This paper explores the utility of NOAA AVHRR data for evaluating a key component in energy and water balance modeling, namely the evapotranspiration (ET). Four scenes covering a wide range of meteorological, vegetation and soil moisture conditions were processed. Values of daily ET are calculated using a simplified, yet operational, model which requires as primary input an estimate of surface temperature, T_s, and air temperature, T_a, near the time of maximum heating. Average values of these temperatures for the watershed are computed and rational estimates of daily ET are determined using, model coefficients found for a more extensive data set taken over an arid-semiarid region in Africa. s-values derived from satellite thermal-IR data due to atmospheric effects, sensor calibration, surface emissivity and image registration are discussed. Comparisons between the model and ground-based estimates of average values of T_s and ET for the basin are presented.

The main emphasis of these models is to evaluate regional ET. Significant strides have been made in developing models which have much of the physics involved in simulating soil-plant-atmosphere processes.²³ However, difficulties involved in specifying a large number of input parameters which depend upon soil and vegetation structure at regional scales limits their applicability to relatively uniform surfaces or areas with detailed hydrologic information. Moreover, effects of heterogeneous surfaces on remote sensing data may actually be a more important issue than how complex the surface-atmosphere interface is modeled.⁵ Finally, accounting for atmospheric attenuation, surface emissivity and sensor calibration are other important issues.¹⁶

This paper discusses the feasibility of using NOAA-11 AVHRR (Advanced Very High Resolution Radiometer) during the dry and wet (monsoon) seasons in the semiarid southwestern United States for energy and water balance modeling at regional scales. The study area was the Walnut Gulch Experimental Watershed (31? 43' N 110? W) operated by the USDA - Agricultural Research Service. During the experimental periods ground-based hydrometeorological data along with aircraft and satellite remote sensing data were collected. A summary of the study is given by Kustas et al.¹²

Issues related to correcting AVHRR data for atmospheric effects, including cloud screening, sensor calibration, emissivity and image registration will be discussed. A semi-empirical approach which uses processed AVHRR data in the thermal-IR for estimating daily ET ^{10, 18}  is evaluated.

Comparison of the daytime fluxes between the variance method and more accurate measurements using eddy correlation²⁷ showed satisfactory agreement.¹²

Data from the METFLUX sites were collected continuously during the experiment and 20 minute averages were recorded. The field campaign in June, during the dry season, had 2 METFLUX stations operating, one was in a grass and the other in a primarily shrub area. The field study during the wet or "monsoon" season had all 8 stations collecting data.

For the dry season, NOAA-11 AVHRR data (a day-night pair) were collected on June 5, Day 156. For the wet season, day - night images were obtained for July 25-28 (Day 206-209), August 8-9 (Day 220-221) and day scenes for July 31 (Day 212) and August 4 (Day 216). From this original set of images, further processing suggested that observations on Day 156, 209, 212 and 216 were suitable for further analyses.

This set of days covered a range of environmental conditions. Table 1 summarizes the general surface and weather conditions for each day. In brief, Day 156 had low soil moisture and senescent vegetation; Day 209 had a dry soil surface and actively transpiring vegetation; Day 212 had a dry surface and drying subsurface soil moisture with vegetation showing some stress; Day 216 had relatively wet soils and unstressed vegetation for a majority of the watershed.

Table One

* Day 216 followed a major rainfall event on Day 213 which ranged from 0 mm being recorded around the west end of the watershed to over 50 mm for the northeast section of the watershed. Several smaller (15 mm or less) and more localized storms occurred on Day 214 and 215 which primarily rewetted areas hit by the rainfall event on Day 213. Hence, a small area of the watershed probably maintained soil moisture conditions similar to Day 212. The influence of spatial variability in rainfall on estimates of representative values for the surface energy balance and surface temperature for the basin is beyond the scope of this paper.¹²

Approach for Estimating Evapotranspiration

A method which estimates daily ET using the difference in surface temperature and air temperature around midday was first proposed by Jackson et al.¹⁰ Their approach was essentially statistical whereby the daily surface energy balance could be expressed as follows:

LE_d - Rn_d - A - B (T_s - T_a) _i (2)

where the subscripts d and i represent daily fluxes in mm/day and "instantaneous" values obtained remotely, and A and B are coefficients. An observational and theoretical study by Seguin and Itier ¹⁹ showed that (2) had potentially broader application. Since then, Eq. (2) and derivatives of it have been compared with ground-truth observations, and more physically-based model estimates ^{3, 5, 11, 13, 18, 24} In general, these studies suggest that (2) is a useful and relatively simple technique for evaluating ET; but reliable results may be restricted to mostly clear days. Nevertheless, its simplicity allows it to be operational and makes it an attractive approach for generating ET maps over large areas. ^{11, 18}

Figure One

Surface Measurements

The components of the surface energy balance (Eq. 1) from the 8 METFLUX sites for the days used in the analysis are plotted as daily totals (mm/day) in Figure 1. Variation among the sites is not substantial, but it is yet to be seen whether differences in vegetation type and density, and soil moisture conditions are correlated with variation in the fluxes. Figure 2 is an example of the diurnal trace in T_s and T_a from the 8 locations. The large variation in T_s among the sites is mainly attributed to the amount of bare soil monitored by the infrared thermometer (IRT). Emissivity for the IRTs was set to 0.99. These sensors having a 15? field of view observed a surface area of order 0.33 m ² at a measurement height of 2.5 m. Hence, the spatial sampling for many of the sparsely vegetated sites may not have been adequate. Air temperature exhibited less variation where differences were no greater than 2 C on average.

Figure Two

Satellite Data

The five-channel NOAA-11 AVHRR LAC data were obtained in digital 1B format. The visible, near-infrared and two thermal channel data (channels 1, 2, 4 and 5) were processed. The image was registered using the Digital Line Graph data provided by the U.S. Geological Survey. The data were resampled once to a Universal Transverse Mercator (UTM) projections. The accuracy of the data is about 1 pixel (1. 1 km). Calibration of the visible and thermal data were performed according to NOAA guidelines. ¹⁴ The thermal channel data were corrected for the nonlinear response of the sensors.

Potential Errors in NOAA 11 AVHRR Remotely Sensed Surface Temperatures

In this study, registration is assumed to be accurate to within 1 pixel (1.1 km). All images used were coregistered to within 1 pixel using a USGS 1:2,000 000 Digital Line Graph (DLG) showing regional river systems. The resulting error in T_s from a 1 pixel registration error can be approximated by calculating the average of the differences in temperature between a pixel and its 4 closest neighbors. The average differences were calculated for each of the watershed pixels (163 in total) for Day 156, 209 and 216. The average registration errors for the dry cases (before rain events) were less than 0.5 K. For Day 216 (August 4), the errors were much larger, with an average of approximately 1.1 K. Since the average value of T_s for the entire watershed was used in this study, registration effects are considered negligible because an offset of one pixel in any direction is not likely to affect the average of 163 pixels. However, studies comparing values from an AVHRR LAC (Local Area Coverage) scene to ground measurements may be subject to registration errors greater than 1.0 K.

The relation between AVHRR voltage (digital counts) and scene radiance is not linear. NOAA has published corrections for this non-linearity error as a function of scene brightness temperature and internal blackbody temperature. ¹⁴  These corrections were applied to the AVHRR data used in this study. However for the scenes considered, the brightness temperatures in the watershed are close to or only slightly exceed the maximum brightness temperature value used in the calibration equation. Therefore calibration errors in channels 4 and 5 should be close to the estimated uncertainty in the calibration equation, which is of order 0.2 K.⁸ Yet it should be noted that as brightness temperatures exceed the non-linearity calibration equation, the uncertainty will be larger. In estimating the land surface temperature, the difference in the radiometric temperatures for the two thermal channels is used. Consequently, using the mean sum of squares, the uncertainty would be around 0.3 K for this difference.

The effects of clouds were evident for Day 212 and 216. No attempt was made to correct cloud contaminated pixels, so any pixel that appeared to be partly or entirely cloudy was eliminated in the analysis. For Day 212 (July 31) data, a cloud composite was developed using channel 4 brightness temperature, NDVI (Normalized Difference Vegetation Index) and channel 1 albedo. This composite image was used to select two cloud-free areas totaling 63 pixels. For Day 216 (August 4), a cloud composite was also developed that showed the upper northwest corner of the watershed to be contaminated by haze. After examination of all pixels in the watershed, the pixels with a water vapor correction exceeding 50 % of the channel 4 brightness temperature were deleted leaving 150 pixels for further analysis.

The split window model developed by Price ¹⁶ was used to correct for atmospheric attenuation of the radiance received by the satellite in the thermal channels. For the present data set, this method was found suitable for the water vapor correction. ⁷ The equation essentially uses the difference in the water absorption in the two thermal channels which results in a relatively simple formula for calculating the surface temperature:

where T_ch4and T_ch5 are channels 4 and 5 brightness temperatures (K), C = 1/(R-1) which is about 3.33 for the present study, and 0_ch4and 0_ch5 are the emissivities of the surface for channels 4 and 5. The uncertainty in R, the ratio of wavelength dependent scaling factors of water vapor absorption, may cause significant errors in the value of T_s. This can be assessed by taking the derivative of T_s in Eq. (3) with respect to R:

Price¹⁶ suggests a value of 0.03 as an estimate of the uncertainty in R. For the monsoon season, (T_ch4 - T_ch5) were typically 3 to 5 K. Thus considering a 5 K difference with an uncertainty of 0.03 in R, the resulting error in T_s would be around 1. 5 K.

Emissivity errors in the split window estimation of T_s have been studied by Becker and Li.² They demonstrate that errors in excess of 3 K are produced by assuming theemissivity in channels 4 and 5 are equaltounity. Unfortunately, emissivity values for the watershed in channels 4 and 5 are not available; hence emissivity errors cannot be accounted for in this study. An estimate of the uncertainty in the value of T_s can be obtained by taking the derivative of Eq.(3) with respect to channel 4 and 5 emissivity:

where *0_ch4 and *0_ch5 represent the uncertainties in the values of channels 4 and 5 emissivities. For example, given brightness temperatures T_ch5 = 307 K and T_ch4 = 310 K and true emissivity values of 0_ch5 = 0.98 and 0_ch4 = 0.975, the resulting error in T_s from Eq. (5) would be 2.9 K. From Eq. (3) it is easily seen that larger differences in channel 4 and 5 emissivities results in a greater absolute error in T_s. Conversely, as the differences in emissivity approaches zero, the final term in Eq. (3) becomes negligible.

Since emissivity values in the thermal bands are unknown, the radiometric surface temperature was evaluated with Eq. (3) neglecting the last term:

An estimate of total potential uncertainty is obtained by combining a conservative estimate of a ? 1 K error for the calibration and a ? 2 K error resulting from the uncertainty in the absorption factor R. The sum of squares error is approximately ? 2.5 K.

Histograms of surface temperature values for the watershed extracted from the NOAA-11 AVHRR images for Day 156 and Day 209 are shown in Figure 3 .

Figure Three

This figure suggests that surface temperature values in the basin during the dry season are more uniformthan during the wet (monsoon) season. This is primarily the result of larger variations in soil moisture and green vegetation cover caused by more frequent and localized rainfall during the monsoon season. Figure 4 illustrates the comparison between the ground observations of surface temperature (T_s averaged for all METFLUX sites) and the mean value for the watershed given by the satellite.

Figure Four

The vertical and horizontal bars represent 1 standard deviation from the mean given by the METFLUX and satellite data, respectively. The line represents perfect agreement. Although the standard deviation is quite large for T_sestimated from the ground station data, it appears that satellite derived values are significantly less than the ground-based estimates of T_s for Day 156 and 209 while there is good agreement for the other two days. Day 156 and 209, incidently, had clear skies and relatively uniform surface soil moisture conditions; thus atmospheric effects or the smaller area sampled by the METFLUX network are not the likely causes for the disagreement. Further analysis to determine the factors causing disagreement between ground and satellite estimates of T _s will be performed.

Table 2 lists the daily average and standard deviation of the components of the surface energy balance estimated from the METFLUX network as well as the surface-air temperature difference using a mean T_a from the station data and T_s from theNOAA-11 AVHRR satellite.

Since there are only four points in the present study, it was decided to plot Eq. (2) with coefficients derived from a more extensive data set over an arid-semiarid region (Kerr et al.¹⁸) and see how closely the four points lie to the curve. Additionally, because a daily value of G was calculated and found not to be negligible it was included in Eq. (2), i.e.,

The constant A has a value of 1 and coefficient B equaled 0.2. The comparison between model estimates of the right hand side of (7) given (T_s - T_a)_i and mean values from the METFLUX network is shown in Figure 5.

Table Two

Figure Five

The coefficients A and B were determined from the results of Kerr et al.¹⁸ Values of T_s and T_a for the basin were derived from the satellite and METFLUX network, respectively. The vertical and horizontal bars represent estimated errors in temperatures (? 2.5 K) and fluxes (? 0.5 mm/day). Only one point, which is from the observations taken on Day 156 (dry season), falls significantly away from the curve; however, all points fall within the scatter observed in other studies (e.g., Seguin and Itier¹⁹, Fig. 2). A more quantitative analysis is given in Table 3 which lists ET estimated using Eq. (7) and mean ET-values measured by the METFLUX network. An estimate of error between the model and observations is given by the Root Mean Square Error (RMSE; see Table 3 for definition).

Table Three

This result suggests that if Rn_d and G_d can be accurately determined, ET estimates using Eq (7) for semiarid regions are probably within 1 mm/day (i.e., RMSE = 0.7 mm/day from Table 3).Given the simplicity of this approach, an error of this magnitude is acceptable and could serve as a first order approximation to test more complex regional ET-model calculations.

In this study, the feasibility of estimating daily ET at the basin scale was tested with NOAA-11 AVHRR data collected during MONSOON 90. Images on four days could be processed which allowed an average surface temperature T_s for the basin to be computed. Potential errors in T_s associated with scene registration, instrument calibration, atmospheric attenuation (including clouds), and surface emissivity suggested an uncertainty of order 2.5 K. A comparison between the basin average T_s from the satellite and ground-based measurements from the METFLUX network suggested a significant disagreement when T_s-values were of order 50 C.

Three of the four values of LE_d - Rn_d + G_d versus (T_s - T_a)_i fell relatively close to the line given by Eq. (7), which is similar to the one obtained in a arid-semiarid region in Africa.¹⁹  The outlier, Day 156, requires a 20 K surface-air temperature difference for satisfactory model agreement with measured LE_d. However, differences of this magnitude with the satellite data were not observed in the watershed. The estimated values of basin-scale ET using Eq. (7) were generally within 1 mm/day of the observations averaged for the METFLUX network.

The fact that only four daytime scenes out of a possible twenty (or 20 %) were useable for this analysis due to cloud contamination reveals the problem in using satellite data in the optical wavebands, which is compounded by the relatively low frequency of coverage. Furthermore, the lack of a significant number of cloud free images makes it difficult to calibrate and use operationally simple ET models like Eq. (2). The use of geostationary satellites may alleviate the problem of obtaining remote sensing information on a daily basis over the area of interest, but it is unlikely that it can be used in evaluating ET with Eq. (7).

This study reveals the need for a regional ET model which can be initialized with remote sensing and meteorological data as described here and periodically updated and checked when conditions are again favorable for satellite acquisition. Such approaches have been developed and are being tested at the regional scale.¹⁵

1. Andre, JC, JP Goutorbe and A. Perrier (1986) HAPEX-MOBILHY: A Hydrologic Atmospheric Experiment for the study of Water Budget and Evaporation Flux at the Climatic Scale. Bull. Amer. Meteor. Soc., 67, 138-144.
2. Becker, F and Z-L Li (1990) Towards a local split window method over land surfaces. Int. J. Remote Sens. 11, 369-393
3. Brunel, JP (1989) Estimation of sensible heat flux from measurements of surface radiative temperature and air temperature at two meters: Application to determine actual evaporation rate. Agric. For. Meteor. 46, 179-191
4. Carlson TN (1986) Regional-scale estimates of surface moisture availability and thermal inertia using remote thermal measurements. Remote Sens. Rev. 1, 197-247
5. Carlson, TN and MJ Buffum (1989) On estimating total daily evapotranspiration from remote surface temperature measurements. Remote Sens. Env. 29, 197-207
6. Carlson, TN, EM Perry and TJ Schmugge (1990) Remote estimation of soil moisture availability and fractional vegetation cover for agricultural fields. Agric. For. Meteor. 52, 45-69
7. Doraiswamy, PC, and EM Perry (1991) The relationship between satellite-based surface temperature and vegetation indices over a semi-arid rangeland watershed. Proceedings of AMS Special Session on Hydrometeorology, Salt Lake City, Utah Sept. 9-13, (in Press).
8. ITT Aerospace (1980) AVHRR/2 Advanced Very High Resolution Radiometer technical description. Contract NAS-526771, ITT Aerospace, Optical Division, Fort Wayne, IN, 304 pp.
9. Jackson, RD (1985) Evaluating evapotranspiration at local and regional scales. Proc. IEEE 73, 1086-1095.
10. Jackson RD, RJ Reginato and SB Idso, (1977) Wheat canopy temperature: A practical tool for evaluating water requirements. Water Resour. Res. 13, 651-656.
11. Kerr, YH, E. Assad, JP Freteaud, JP Lagouarde and B. Seguin (1987) Estimation of evapotranspiration in the Sahelian zone by use of METEOSAT and NOAA AVHRR data. Adv. Space Res. 7, 161-164.
12. Kustas, WP, DC Goodrich, MS Moran, SA Amer, LB Bach, JH Blanford, A. Chehbouni, H. Claassen, WE Clements, PC Doraiswamy, P. Dubois, TR Clarke, CST Daughtry, D. Gellman, TA Grant, LE Hipps, AR Huete, KS Humes, TJ Jackson, TO Keefer, WD Nichols, R Parry, EM Perry, RT Pinker, PJ Pinter, Jr, J Qi, A Riggs, TJ Schmugge, AM Shutko, DI Stannard, E Swaitek, WJD van Leeuwen, J van Zyl, A Vidal, J Washburne, and MA Weltz (1991) An interdisciplinary field study of the energy and water fluxes in the atmosphere-biosphere system over semiarid rangelands: Description and some preliminary results. Bull. Amer. Meteor. Soc. (in press)
13. Nieuwenhuis, GJ, AEH Smidt and HAM Thunnissen (1985) Estimation of regional evapotranspiration of arable crops from thermal infrared images. Int. Jour. Remote Sens. 6, 1319-1334.
14. NOAA (1988) Data extraction and calibration of TIROS-N NOAA radiometer. (Editor WC Planet) NOAA Technical Mem. NESS 107.
15. Ottl?, C, D Vidal-Madjar, and G Girard (1989) Remote sensing applications to hydrological modeling. Jour. Hydrology 105, 369-384.
16. Price, JC (1984) Land surface temperature measurements from the split window channels of the NOAA 7 Advanced Very High Resolution Radiometer. Jour. Geophys. Res. 89, 7231-7237.
17. Schmugge, TJ and F Becker (1990) Remote sensing observations for the monitoring of land-surface fluxes and water budgets. In Land Surface Evaporation: Measurement and Parameterization (TJ Schmugge and J-C Andre Editors) 424 pp.
18. Seguin, B, E Assad, JP Freteaud, J Imbernon, Y Kerr and JP Lagouarde (1989) Use of meteorological satellites for water balance monitoring in the Sahelian regions. Int. Jour. Remote Sens. 10, 1101-1117.
19. Seguin B and B Itier (1993) Using midday surface temperature to estimate daily evaporation from satellite thermal IR data. Int. Jour. Remote Sens. 4, 371-383.
20. Sellers, PJ, FG Hall, G Asrar, DE Strebel and RE Murphy (1988) The First ISLSCP Field Experiment (FIFE). Bull. Amer. Meteor. Soc. 69, 22-27.
21. Tillman, JE (1972) The indirect determination of stability, beat and momentum fluxes in the atmospheric boundary layer from simple scalar variables during dry unstable conditions. Jour. Appl. Meteor. 11, 783-792.
22. van de Griend, AA, M Owe, HF Vugts and SD Prince (1989) Water and surface energy balance modeling in Botswana. Bull. Amer. Meteor. Soc. 70, 19-24.
23. van de Griend, AA and JR van Boxel (1989) Water and surface energy balance model with a multilayer canopy representation for remote sensing purposes. Water Resour. Res. 25, 949-971.
24. Vidal, A and A Perrier (1989) Analysis of a simplified relation for estimating daily evapotranspiration from satellite thermal IR data. Int. Jour. Remote Sens. 10, 1327-1337.
25. Weaver, HL (1990) Temperature and humidity flux-variance relations determined by one-dimensional eddy correlation. Boundary-Layer Meteor. 53, 77-91.
26. Wesely, ML (1988) Use of variance techniques to measure dry air-surface exchange rates. Boundary-layer Meteor. 44, 13-21.
27. Wesely, ML, GW Thurtell and CB Tanner (1970) Eddy correlation measurements of sensible heat flux near the earth's surface. Jour. Appl. Meteor. 9, 45-50.

Copyright ? 1991by the International Astronautical Federation. The U. S. Government has a royalty-free license to exercise all rights under the copyright claimed herein for Governmental purposes. All other rights are reserved by the copyright owner.