Flooding is one of the most destructive and costly natural disasters, and climatechanges would further increase risks globally. This work presents a novel mul-timodal machine learning approach for multi-year global flood risk prediction,combining geographical information and historical natural disaster dataset. Ourmultimodal framework employs state-of-the-art processing techniques to extractembeddings from each data modality, including text-based geographical data andtabular-based time-series data. Experiments demonstrate that a multimodal ap-proach, that is combining text and statistical data, outperforms a single-modalityapproach. Our most advanced architecture, employing embeddings extracted us-ing transfer learning upon DistilBert model, achieves 75%-77% ROCAUC scorein predicting the next 1-5 year flooding event in historically flooded locations.This work demonstrates the potentials of using machine learning for long-termplanning in natural disaster management