Rarell/global-drought-prediction-with-credit

Code to process data to train a machine learning DroughtFormer, built in the CREDIT framework, to predict drought related variables (ML model is trained globally, but there is a regional focus on Africa). Also contains code to analyze DroughtFormer predictions. # DroughtFormer: A global CrossFormer model for drought prediction developed in the CREDIT framework This project is for the development of DroughtFormer, a machine learning model focused on predicting global drought, with emphasis on Africa, and developed within the Community Research Earth Digitial Information Twin (CREDIT) framework and Python environment. Specifically, this repository contains the scripts used to collect, process, and orgnize the datasets for DroughtFormer into yearly .zarr formats. The scripts folder contains code used to postprocess forecasts made by the most recently trained version of DroughtFormer. Figures made from the postprocessing, as well as the weights and learning history of the most recent version are found in results. Lastly, customized versions of the CREDIT code, from testing different parameters, are also found in modified_credit_files.