FireCastRL Spatiotemporal Wildfire Dataset
收藏资源简介:
该数据集由纽约州立大学布法罗分校团队构建,是一个大规模时空野火预测数据集,整合了IRWIN野火事件数据库和GRIDMET气象数据,包含950万条样本,覆盖2014-2025年美国本土的50,720起野火事件及合成的76,080个负样本。数据经过去重、空间平衡和75天时间窗口特征工程处理,包含21项气象与火灾指数特征,如降水量、风速、燃烧指数等。其创新性在于首次将预测性环境变量与强化学习模拟环境相结合,为AI驱动的野火主动防控系统提供训练基础,可支持从早期预警到战术调度的全链条研究。
This dataset was developed by a research team from the University at Buffalo, The State University of New York. It is a large-scale spatiotemporal wildfire prediction dataset that integrates the IRWIN Wildfire Incident Database and GRIDMET meteorological data. The dataset contains 9.5 million samples, covering 50,720 wildfire incidents and 76,080 synthesized negative samples across the contiguous United States from 2014 to 2025. The data has undergone deduplication, spatial balancing, and feature engineering with a 75-day time window, and includes 21 meteorological and fire index features such as precipitation, wind speed, and burning index. Its innovation lies in the first combination of predictive environmental variables and reinforcement learning simulation environments, which provides a training foundation for AI-driven proactive wildfire prevention and control systems, and supports full-chain research ranging from early warning to tactical scheduling.




