Correcting the Edge Effect for Sensor Spatial Response in Evapotranspiration Estimation through Remote Sensing

Abstract

Recent advances in remote-sensing technology using multispectral images provide a powerful tool to estimate crop-water evapotranspiration (ET) at the watershed scale with large spatial and temporal precision. However, most current sensors onboard operational satellites that capture radiances from the thermal infrared (TIR) spectral region are still constrained by limited spatial resolution, which creates a potential error called the edge effect. This consists of information intrusion from surrounding areas that crosses the boundary of the region of interest due to coarse pixel resolution. To reduce this error, a buffer zone is defined as a fixed distance from the field boundary, and the information contained in this zone is excluded from the sample. This study evaluates the optimal buffer distance needed to minimize the bias associated with the aggregated edge effect in annual ET estimates when using Landsat-7 enhanced thematic mapper plus (ETM+) data. Theoretical and statistical analyses indicate that a buffer of 2-TIR pixel equivalent distance ensures that the remaining field area contain radiances free of edge effects with 98.4% confidence. However, the analysis shows that in a circular turfgrass field of 50.9 ha, this process would eliminate half of the area. The analysis using annual ET data shows that applying a buffer distance of 0.75 TIR pixel is sufficient to mitigate the cumulative edge effect while preserving field representative valuable data. Although this analysis was based on Landsat-7 ETM+ images, the results could also be applied to annual ET maps with different thermal infrared pixel sizes.