Aggregation and Disaggregation Techniques Applied on Remotely Sensed Data to Obtain Optimum Resolution for Surface Energy Fluxes Estimation

Continuous monitoring of surface energy fluxes provides an important tool for precision agriculture management. It is, therefore, desirable to obtain these fluxes at agricultural field size (length scale ~ 10-100 m). To date, land surface temperature (LST), a fundamental input required for flux computations, is usually available at a nominal resolution of 1 km, which disables field-scale monitoring. Disaggregating LST data into field-scale sub-pixels was found to be possible, with deterioration in temperature accuracy as sub-pixel size is reduced. In contrast to LST, land use and fractional vegetation cover (LU and FC, additional key inputs) are available at high spatial resolution (e.g., 30 m). Aggregation of LU and FC to meet the lower resolution LST data introduces errors when aggregating to larger pixel sizes. The objective of this research is to find the optimum resolution that will minimize the errors due to aggregation of LU/FC and disaggregation of LST data, to provide continuous estimates of field scale surface energy fluxes. Data were used from the 2002 Soil Moisture-Atmosphere Coupling Experiment (SMACEX02) conducted over the upper Midwest corn and soybean production region of Iowa. Three dates during the period of rapid crops growth (June 23, July 1, and July 8) for which Landsat TM images are available were analyzed. The original pixels were aggregated to form 960 m pixels (to mimic thermal data currently available from MODIS) and were then disaggregated following the procedure suggested by Kustas et al. (2003)* to form 60, 120, and 240 m sub-pixels. LU and FC were obtained at 30 m resolution and then aggregated to 60, 120, 240, and 960 m. The Two-Source-Model was run at each of the resolutions using the pertinent inputs. The model output at 60 m resolution, using the original LST data was considered the base line, to which all other outputs were compared. For comparing the flux results at the lower resolutions, the 60 m flux output was aggregated. The results indicate that disaggregation of the currently available lower resolution LST to field scale sub-pixel resolutions for enabling surface energy flux monitoring for this region (LST range ~20-45C) can induce RMSE of 0.7-2.1C, increasing with resolution. This suggests higher resolution LST data is still valuable at all times and crucial under certain conditions. Errors in flux calculations at the different resolutions will be presented. * Kustas W.P., et al. 2003. Remote Sensing of Environment, 85, 429-440.