An intercomparison of three remote sensing-based surface energy balance algorithms over a corn and soybean production region (Iowa, U.S.) during SMACEX

Reliable estimation of the surface energy balance from local to regional scales is crucial for many applications including weather forecasting, hydrologic modeling, irrigation scheduling, water resource management, and climate change research. Numerous models have been developed using remote sensing, which permits spatially distributed mapping of the surface energy balance over large areas. This study compares flux maps over a relatively simple agricultural landscape in central Iowa, comprised of soybean and corn fields, generated with three different remote sensing-based surface energy balance models: the Two-Source Energy Balance (TSEB) model, Mapping EvapoTranspiration at high Resolution using Internalized Calibration (METRIC), and the Trapezoid Interpolation Model (TIM). The three models have different levels of complexity and input requirements, but all have operational capabilities. METRIC and TIM make use of the remotely sensed surface temperature-vegetation cover relation to define key model variables linked to wet and dry hydrologic extremes, while TSEB uses these remotely sensed inputs to define component soil and canopy temperatures, aerodynamic resistances, and fluxes. The models were run using Landsat imagery collected during the Soil Moisture Atmosphere Coupling Experiment (SMACEX) in 2002 and model results were compared with observations from a network of flux towers deployed within the study area. While TSEB and METRIC yielded similar and reasonable agreement with measured heat fluxes, with root-mean-square errors (RMSE) of approximately 50-75W/m², errors for TIM exceeded 100W/m². Despite the good agreement between TSEB and METRIC at discrete locations sampled by the flux towers, a spatial intercomparison of gridded model output (i.e., comparing output on a pixel-by-pixel basis) revealed significant discrepancies in modeled turbulent heat flux patterns that were largely correlated with vegetation density. Generally, the largest discrepancies, primarily a bias in H, between these two models occurred in areas with partial vegetation cover and a leaf area index (LAI)<2.0. Adjustment of the minimum LE assumed for the hot/dry hydrologic extreme condition in METRIC reduced the bias in H between METRIC and TSEB, but caused a significant increase in bias in LE between the models. Spatial intercomparison of modeled flux patterns over a variety of landscapes will be required to better assess uncertainties in remote sensing surface energy balance models, and to work toward an improved hybrid modeling system.