Data-Driven Agroclimate Modeling and Forecasting Based on Earth Observations and Predictions: A Study of Evapotranspiration, Precipitation, and Crop Yields

Abstract

Meeting the growing food demand in a sustainable manner is the main challenge faced by agriculture, at a time when the climate changes increasingly threaten food security. Innovative agro-climate approaches are needed to convert the growing amount of information emerging thanks to technological advances into products that help to adopt better decisions. Since water intervenes much of the climate change impacts on agriculture, accurately forecasting of precipitation and evapotranspiration is of the uppermost importance for minimizing the effect of adverse weather. The improvements of the numerical weather prediction (NWP) models and the statistical post-processing techniques provide unprecedented opportunities to better anticipate the changes in precipitation and evapotranspiration. The use of satellite remote sensing techniques for in-season forecasting of major crop yields over large areas is also of special interest, as it provides proxies of food security and food prices. While datasets from the moderate resolution imaging spectroradiometer (MODIS) are advantageous for crop forecasting, more research is needed on how factors such as the type of MODIS product, the statistical model or the training domain affect the crop yield forecasts. This study has been aimed to develop and evaluate new data-driven approaches for agro-climate forecasting, which combines NWP forecasts, remote sensing data, numerical modeling and machine learning techniques for improving crop water demand and crop yields forecasting in agricultural ecosystems. The manuscript is divided into six main chapters. In Chapter I, I provide a general introduction of the research. In Chapter II, single and multimodel ensemble forecasts of daily reference evapotranspiration, based on three leading NWP models over the continental U.S (CONUS), are produce and evaluated. The ability of three states of art probabilistic post-processing methods for improving NWP-based daily and weekly reference evapotranspiration forecasts over the CONUS in evaluated in Chapter III. In Chapter IV I evaluate leading NWP models and post-processing methods for improving ensemble precipitation forecasts over Brazil. In Chapter V it is evaluated a new optimization framework for the MODIS-based county and state-level corn yield forecasting over major producing states of the U.S. Finally, Chapter VI provides concluding remarks. The results represent a step forward in the efforts for improving precipitation, evapotranspiration and crop yield forecasting across multiple scales.