Emulators for dynamic vegetation models - Supervised learning in large data sets

Abstract

The observed and expected changes in the environment due to human actions implies risks that future food production will be insufficient. Pre- dicting the impact these changes have on the agricultural system could be beneficial by allowing for proactive mitigating efforts. The prediction of these impacts often involve large computer programs that simulate the behavior of the environment. By implementing a statistical representation of the simulator, called an emulator, our hope is that these predictions could be obtain at a lower computational cost. This master thesis has implemented and evaluated a Gaussian process emulator for a vegeta- tion model that is used for predicting the annual production of spring wheat based on climate data at different locations around the world. The problem of accurately modeling the simulator using a Gaussian process approach was split into two parts. The first part was to model the average yield at each location given average climate input at that location. The second part was to model the yield at a specific year for a location given the average yield at that location and the climate input anomalies dur- ing that year. The results was far from satisfactory and a more complex approach is probably needed before the emulator can be of any practical use. Based on our findings, possible extensions that might improve results are discussed.

Abstract from the paper’s OpenAlex record, shown for indexing and research. Copyright remains with the authors or the publisher. Read at the source ↗

The text the model read, with the source record’s inline markup (italics, sub- and superscripts) rendered and its character entities decoded; quote offsets refer to the raw text.