The impact on area-averaged heat fluxes from using remotely sensed data at different resolutions: A case study with Washita '92 data

Abstract

Landscape-scale fluxes are derived using an energy balance model with remotely sensed input data at different pixel resolutions. The model explicitly evaluates energy flux contributions from the soil and vegetation using remotely sensed near-surface soil moisture and normalized difference vegetation index (NDVI) to define key model variables. The remotely sensed data used in the model were collected as part of the Washita '92 Experiment conducted in the Little Washita watershed, a subhumid catchment in southwest Oklahoma. Near-surface soil moisture maps covering approximately an 800 km2 area at 0.2 km resolution were generated using aircraft-based passive microwave observations from the L band electronically scanned thinned array radiometer (ESTAR). A spatially distributed NDVI map was derived from a SPOT satellite image. Daily values of the midday Bowen ratio (B(O), ratio of sensible to latent heat flux) computed by the model indicated that areas with low vegetation cover or bare soil and senescent vegetation were drying out significantly (i.e., dramatic increases in B(O)) while other areas with higher vegetation cover showed smaller increases in B(O) in response to a drying soil surface. After a few days of drying, B(O) values computed by the model ranged from ~0.1 to values ≥2 over the image. The resulting spatially distributed maps of B(O) suggested a heterogeneous surface at the field or patch scale. A satellite-based L band ESTAR would have a pixel resolution on the order of 101 km or larger (i.e., landscape scale). Since fluxes are nonlinearly related to model input variables/parameters, defining model inputs at resolutions significantly larger than the patch scale may cause significant errors in flux calculations. Potential errors in flux calculations were investigated by evaluating differences in area-averaged flux estimates for the ~18 x 45 km study area using the model with the remotely sensed near-surface soil moisture and NDVI data at the following pixel resolutions: (1) 0.2 km (original pixel resolution), (2) 9 km, and (3) the whole image, which has an effective pixel resolution on the order of 28 km. Differences in the area-averaged sensible and latent fluxes computed by the model at the three resolutions were within 10%. Such minor discrepancies in the area-average fluxes suggests that the heterogeneity in soil moisture and vegetation type and cover was not of the type to affect regional flux predictions using significantly coarser resolution information. There was a consistent decrease in area-averaged latent heat flux estimates at the coarser resolution caused by biases in area-averaged values of the other three energy flux components. The consistent decrease in area-average latent heat flux caused the area-averaged midday B(O) to increase with decreasing pixel resolution resulting in ≃15% increase from 200 m to using the full image. These results, however, depended on the method used in aggregating leaf area index to the coarser resolutions, indicating that techniques for scaling up key model parameters have a significant effect on area-average flux predictions.