A Machine Learning Approach to Long-Term Drought Prediction using Normalized Difference Indices Computed on a Spatiotemporal Dataset

Earlier research has shown that the Normalized Difference Drought Index(NDDI), combining information from both NDVI and NDMI, can be an accurate earlyindicator of drought conditions. NDDI is computed with information fromvisible, near-infrared, and short-wave infrared channels, and demonstratesincreased sensitivity as a drought indicator than other indices. In this work,we aim to determine whether NDDI can serve as an early indicator of drought ordramatic environmental change, by computing NDDI using data from landscapesaround bodies of water in Europe, which are not as drought-prone as the centralUS grasslands where NDDI was initially evaluated on. We use the datasetSEN2DWATER (SEN2DWATER: A Novel Multitemporal Dataset and Deep LearningBenchmark For Water Resources Analysis), a 2-Dimensional spatiotemporal datasetcreated from multispectral Sentinel-2 data collected over water bodies fromJuly 2016 to December 2022. SEN2DWATER contains data from all 13 bands ofSentinel-2, making it a suitable dataset for our research. We leverage twoCNNs, each learning trends in NDVI and NDMI values respectively using timeseries of images obtained from the SEN2DWATER dataset. By using the CNNsoutputs, the predicted NDVI and NDMI values, we propose to compute a predictedNDDI, with the goal of investigating its accuracy. Preliminary results showthat NDDI can be effectively forecasted with good accuracy by using ML methods,and the SEND2DWATER dataset could allow to calculate NDDI as a useful methodfor predicting climate and ecological change. Moreover, such predictions couldbe highly useful also in mitigating, or even preventing, any harmful effects ofclimate and ecological change, by supporting policy decisions.