INSAIT-Institute/FireScope-Bench

FireScope-Bench数据集是一个用于野火风险预测和遥感分析的基准数据集，主要包含美国（USA）和欧洲（Europe）的野火相关数据。数据集分为两部分：1) wildfire_rasters（野火风险栅格）部分，包括Sentinel-2卫星图像（以2×2瓦片融合，分辨率约2K）和对应的野火风险栅格（来自Wildfire Risk to Communities项目，以2×2瓦片融合，分辨率为682×682），以及一个小样本子集用于FireScope实验；2) wildfire_events（野火事件）部分，包括野火事件前的Sentinel-2图像（按年份组织，对应下一年的事件）、野火事件掩码（美国数据来自MTBS，欧洲数据来自EFFIS Burnt Areas数据集）、控制区域掩码和对应图像。此外，数据集还提供气候数据（climate_data.json），对应坐标点的气候信息。数据集支持图像-文本到图像、图像到图像等任务，旨在通过链式思维（CoT）推理进行野火风险栅格预测，基于多源数据（如Sentinel-2、NASA POWER等）构建，遵循CC-BY-4.0许可。The FireScope-Bench dataset is a benchmark dataset for wildfire risk prediction and remote sensing analysis, primarily containing wildfire-related data for the USA and Europe. The dataset consists of two main parts: 1) the wildfire_rasters section, which includes Sentinel-2 satellite images (fused in 2×2 tiles with ~2K resolution) and corresponding wildfire risk rasters (from the Wildfire Risk to Communities project, fused in 2×2 tiles with 682×682 resolution), along with a small sample subset used for FireScope experiments; 2) the wildfire_events section, which includes pre-event Sentinel-2 images (organized by year, corresponding to events in the next year), wildfire event masks (from MTBS for the USA and EFFIS Burnt Areas dataset for Europe), control area masks, and corresponding images. Additionally, the dataset provides climate data (climate_data.json) with climate information for various coordinates. It supports tasks such as image-text-to-image and image-to-image, aiming to predict wildfire risk rasters through chain-of-thought (CoT) reasoning, and is built from multiple sources (e.g., Sentinel-2, NASA POWER) under the CC-BY-4.0 license.