Spatial Estimation of Wheat Yields from Landsat’s Visible, near Infrared and Thermal Reflectance Bands

Abstract

At a field level spatial crop yield patterns are mainly determined by spatially changing soil properties (e.g. soil moisture) in interaction with seasonal climate conditions and weather patterns at critical crop growth stages in the crop development. In this research we combined remote sensing technologies, local weather and canopy condition to describe spatial yield patterns of a 1400ha wheat field during the 2011 winter cropping season. More specifically, we determined the ability of remotely sensed derived indices, within the visible and thermal domains, to predict final harvested wheat yield at field scale. Weather and crop variables were continuously monitored by installing three automatic weather stations in a transect covering different soil types. Weather variables included rainfall, minimum and maximum temperatures and relative humidity, while crop canopy temperature was also measured. Satellite imagery Landsat TM 5 and 7 was obtained at five different stages in the crop cycle. Weather variables and crop characteristics were used to calculate a crop water stress index (CSIws) at the location of each weather station. Field data was used to validate a crop stress index from satellite imagery. Yield data was acquired from the combine harvester at different locations in the field. We used visible and near-infrared bands to calculate the enhanced vegetation index (EVI). Thermal bands and EVI were used to derive a crop stress indices (CSIsat) as well as a moisture stress index (MSIsat), based on Moran’s trapezoid approach, at several times during the crop growth period. Weather station data were used to ground truth the satellite derived indices. Results showed that spatial variations in crop yield were related to a satellite derived canopy stress index (CSIsat) and a moisture stress index (MSIsat). At field level the canopy stress index (CSIws) calculated at midday was correlated to the CSIsat at late morning close to the time of Landsat satellite pass-over. Harvested yield was moderately correlated (R2 = 0.67) to CSIsat for a fix date across all fields. This relationship noticeably improved (adjusted R2 = 0.95), using both indices from all five dates across all fields during the crop growth period. Here we showed that satellite derived crop attributes (CSIsat and MSIsat) can account for most of the variability in actual crop yield and that they could be used to predict aggregated field scale wheat yields (deviation of +2.6%). The application and value of such an approach to the grains industry is also discussed.