Earth Observation Based Canadian Crop Yield Forecasting -- Impact of Spatial Modeling Scale

Abstract:

Earth Observation (EO) based yield modelling has long been in development as an alternative method to the traditional survey based methods in forecasting the regional and global crop yield. However, it is only in last decade or so, with availability of high quality regional EO data in near real time (NRT), EO-based crop yield forecasting has become practical enough to be applied towards operational crop yield reporting. The Canadian Crop Yield Forecaster (CCYF) is one of such modelling tool that designed to provide regional and national crop yield outlooks during and shortly after the growing season. The CCYF integrates climate, remote sensing and other earth observation information (e.g., historical yields, soil and crop maps) using a physical based soil moisture budget model and a statistical based yield forecasting model. One of the major challenges for CCYF and many other EO-based crop yield forecasting systems is to determine a proper spatial modelling scale that could be easily aggregated to various required yield reporting units, yet still retain the statistical sensitivity of crop yield to variations in climate, soil and remote sensing vegetation indices. In this study, we have compared yield modelling using CCYF at three different administrative scales, i.e. township, Census Agricultural Regions (CARs) and province for four crops (spring wheat, canola, corn and soybeans) in the agricultural regions of Manitoba, Canada. Due to the shorter available historical yield records at the township scale, different modelling scheme is applied for township scale modelling compared to the other two larger scales. The modelling at provincial scale did not capture the yield variability, while the modelling at CAR level provided reasonable results for some CARs while failed for others. The modelling at township scale captured most of the yield variability, yet its performance and implementation is restricted by the availability of the yield data at this scale.