可解释的基于补丁的深度学习用于从集合模拟预测野火蔓延
Interpretable Patch-Based Deep Learning for Wildfire Spread Prediction from Ensemble Simulations
这个论文研究用深度学习模型预测野火蔓延，比较了四种模型，发现地表燃料负荷这个变量很重要，能降低21%的预测误差，还分析了模型如何学习，挺有意思的。
这项研究比较了四种深度学习架构（包括一个基于补丁的U-Net、一个迁移学习的ResNet-50、一个受风驱动的对流方程约束的物理信息网络和一个Swin-Unet转换器）在西班牙加泰罗尼亚的Rectoret地区，使用10,584个2米分辨率的火灾蔓延模拟数据集进行训练。
研究发现，在地形和植被变量中，只有地表燃料负荷与燃烧概率有较强的相关性（r=0.27），包含它可以将预测误差降低21%。
卷积模型主要依赖距离当前火前线的距离，而Swin-Unet则更侧重于燃料和地形。
Interpretable Patch-Based Deep Learning for Wildfire Spread Prediction from Ensemble Simulations
Wildfire spread is traditionally predicted using physics-based simulators, which are physically interpretable but whose cost increases with each additional ensemble member. We ask how well deep learning surrogates can reproduce these simulations at a fraction of this cost, training them on 10,584 fire spread simulations at 2m resolution for the Rectoret region in Catalonia, Spain. Four architectures are compared: a patch-based U-Net, a transfer-learned ResNet-50, a physics-informed network constrained by the wind-driven advection equation and a Swin-Unet transformer. Among the terrain and vegetation variables, only surface fuel load predicts burn probability with any strength (r = 0.27) and including it lowers prediction error by 21%. The remaining variables correlate weakly and are highly duplicative. Next, an experiment with saliency, occlusion and rotation demonstrates the models' learning. Convolutional models rely primarily on distance from the current fire front, while Swin-Unet assigns more weight to fuel and terrain, a finding also noted in an unrelated wildfire dataset. When applied without retraining to the second region, Pedriza, all three convolutional models still predict fire spread, losing accuracy by a small but systematic margin.
本条情报归属 AITOP「论文」核心栏目，重点聚焦欧洲人工智能治理与技术生态。在跨区域产业观察坐标中，该动态直接关联「AI 开发者工具链、Coding Agent 与工程化实践」研究主线，反映出当前技术演进与商业落地的最新风向。涉及核心实体与技术议题包括：可解释、深度学习、野火预测、物理信息网络，读者可通过下方标签进入对应专题页查阅历史脉络与关联对比。
信源核验说明：本文原始线索采编自「arXiv cs.AI」（@Marcin Lawenda, Aleksandra Krasicka, David Caballero, Luis Torres, Łukasz Szustak），经 AITOP 多维评分管线结构化提炼。
