In this chapter, numerous methods were reviewed for using remote sensing to estimate XE. Based on a similar review conducted by Kustas and Norman (1996), a series of issues were identified as important for remote sensing of XE from measurements, modeling studies, and theoretical considerations. A slightly revised list of these issues is included here:
2. Most models are sensitive to errors in raero — Ta and u, yet the measurement of Ta and u at the time and location of the Tnd observation is not typically available.
3. Tmi dependence on view angle cannot generally be neglected because differences in vegetation and soil temperatures can be significant depending on soil moisture conditions.
4. Thermal emissivity is only known approximately on the pixel scale.
5. Atmospheric corrections and satellite calibrations contribute significant errors in the measurements of pA/ and Trii that are not always adequately known.
6. Remote observations are instantaneous, while integrated fluxes are desired on hourly, daily, or longer time scales.
7. Satellites with larger pixel sizes (l to 4 km) can provide sufficiently frequent observations in time (i.e., GOES), but may have uncertainties related to the averaging over heterogeneous subpixel areas.
8. Continuous (hourly or daily) surface flux estimates are most useful, and clouds cause remote observations to be intermittent.
Kustas and Norman (1996) provided a representative list of models using remote observations to estimate XE and attempted to characterize which of the above eight issues each of these models addressed. None of the models address all the important issues at the present time, but several of the models address some of the important issues (l, 3, 4, and 6). Fewer models addressed the most critical issues of spatially distributed meteorological data and atmospheric correction of satellite image data (2 and 5). Related to issue 2, meteorological data acquired at a time or location other than that of the Tmd or VI observation can cause substantial error in the estimate of XE. Moran and Jackson (1991) reported that errors in extrapolation of Ta greater than 1°C were unacceptable for estimation of XE using the energy balance approach. They also reported that measurements of Ta measured at 2 m height over adjacent fields of bare soil and lush vegetation differed by up to 3°C at midday. Similarly disturbing results have been reported for wind speed estimation. Rahman (1996) compared a wind speed map constructed by simple interpolation of u values from local weather stations with a map of wind speed derived from the Regional Atmospheric Modeling System (RAMS; Pielke et al., 1992) that accounted for topographic effects. The RAMS-derived map of u was a substantial improvement over the simple interpolation because it accounted for the relatively strong winds in the passes between mountain ranges and relatively light winds in the lee of the ranges.
Related to issue 5, accounting for the attenuation of the radiances received by satellite-based sensors is not a trivial matter (Kaufman, 1989; Price, 1989). In correcting thermal-infrared data, whether using radiative transfer models or split-window techniques, the uncertainty is l to 3°C over land surfaces (Becker and Li, 1990; Perry and Moran, 1994). Model sensitivity to such an uncertainty in Tmd can be significant, especially over large vegetation where errors can be M00W/m2 for hourly to daily time scales (Norman et al., 1995a). However, the l50W/m2 uncertainty in estimating sensible heat flux from radiometric surface temperature observations suggested by Sellers et al. (1995b) is in many cases two to three times larger than errors reported by other researchers (Choudhury, 1994). All the methods reviewed in this chapter are based on the assumption that accurate remotely sensed estimates of surface reflectance, temperature, and backscatter will be readily available. At this time, they are not. A primary challenge will be to improve the accuracy and consistency of remotely sensed information with an insight into the accuracy requirements of operational models and algorithms.
None of the models explicitly addressed the issue of subpixel averaging, often termed aggregation (issue 7). Aggregation refers to spatial averaging of some heterogeneous surface variable to obtain an effective value representative of an area. In an assessment of the state of the art in aggregation research, Michaud and Shuttleworth (1997) concluded that, over flat terrain, simple aggregation rules applied to surface properties could result in simulated values of XE within 10% of fluxes from models with full representation of heterogeneity. Furthermore, they concluded that aggregation rules for vegetation characteristics were relatively straightforward in the case of patch-scale heterogeneity (variability of 100 to 1000 m). However, mesoscale heterogeneity (10 to 100 km) in surface cover will need to be addressed through more complicated types of parameterization and, in mountainous terrain, the influence of topography on near-surface meteorology must be considered. In an aggregation study related to the use of remote-sensing data for energy balance evaluation, Moran et al. (1997a) found that aggregation of remotely sensed measurements in sparse canopies could be accomplished with little error (such as aggregation of TriA from 1 m2 to 1 km2) but not others (such as aggregation of H to 1 km2). Kustas and Humes (1996) applied the Norman et al. (1995b) dual-source model for computing basin-scale fluxes with TnA at 120-, 1000-, and ~8000-m pixel resolution over a semiarid rangeland landscape. They found minor changes in the fluxes aggregated from the different resolutions. Sellers et al. (1995a) investigated the impact of spatial variation in topography, vegetative cover, and soil moisture on area-averaged fluxes simulated by a SVAT model over a 2 x 15 km domain. They found simple averages of these parameters introduced minor errors in the SVAT simulations of the area-averaged fluxes. Still, other studies (Crosson et al., 1993; Sellers et al., 1992) suggest that issue 7 may be a significant problem at the 1-km scale but may average out at the 10-km scale (Norman and Divakarla, 1995).
None of the current models address the issue of continuous surface fluxes even with clouds, but studies are in progress to combine the thermal infrared remote-sensing approaches discussed in this chapter with mesoscale models and with a simplified land-atmosphere exchange model (Anderson et al., 2000). If issues 1 to 7 are addressed adequately, issue 8 will not limit remote estimation of regional XE fluxes.
Was this article helpful?