Physics-aware machine learning for data-driven fire risk prediction

2019/20澳大利亚火灾季节的范围、影响和消防机构的反应都是前所未有的。在这个项目中，我们的目标是回答这样一个问题：这些火灾的规模是由已知的火灾驱动因素(干旱、天气、燃料和点火)驱动的，还是涉及根本上新的未描述的过程和现象？我们将通过开发一种创新的、物理感知的火灾风险和蔓延的机器学习模型来实现这一点，并在20年的卫星火灾记录上进行训练和验证。该模型的预测能力将在2019/20年火灾季节进行测试，以确定是否可以确定新的火灾驱动因素，并且该模型本身将被操作为一种新的中短期火灾风险预测工具。

The 2019/20 Australian fire season was unprecedented in its extent, impact, and the response of fire agencies. In this project, we aim to answer the question: was the scale of these fires driven by known drivers of fire (drought, weather, fuels and ignitions), or were fundamentally new undescribed processes and phenomena involved? We will accomplish this by developing an innovative, physics-aware machine learning model of fire risk and spread, trained and validated on a two-decade satellite fire record. The predictive ability of the model will be tested on the 2019/20 fire season to determine if novel drivers of fire can be identified, and the model itself will be operationalised into a novel short-to-mid term fire risk prediction tool.