Explainable deep learning for wildfire danger estimation

<p>Deep learning models have been remarkably successful in a number of different fields, yet their application to disaster management is obstructed by the lack of transparency and trust which characterises artificial neural networks. This is particularly relevant in the field of Earth sciences where fitting is only a tiny part of the problem, and process understanding becomes more relevant [1,2]. In this regard, plenty of eXplainable Artificial Intelligence (XAI) algorithms have been proposed in the literature over the past few years [3]. We suggest that combining saliency maps with interpretable approximations, such as LIME, is useful to extract complementary insights and reach robust explanations. We address the problem of wildfire forecasting for which interpreting the model's predictions is of crucial importance to put into action effective mitigation strategies. Daily risk maps have been obtained by training a convolutional LSTM with ten years of data of spatio-temporal features, including weather variables, remote sensing indices and static layers for land characteristics [4]. We show how the usage of XAI allows us to interpret the predicted fire danger, thereby shortening the gap between black-box approaches and disaster management.</p><p>&#160;</p><p>[1] <strong>Deep learning for the Earth Sciences: A comprehensive approach to remote sensing, climate science and geosciences</strong></p><p>Gustau Camps-Valls, Devis Tuia, Xiao Xiang Zhu, Markus Reichstein (Editors)</p><p>Wiley \& Sons 2021</p><p>[2] <strong>Deep learning and process understanding for data-driven Earth System Science</strong></p><p>Reichstein, M. and Camps-Valls, G. and Stevens, B. and Denzler, J. and Carvalhais, N. and Jung, M. and Prabhat</p><p><em>Nature </em>566 :195-204, 2019</p><p>[3] <strong>Explainable AI: Interpreting, Explaining and Visualizing Deep Learning</strong></p><p><strong>&#160;</strong>Wojciech Samek, Gr&#233;goire Montavon, Andrea Vedaldi, Lars Kai Hansen, Klaus-Robert M&#252;ller (Editors)</p><p>LNCS, volume 11700, Springer&#160;</p><p>[4] <strong>Deep Learning Methods for Daily Wildfire Danger Forecasting</strong></p><p>Ioannis Prapas, Spyros Kondylatos, Ioannis Papoutsis, Gustau Camps-Valls, Michele Ronco, Miguel-&#193;ngel Fern&#225;ndez-Torres, Maria Piles Guillem, Nuno Carvalhais</p><p>arXiv: 2111.02736</p><p><br>&#160;</p>