INSAIT-Institute/FireScope-Bench

---
license: cc-by-4.0
pretty_name: FireScope-Bench
tags:
- wildfire
- remote-sensing
- geospatial
- reasoning
- CoT
task_categories:
- image-text-to-image
- image-to-image
---

# FireScope-Bench dataset

This dataset accompanies the paper:

**FireScope: Wildfire Risk Prediction with a Chain-of-Thought Oracle**\
Accepted to **CVPR 2026**\
Paper: https://arxiv.org/abs/2511.17171 \
GitHub: https://github.com/insait-institute/FireScope

## File and Folder Description

### `climate_data.json`

Contains climate data for various coordinates.\
There is a correspondence between coordinates encoded in tile/raster
filenames and dictionary keys in this file.

------------------------------------------------------------------------

### `usa/wildfire_rasters`

#### `images/`

Sentinel-2 satellite images corresponding to wildfire risk rasters.
Images are fused in **2×2 tiles**, resulting in **\~2K resolution
images**.

#### `rasters/`

Wildfire risk rasters from the **Wildfire Risk to Communities**
project.\
Rasters are fused in **2×2 tiles**, resulting in **682×682 resolution
rasters**.

#### `small_sample/`

Small sample subset used for the majority of the **FireScope
experiments**.

- `images/` -- Sentinel-2 images for the small sample
- `rasters/` -- wildfire risk rasters for the small sample

------------------------------------------------------------------------

### `usa/wildfire_events`

#### `images/`

Sentinel-2 images for wildfire events, organized in subdirectories by
**year**.

Images are **pre-event** images.\
Therefore, images located in folder **Y** correspond to wildfire events
in **year Y+1** and are named accordingly to match the corresponding
event mask.

#### `masks/`

Wildfire event masks from the **MTBS (Monitoring Trends in Burn
Severity)** dataset.

#### `controls/`

Masks for control areas (zero-masks).

#### `controls_images/`

Sentinel-2 images corresponding to the control areas.

------------------------------------------------------------------------

### `europe/wildfire_events`

#### `images/`

Sentinel-2 images for wildfire events, organized in subdirectories by
**year**.

Images are **pre-event** images.\
Therefore, images located in folder **Y** correspond to wildfire events
in **year Y+1** and are named accordingly to match the corresponding
event mask.

#### `masks/`

Wildfire event masks from the **EFFIS Burnt Areas dataset**.

#### `controls/`

Masks for control areas (zero-masks).

#### `controls_images/`

Sentinel-2 images corresponding to the control areas.

------------------------------------------------------------------------

## Licensing

This dataset aggregates data from multiple sources and is released under
**Creative Commons Attribution 4.0 (CC-BY-4.0)**.

Underlying data sources include:

- **Sentinel-2 satellite imagery** (Copernicus programme)
- **Wildfire Risk to Communities** rasters
- **MTBS (Monitoring Trends in Burn Severity)** wildfire event data
- **EFFIS Burnt Areas dataset**
- **NASA POWER climate data**

Users should also comply with the licensing and attribution requirements
of the original data providers.

------------------------------------------------------------------------

## Citation

If you find this work useful in your research, please cite:

``` bibtex
@article{markov2025firescope,
title={FireScope: Wildfire Risk Prediction with a Chain-of-Thought Oracle},
author={Markov, Mario and Ailuro, Stefan Maria and Van Gool, Luc and Schindler, Konrad and Paudel, Danda Pani},
journal={arXiv preprint arXiv:2511.17171},
year={2025}
}
```

------------------------------------------------------------------------

许可证: cc-by-4.0
友好名称: FireScope-Bench
标签:
- 野火
- 遥感
- 地理空间
- 推理
- 思维链（Chain-of-Thought，CoT）
任务类别:
- 图像-文本到图像
- 图像到图像

# FireScope-Bench 数据集

本数据集配套论文：
**《FireScope：基于思维链神谕的野火风险预测》**
已被**CVPR 2026**收录
论文链接：https://arxiv.org/abs/2511.17171

## 文件与文件夹说明

### `climate_data.json`

包含各类坐标对应的气候数据。
瓦片/栅格文件名中编码的坐标与该文件中的字典键一一对应。

------------------------------------------------------------------------

### `usa/wildfire_rasters`

#### `images/`

与野火风险栅格对应的Sentinel-2卫星影像。
影像以**2×2瓦片**拼接而成，分辨率约为**2K**。

#### `rasters/`

源自**Wildfire Risk to Communities**项目的野火风险栅格。
栅格以**2×2瓦片**拼接而成，分辨率为**682×682**。

#### `small_sample/`

用于绝大多数**FireScope实验**的小子样本子集。
- `images/`：小子样本对应的Sentinel-2卫星影像
- `rasters/`：小子样本对应的野火风险栅格

------------------------------------------------------------------------

### `usa/wildfire_events`

#### `images/`

野火事件对应的Sentinel-2卫星影像，按**年份**分为子文件夹。
影像为**事件前**影像。
因此，**Y**文件夹中的影像对应**Y+1**年发生的野火事件，并按照匹配对应事件掩码的规则命名。

#### `masks/`

源自**MTBS（火灾严重性监测趋势，Monitoring Trends in Burn Severity）**数据集的野火事件掩码。

#### `controls/`

对照区域掩码（零掩码）。

#### `controls_images/`

与对照区域对应的Sentinel-2卫星影像。

------------------------------------------------------------------------

### `europe/wildfire_events`

#### `images/`

野火事件对应的Sentinel-2卫星影像，按**年份**分为子文件夹。
影像为**事件前**影像。
因此，**Y**文件夹中的影像对应**Y+1**年发生的野火事件，并按照匹配对应事件掩码的规则命名。

#### `masks/`

源自**EFFIS火灾过火区域数据集**的野火事件掩码。

#### `controls/`

对照区域掩码（零掩码）。

#### `controls_images/`

与对照区域对应的Sentinel-2卫星影像。

------------------------------------------------------------------------

## 许可协议

本数据集整合了多源数据，并以**知识共享署名4.0（CC-BY-4.0）**协议发布。
其底层数据源包括：
- **Sentinel-2卫星影像**（哥白尼计划）
- **Wildfire Risk to Communities**项目栅格数据
- **MTBS（火灾严重性监测趋势，Monitoring Trends in Burn Severity）**野火事件数据集
- **EFFIS火灾过火区域数据集**
- **NASA POWER气候数据**

用户还需遵守原始数据提供商的许可与署名要求。

------------------------------------------------------------------------

## 引用方式

若您的研究中使用了本数据集，请引用如下文献：
bibtex
@article{markov2025firescope,
title={FireScope: Wildfire Risk Prediction with a Chain-of-Thought Oracle},
author={Markov, Mario and Ailuro, Stefan Maria and Van Gool, Luc and Schindler, Konrad and Paudel, Danda Pani},
journal={arXiv preprint arXiv:2511.17171},
year={2025}
}