Drought has serious impacts on health and the economy in the Western U.S., but it’s difficult to predict weeks or months in advance. This study tests a new machine learning approach called graph reservoir computing (Graph RC) to improve drought forecasting. We compare it to ConvLSTM, another widely used model type, using historical weather and climate data. We test whether adding more variables, using higher-resolution data, and including 3-D atmospheric structure can improve predictions. We expect the graph-based model with more detailed input data to perform best and help advance drought forecasting.