Neural Networks for Models of Forward and Inverse Problems in Fire Spread

This dataset contains fire spread simulations generated with a physics-driven and data-driven model. The data-driven model is trained with a TensorFlow applied to data generated with the physics-driven model described in the paper titled "A simple model for wildland fire vortex–sink interactions". The data are separated into- A forward problem (forward_problem directory) that predicts fire spread behavior based on fuel and atmospheric properties.- An inverse problem (inverse_problem directory) that predicts fuel and atmospheric properties based on fire spread behavior.Three different networks (CNN, Unet, and FC-Unet) are considered for the forward problem, and one network (CNN-FC) is considered for the inverse problem.This work has been published: Tong, X. & Quaife, B. (2025). Data-driven fire modeling: Learning first arrival times and model parameters with neural networks. Environmental Modelling & Software, 183, 106253.These codes and data are identical to https://github.com/quaife/ML_FAT_data (commit 297de500888b11170df4fec7ba9b79d1be923ecd)