE SAN PEDRO RIPARIAN
AREA WITH REMOTE AND IN SITU MEASUREMENTS
AREA WITH REMOTE AND IN SITU MEASUREMENTS
J. Qi(1), M. S. Moran(1), D. C. Goodrich(2), R. Marsett(1), R. Scott(3), A. Chehbouni(4), S. Schaeffer(3), J.
Schieldge(5), D. Williams(3), T. Keefer (2); D. Cooper(6), L. Hipps(7), W. Eichinger(8), and W. Ni(1)
(1) USDA-ARS, Water Conservation Laboratory, Phoenix, Arizona, USA;(2) USDA-ARS, Southwest Watershed Research Center, Tucson, Arizona, USA; (3) University of Arizona, Tucson, AZ, USA; (4) ORSTOM/IMADES, Hermosillo, Sonora, Mexico; (5) Jet Propulsion Lab., Pasadena, CA, USA;(6) Los Alamos Nat. Lab., Los Alamos, NM, USA; (7) Utah State Univ., Logan, UT, USA; (8) Univ. of Iowa, Iowa City, IA USA
* Corresponding author: USDA-ARS, Water Conservation Laboratory, 2000 E. Allen Road, Tucson, Arizona, USA email: qi@tucson.ars.ag.gov
1. INTRODUCTION
Water in the southwest US has been and remains an issue of paramount importance. Its quality and usages have been the source of political and legal activity for over 100 years(Carey Act, 1894). With increases in regional population and on going use of irrigation, the issue of water continues to become more important. With ever increasing water demands and decreasing ground water tables, it becomes apparent that knowledge of water recharge, its spatial distribution, and its partitioning into ground infiltration, surface runoff, and evaporative losses to the atmosphere, is critical for properly addressing water issues in semi-arid regions. Apart from these societal issues, water in arid and semi-arid regions is also an important component in the hydrological cycle. Its spatial and temporal variations are key inputs to hydrological models and determine rainfall patterns. Therefore, its spatial and temporal distribution, especially the evaporative losses, not only provides an input to the water budget equation, but also improves our understanding of the hydrological process in the semi-arid environment.
Aimed at addressing these issues, the SALSA (Semi-Arid Land Surface-Atmosphere) Program was initiated at the USDA-Agricultural Research Service, Southwest Watershed Research Center, to provide an infrastructure for long term monitoring of some of the key variables in understanding the hydrological processes and its environmental impacts in the riparian area along the San Pedro River (Goodrich et al., this issue). The objective of this study, as part of the integrated SALSA program objectives, is to develop practical algorithms to estimate seasonal/annual/inter-annual, and decadal variations in the evaporative water losses, using remotely sensed imagery and in situ measurements. This paper presents some preliminary results of evaporative water losses over a spatially heterogeneous portion of the San Pedro River basin with a focus on a portion of the riparian corridor from Lewis Spring to Fairbanks (See Figure 1 in Goodrich et al., in this issue).
2. FIELD EXPERIMENT
The Lewis Spring site(see Figures 1 and 2 in Goodrich et al., this issue) is located north of State Highway 90 where it crosses the San Pedro River. It is at this location that intensive ground in situ measurements were made. This site is composed of riparian cottonwood along the corridor, sacaton grasses on either side of the bank, and mesquites further away from the banks(Figure 1). A total of six intensive field campaigns was conducted at this site, four with aircraft remote sensing measurements in April, July, August, and October 1997, dedicated to acquiring ground truth data for algorithm development and hydrological model validation.
During an intensive field campaign in August, remote sensing images from Landsat satellite and multi-altitude aircraft sensors were acquired at spatial resolutions ranging from 0.63m to 30m and with different spectral bands ranging from optical to thermal infrared to microwave. These images were georegistered and calibrated to obtain surface reflectances and temperatures(Moran et al.,this issue). In particular, low altitude aircraft images of surface temperatures acquired by Agrometrics were used in this analysis, together with in situ ground measurements, to map the spatial variations of water losses through the evapotranspiration process and temporal changes through the season over the four campaigns.
Ground in situ measurements of surface temperatures, evaporative water losses, sensible heat fluxes, and the available incoming solar radiation, along with other meteorological variables, were made at this site (Figure 1) to document the variations of these physical parameters during the experiment period. The evaporative water losses during the intensive field campaign in August ( 7- 15) 1997, were measured with the following devices:
1) Bowen Ratio: This is a long-term effort to record evapotranspiration (ET) to document both diurnal and seasonal changes of water loss through evapotranspiration. Two towers were placed on the site, one being located within the mesquite and another within the sacaton sites (Scott et al., this issue). Due to spatial heterogeneity, some mesquites were found to exist at the sacaton site within the fetch of the Bowen Ratio devices and vice verse for the mesquite site.
2) Scintillometer: This device was first placed over the sacaton grass-dominated site (see Figure 1) from August 7 to 10 and moved to approximately 1.5 km west of the site to measure the sensible heat flux across the mixture of sacaton grasses and mesquite shrubs (Chehbouni et al., this issue). This device measured the sensible heat fluxes, from which evaporative water fluxes can be computed, using the available energy measurements by the Bowen ratio tower.
3) 3-D Sonic: This 3-dimensional sonic device simultaneously measures both sensible and latent heat fluxes of a representative area (Schieldge et al., this issue). The device was deployed at the site to measure the fluxes of sacaton grasses.
4) Sapflow: The rate of water flowing across a sapwood section was measured(Schaeffer and Williams, this issue). The flow rate was later translated into evapotranspiration rates per canopy area, using information of sapwood area and sizes of the trees. This conversion was very empirical at the time of this study due to a lack of sophisticated measurements of the dimensions of the trees, distribution and densities of branches. It is expected that precise measurements of these physical properties, including green leaf area index (LAI) will be used to derive ET of cottonwood canopy in the near future. In addition, a third tower was placed near one of the trees where sapflow was measured. On this tower were infrared thermometers to record both air temperature and the canopy surface temperature of the cottonwood trees.
All devices for ET measurements were cross-calibrated to ensure compatibility before the August intensive field campaign. The ET flux measurements were synchronized at a 30 minute time interval (except the Bowen ratio device which was at a 20 minute interval) so that comparison could be easily made.
3. METHODOLOGY
Water loss via evapotranspiration is primarily driven by the water vapor difference between the substomatal cavities in the leaf and the ambient air. As water evaporates, heat is extracted both from the canopy leaves and from the air that is near the canopy surface. This results in a lower canopy temperature than the ambient air temperature. It is therefore a reasonable assumption that the ET rate is a function of differential temperature dT=Ts - Ta, where Ts is the canopy surface temperature and Ta is air temperature. This can be theoretically derived from energy balance equation:
ET = Rn - G - H (1)
where Rn is net radiation (W/m2), G is soil heat flux, and H is sensible heat flux. Jackson et al(1977) suggested relating daily sensible heat to differential temperature dT:
ET = Rn -G - B (dT) (2)
where B is a constant depending primarily on surface roughness and wind speed. This approach was used in various studies (Seguin and Itier, 1983; and Moran et al., 1994). The soil heat flux G was assumed to be zero when daily average was used. Using equation (2) with instantaneous remote sensing imagery requires further assumptions, because G can be relatively large and and Rn may vary from location to location. As a first approximation, we rewrite equation (2) in the following format:
ET = A - B (dT) (3)
where A = Rn - G, which is sometimes termed the "available energy". Parameters A and B can be determined empirically using ground data. Thus ET values can be estimated with the differential temperature dT which can be obtained from remote sensing images.
To investigate the feasibility of this approach, ET measurements with Bowen ratio devices over the sacaton and mesquite sites were averaged from 9:00am to 10:00am, corresponding to the time period of the Agrometrics overpasses. The differential temperature dT was also computed during this period to establish the ET - dT relationship. The sapflow rate of the cottonwood canopy was initially recorded in the unit of cm3H20/cm2/hr, where the cm2 was the sapwood area. This unit was converted to ET per unit canopy area in order to use equation (3).
4. RESULTS AND DISCUSSION
The hourly averaged ET data were plotted against their corresponding
differential temperatures,dT in Figure 2. 
A
linear regression model was fit to the data, where A=184, B=36.8 and r2=0.45.
Note that the data were acquired from three types of vegetation: cottonwood,
sacaton, and mesquite. Whether one can fit a single line through different
vegetation types remains questionable. Since we did not have enough data
points for each vegetation type, we assumed that parameters A and B were
constant, or similar in magnitudes for the three vegetation types, thus
allowing us to fit a single line for all data points.
As can be seen, the regression line reaches zero at dT=5. This suggests that the differential temperature dT is not the only driving force for evapotranspiration. Other factors such as the water vapor and wind speed are also important environmental variables that affect evapotranspiration. For the sacaton grasses, the surface temperature is generally higher than the ambient air temperature, having minimum ET values compared with mesquite and the cottonwood canopy. The cottonwood appears much cooler than the ambient temperature. It should be pointed out here that the ambient temperature of the cottonwood was recorded at a short distance from the cottonwood tree. The water flux may well affect the ambient temperature. The proper distance from the canopy surface for measurement of ambient temperature is still under investigation. In this study, the ambient temperatures at 2 meters above the sacaton and mesquite sites were used, whereas the ambient temperature at 2 feet adjacent to the cottonwood canopy was used for cottonwood site.
The ET values estimated with remotely sensed surface temperature were
compared with in situ measurements acquired with ET values measured with
Scintillometer, 3-D sonic and Bowen ratio techniques in Figures 3-5 for
the three major vegetation types. For the sacaton site (Figure 3), the
ET value estimated with remote sensing techniques was very close to the
ET values measured with Bowen ratio device, but lower than those estimated
with the other two devices. 
It
should be pointed out here, the comparison between remote sensing and Bowen
ratio estimated ET values does not serve as a validation, since the equation
was derived based on the Bowen ratio data. Comparison of ET values with
the other two devices indicated that the remote sensing technique underestimated
the ET values of the sacaton canopy. It should be further noted that an
average Ts value of all sacaton areas, including those that were covered
by the other three devices, was used in the estimation of ET values with
remote sensing techniques. Therefore, the remotely sensed ET values of
sacaton represent a larger area than those covered by the other three devices.
This may have resulted in some differences between remotely sensed ET and
values from other devices.
For the mesquite site, ET values estimated with all four techniques
are presented in Figure 4. It appeared that the Bowen ratio underestimated
ET values of the mesquite shrubs compared with the other techniques, including
remote sensing techniques. It is interesting to note that the remotely
sensed ET value is very close to the values derived from Scintillometer
and 3-D sonic devices, although the equation was derived based on data
from Bowen ratio device. The difference between ET values from remote sensing
and Bowen ratio was expected from equation (1) and Figure 2, where the
equation line is above the mesquite data points. The estimate values with
remote sensing techniques, however, were close to the ET values derived
by the other two devices. Again, it should be kept in mind that the remotely
sensed ET values represent a much larger area than those covered by any
of the devices used, because the entire mesquite area of the Lewis Spring
site was used in the ET calculations. 
Because other devices were not deployed to measure the ET values of
the cottonwood canopy, the remote sensing ET estimates of the cottonwood
were compared in Figure 5 for different trees with the values obtained
with sapflow techniques. ET estimates from remotely sensed images generally
overestimated the ET values compared with the sapflow measurements. For
all trees where the sapflow measurements were made, surface temperature
Ts was obtained from remote sensing images. The ET values were estimated
with these Ts and the ambient air temperature recorded at tree 203, assuming
that the Ta was constant for all trees studied. It should also be noted
that the tree sizes were empirical estimates based on the sapwood areas.
Even if accurate measurements were made about these tree sizes, it is difficult
to translate the sapflow rate to ET, because the effective canopy areas
are almost impossible to be inferred from the physical sizes of the trees,
and even more difficult from the sapwood areas estimates. Scaling from
individual tree estimates of ET to canopy level is another issue to be
further studied. Nevertheless, an actual measurement of these tree sizes
may improve the accuracy of the ET estimates. 
With
these issues being considered, the ET values estimated from remotely sensed
data are a reasonable approximation to the actual water loss from the cottonwood
trees.
The spatial distribution of ET rates at the Lewis Spring site was mapped using the thermal images acquired with Agrometrics sensor (Figure 6) and equation (3). The cottonwood transpired water at a much higher rate than the sacaton grasses and mesquite shrubs. The lowest ET rate was found for the sacaton grasses. Field records indicate that most sacaton grasses were almost scenesed during the August field campaign while the cottonwood and mesquites reached maximum greenness. Much of the scenesed sacaton and litter covered new growth, preventing water evaporation and transpiration. It is also noted that the sacaton grasses were clumped and, therefore, there were exposed bare soil surfaces. This resulted in higher surface temperature, and the reduced ET rates.
The water loss rates at the Lewis Spring site can be estimated based on the ET map(Figure 6). The total water loss rate from the three major vegetation types, cottonwood, mesquite, and sacaton, was estimated to be 4125, 3465, and 1335 kilograms of water per hectare per hour (Table 1). Assuming that these numbers were typical values at 9:00am-10:00am period, and this rate lasts for 9.5 hours based on the sapflow data (Schaeffer and Williams, this issue), the total daily loss of water via evapotranspiration from these three classes were 39118, 32918, and 12683 kilograms of water per hectare.
Using another remote sensing image (TIMS data acquired by JPL in May 1996) that covered the entire corridor, and assuming that these water loss rates are representatives of the whole corridor (range from Lewis Spring to Fairbanks; see Figure 1 in Goodrich et al., in this issue), then the total estimates of water loss for the entire corridor were approximately 4336, 41637, and 2296 tons of water per day from cottonwood, mesquites, and sacaton grasses (these three classes account for up to 52% of all entire corridor). Water losses from other classes were expected to be much less than these three, because the remaining classes were primarily bare soils. The total of water loss from the corridor was summed to be about 48270 tons per day. These values are equivalent to 176 thousands gallons of water per day per hectare evaporated from these three major vegetation. The daily evaporative water loss for the entire riparian corridor was estimated to be about 1.0x 108 gallons.
5. FUTURE WORK
It is expected that when multitemporal remote sensing images are processed, temporal interpolation and extrapolation with simple models (Goodrich et al.(b), in issue) between and beyond remote sensing acquisition dates, can be accomplished to establish the seasonal variation of water loss at the corridor and total consumptive water use of the riparian vegetation in the near future.
6. ACKNOWLEDGMENT
Financial support from the USDA-ARS Global Change Research Program, NASA grant W-18,997, NASA Landsat Science Team, grant #S-41396-F, USDA National Research Initiative Grant Program, Arizona Department of Water Resources, ORSTOM, the French Remote Sensing Program (PNTS) via the VEGETATION Project and the ERS2/ATSR2 Project. Assistance was also provided in part by the NASA/EOS grant NAGW2425, and Ft. Huachuca; this support is also gratefully acknowledged. Special thanks are extended to the ARS staff located in Tombstone, Arizona for their diligent efforts and to USDA-ARS Weslaco for pilot and aircraft support.
7. REFERENCES
Carey Act, 1894, Federal government ceded land to the state for irrigation.
Chehbouni, Hartogensis, Hipps, Brunel, Watts, Rodriguez, Boulet, Dedieu, Kerr, De Bruin, 1998, Sensible heat flux measurements using a large aperture scintillometer over heterogeneous surfaces.
Goodrich D. C., A. Chehbouni, B. Goff, et al., 1998(a), An overview of the 1997 activities of the semi-arid land-surface-atmosphere (Salsa) program.
Goodrich, G. D., M.S. Moran, R. Scott, J. Qi et al , 1998(b) Seasonal Estimates of Riparian Vegetation evapotranspiration (Consumptive Water Use) using Remote and In-Situ Measurements.
Jackson R. D. R. J. Reginato, S. B. Idso, 1977, Wheat canopy temperature: A practical tool for evaluating water requirements, Water Resourc. Res., 13, 651-656.
Moran S. M., T. R. Clarke, W. P. Kustas, M. Weltz, and S. A. Amer, 1994, Water Resourc. Res. Vol. 30, no. 5, 1287-1297.
Moran S. M., D. Williams, D. Goodrich, et al., 1998, Overview of remote sensing of semi-arid ecosystem function in the upper San Pedro River Basin, Arizona.
Schaeffer, S., and D. Williams, 1998, Transpiration of desert riparian forest canopies estimated from sap flux.
Schieldge, J., and A. Kahle, Use of remotely sensed and in situ data to investigate the effects of surface heterogeneity and terrain on the spatial and temporal distribution of sensible and latent heat fluxes in the semi-arid San Pedro basin.
Scott R. L., W. J. Shuttleworth, and D. C. Goodrich,1998, Water use of two dominant riparian vegetation communities in southeastern Arizona, Seguin B., B. Itier, 1983, Using midday surface temperature to estimate daily evaporation from satellite thermal IR data, Int J., Remote Sens. 4, 371-383.
homepage