Crop water productivity estimation with hyperspectral remote sensing
Crop water productivity (CWP) is the ratio of accumulated crop biomass or yield (Y) to the water utilized to produce it, which is typically estimated using transpiration (ETC). CWP is an important metric to test and monitor water-saving strategies in agroecosystems across the globe. Red and near-infrared broadbands have been used to estimate CWP, because they capture biophysical constraints based on crop-light interaction principles at pixel level (e.g., 30-meter resolution) over large areas through time. Hyperspectral remote sensing, which allows for the more precise measurement of crop-light interactions at higher spectral resolution, should in theory provide higher accuracy in CWP estimation but has been underutilized by the remote sensing community due to computational challenges and lack of availability. In this study, a simple methodology is presented to demonstrate how CWP could be estimated using hyperspectral remote sensing. Due to a lack of hyperspectral data, Landsat-7 Enhanced Thematic Mapper Plus (ETM+) data were used for the demonstration. Landsat is a broadband sensor that provides considerable spectral information for CWP estimation. New bands were identified in the workflow outside the typical Landsat bands used to estimate CWP and its components (Y and ETC). Landsat bands 1 and 3 were the most effective at estimating CWP and Y with an R2 of 0.72 (RMSE = 0.50 kg m−3) and 0.64 (RMSE = 0.31 kg m−2), respectively. All of the bands were poor at estimating ETC, with Landsat bands 1 and 7 being the most highly correlated (R2 = 0.13, RMSE = 0.08 m). Future work should train models with multiple estimates of CWP and Y over the growing season, while ETC may be better estimated with thermal infrared bands not considered in this study. Finally, studies should also consider estimating CWP categorically, instead of continuously, if the same objectives of testing and monitoring are met.
|Crop water productivity estimation with hyperspectral remote sensing
|Michael Marshall, Itiya Aneece, Daniel Foley, Cai Xueliang, Trent Biggs
|USGS Publications Warehouse
|Western Geographic Science Center