Seeing is believing I: The use of thermal sensing from satellite imagery to predict crop yield

Abstract

Volatility in crop production has been part of the Australian environment since cropping began with the arrival of the first European settlers. Climate variability is the main factor affecting crop production at national, state and local scales. At field level spatial patterns on yield production are also determined by spatially changing soil properties in interaction with seasonal climate conditions and weather patterns at critical stages in the crop development. Here we used a combination of field level weather records, canopy characteristics, and satellite information to determine the spatial performance of a large field of wheat. The main objective of this research is to determine the ability of remote sensing technologies to capture yield losses due to water stress at the canopy level. The yield, canopy characteristics (i.e. canopy temperature and ground cover) and seasonal conditions of a field of wheat (~1400ha) (-29.402° South and 149.508°, New South Wales, Australia) were continuously monitored during the winter of 2011. Weather and crop variables were continuously monitored by installing three automatic weather stations in a transect covering different positions and soils in the landscape. Weather variables included rainfall, minimum and maximum temperatures and relative humidity, and crop characteristics included ground cover and canopy temperature. Satellite imagery Landsat TM 5 and 7 was collected at five different stages in the crop cycle. Weather variables and crop characteristics were used to calculate a crop stress index (CSI) at point and field scale (39 fields). Field data was used to validate a spatial satellite image derived index. Spatial yield data was downloaded from the harvester at the different locations in the field. We used the thermal band (land surface temperature, LST) and enhanced vegetation index (EVI) bands from the MODIS (250 m for visible bands and 1km for thermal band) and a derived EVI from Landsat TM 7 (25 m for visible and 90m for thermal) satellite platforms. Results showed that spatial variations in crop yield were related to a satellite derived canopy stress index (CSIsat) and a moisture stress index (MSIsat). A weather station level canopy stress index (CSIws) calculated at midday was correlated to the CSIsat at late morning. In addition, a strong linear relationship was observed between EVI and LST at point scale throughout the crop growth period. Differences were smallest at anthesis when the canopy closure was highest. This suggests that LST imagery data around flowering could be used to calculate crop stress over large areas of the crop. The harvested yield was related (R2 = 0.67) to CSIsat using a fix date across all fields. This relationship improved (R2 = 0.92) using both indices from all five dates across all fields during the crop growth period. Here we successfully showed that satellite derived crop attributes (CSIsat and MSIsat) can account for most of the variability in final crop yield and that they can be used to predict crop yield at field scales. Applications of these results could enhance the ability of producers to hedge their financial on -farm crop production losses due to in-season water stress by taking crop insurance. This is likely to further improve their adaptive capacity and thus strengthening the long-term viability of the industry domestically and elsewhere.