Impacts of climate change and variability on crop yields using emulators and empirical models

The thesis assesses impacts of climate change and variability on regional and global crop yields using econometric approaches to analyze global gridded data. Using a large dimension panel data of six Global Gridded Crop Models (GGCMs) for four rainfed crops (maize, rice, soybeans and wheat) an emulator suitable/amenable of being integrated into Integrated Assessment Models (IAMs) is built. The performance of the emulator is evaluated against observational-based, empirical models at regional scale by building a statistical model calibrated on historical observed crop yields data for United States (U.S.) counties. Chapter 1 provides the background of existing research methodologies in agronomic literature. The gaps in existing research and scope for research are laid down as motivation and objectives of the research that follows in the subsequent chapters. Chapter 2 discusses the data, methodology and framework used in the construction of a simple statistical emulator of the response of crops to weather shocks simulated by crop models. To facilitate the integration of the emulator into IAMs, the simplest model using a base specification of linear fixed effect with time trend interactions is developed. Chapter 3 investigates modifications to the base specification with a series of robustness checks exploring the suitability of an additional predictor variable, the stratification of coefficients geographically by groups of Agro-Ecological Zones (AEZs); and most importantly, the role of spatial dependence in variables by applying a spatial model. Chapter 4 compares the performance of the statistical emulator calibrated on crop model results, with an empirical models of crop responses based on historical data. The comparison focuses on U.S. counties. The base specification from Chapter 2 together with historical observed data from the U.S. Department of Agriculture (USDA), are utilized in an inter-comparison exercise for divergence in results and subsequent implications. Collectively, the three chapters (2-4) address several important questions: (1) what do reduced-form statistical response surfaces trained on crop model outputs from various simulation specifications look like; (2) do model-based crop response functions vary systematically over space (e.g., crop suitability zones) and across crop models?, (3) how do model-based crop response functions compare to crop responses estimated using historical observations? and (4) what are the implications for the characterization of future climate risks? Chapter 5 concludes the thesis providing a summary of key contributions and suggestions for future work.