Combatting global floods demands robust predictive models, and this GitHub repository addresses the challenge using Python and deep learning. By harnessing Tropical Rainfall Measuring Mission (TRMM) and Dartmouth Flood Observatory (DFO) data, the project pioneers a novel image-based approach for global flood prediction.
Various deep learning models, including Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and Convolutional Recurrent Neural Networks (ConvRNN), are implemented to forecast flood images from downscaled rainfall data. Notably, Convolutional RNN outperforms other models, achieving an average precision of 0.9252.
Python plays a pivotal role in processing and modeling TRMM and DFO data, demonstrating the feasibility of this innovative technique for global flood prediction.
- trmm-deep-learning.ipynb: Python notebook (Kaggle-based) to predict DFO flood images from TRMM rainfall images using deep learning methods.
- trmm-images.ipynb: Python notebook (Kaggle-based) to process TRMM and DFO data into images suitable for the neural networks.
- Dissertation.pdf: Full project report containing details on the entire research process and results.
Convolutional RNN stands out as the most effective model, achieving an impressive average precision of 0.9252.
Future work could extend the model by incorporating additional variables and leveraging powerful computational systems to enhance spatiotemporal accuracy in predicting global rainfall and flooding.
Feel free to explore the repository, delve into the code, and contribute to advancing the field of global flood prediction with deep learning!