This project aims at using Deep Learning methods to replace Shallow Water Equation(SWE) solvers in flood prediction tasks to boost the prediction speed.
The codes here are made up of three main parts: Simulation of flood with SWE solver, training of models, model evaluation. Training data and test data are stored in (https://drive.google.com/drive/folders/1YoUBOwVOBgkOKDQSRtwJjAlvZYn99jT6?usp=sharing).
The simulation of flood depends on the Shallow Water Equation Solver-Anuga Hydro(https://github.com/GeoscienceAustralia/anuga_core). As ANUGA is developed in Python 2 environment, the simulation part is written in Python 2.7. The Deep Learning architecture and training is precessed with Pytorch. Thus pytorch is needed to run the codes.
In floodsimulation file, you can find the code used for the simulation of floods. In models, the architectures of the models mentioned in the paper are presented. In trainingcode, peole can find the training details for the architectures.Trained models, training data and test data and the covariance matrix used for the posterior adjustment are stored in Google Drive(https://drive.google.com/drive/folders/1YoUBOwVOBgkOKDQSRtwJjAlvZYn99jT6?usp=sharing). To evaluate the results, model_test.ipynb amd model_posterior_update.ipynb can be applied.
If you find this helpful, please kindly cite the paper:
MIT