Honors Capstone Report: Emulating the Global Change Analysis Model with an Expanded Input Space

Multi-sector dynamics (MSD) models are capable of predicting the evolution of highly complex human-earth interactions. The nuanced relationship between these interactions requires sophisticated modeling. This sophistication poses difficulty for researchers performing exploratory analysis, as it increases the time and computational power necessary to run the model. Previous work created a deep neural network emulator of one popular MSD model, with significantly reduced runtime. However, one of the limitations of the emulator is the limited scope of inputs that it accepts compared to the full MSD model. This paper describes work extending this previous emulator to accept an expanded input space – 233 inputs, as opposed to the original 12 inputs – as a step toward more comprehensive emulation of MSD models. In our work we demonstrate that our emulator is highly accurate, obtaining R2 metrics over 0.98 on a heldout test set, mirroring the successful results of past work despite the greater complexity of the emulation problem.