Drought Identification in NLDAS Data using Machine Learning Methods

Though machine learning (ML) methods have been around for decades, they have only more recently been adopted in the geosciences. The availability of existing long data records combined with the capability of ML algorithms to learn highly non-linear relationships between data sources means there is even more potential for the replacement or augmentation of existing scientific analyses with ML methods. Here, I give an example of how I used a convolutional neural network (CNN) for the task of pixelwise classification of the North American Land Data Assimilation System (NLDAS) Total Water Storage data into their corresponding drought levels based on the Palmer Drought Severity Index (PDSI). Promising results indicate there is much to be explored in the application of ML to drought identification and monitoring.