README.md

---
dataset_info:
  features:
  - name: id
    dtype: string
  - name: department_name
    dtype: string
  - name: chm
    dtype:
      array2_d:
        shape:
        - 256
        - 256
        dtype: float16
  - name: no_data_percentage
    dtype: float32
  - name: crs
    dtype: string
  - name: transform
    dtype: string
  - name: bounds
    dtype: string
  - name: resolution
    dtype: float32
  - name: chm_mean_year
    dtype: int16
  - name: rgbnir_ndvi_1
    dtype:
      array3_d:
        shape:
        - 5
        - 256
        - 256
        dtype: uint8
  - name: rgbnir_year_1
    dtype: uint16
  - name: rgbnir_ndvi_2
    dtype:
      array3_d:
        shape:
        - 5
        - 256
        - 256
        dtype: uint8
  - name: rgbnir_year_2
    dtype: uint16
  - name: rgbnir_ndvi_3
    dtype:
      array3_d:
        shape:
        - 5
        - 256
        - 256
        dtype: uint8
  - name: rgbnir_year_3
    dtype: uint16
  splits:
  - name: train
    num_bytes: 880058054404
    num_examples: 785392
  download_size: 730412322573
  dataset_size: 880058054404
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
license: apache-2.0
pretty_name: PrediTree
tags:
- remote-sensing
- multi-temporal
- multi-spectral
- canopy-height-prediction
- 3-pg
- infrared
- rgb
- model
---

# 🌳 PrediTree: A Multi-Temporal Multi-Spectral Sub-Meter Canopy Height Maps Dataset 

[![Dataset](https://img.shields.io/badge/🤗-Dataset-blue.svg)](https://huggingface.co/datasets/hiyam-d/vhr_canopy_height_allier_50cm_small)
[![Paper](https://img.shields.io/badge/📄-Paper-green.svg)](https://arxiv.org/)
[![License](https://img.shields.io/badge/License-Apache--2.0-yellow.svg)](https://www.apache.org/licenses/LICENSE-2.0)

![Sample Panels](./sample.png)

## 📖 Overview
**PrediTree** is a large-scale **multi-temporal, multi-spectral canopy height dataset** designed for 🌍 **remote sensing, forestry monitoring, and environmental analysis**.  
All imagery and canopy height products are **spatially aligned** at **0.5 m resolution**, enabling fine-grained tree growth prediction and ecological studies.

---

## ✨ Key Highlights
- 📊 **Multi-Temporal**: 3 yearly acquisitions (RGB + NIR + NDVI)  
- 🌈 **Multi-Spectral**: High-resolution optical imagery including RGB, NIR, and derived NDVI  
- 🌲 **Canopy Height Models (CHM)**: LiDAR-based data  
- 📏 **Resolution**: 0.5 m 
- 🌍 **Coverage**: France-wide dataset with departmental splits  
- 📦 **Scale**: 785k training patches, ~880 GB of data  

---

## 📂 Dataset Structure
Each sample contains:
| Column | Description |
|--------|-------------|
| `chm` | 🌲 Canopy Height Model (m) |
| `rgbnir_ndvi_[1-3]` | 📸 RGB + NIR + NDVI imagery for three years (5 bands, 256×256) |
| `rgbnir_year_[1-3]` | 📅 Acquisition year for imagery |
| `chm_mean_year` | 🏞️ Average canopy height across years |
| `no_data_percentage` | ❌ % missing pixels |
| `crs`, `transform`, `bounds`, `resolution` | 🗺️ Geospatial metadata |

---

## 📊 Dataset Specs
```yaml
splits:
  train:
    num_examples: 785,392
    256_256px_subtile_examples: 3,141,568
    size: 880 GB
resolution: 0.5 m
dataset_size: 880 GB
license: apache-2.0
```

---

## 🔬 Scientific Context
PrediTree is the **first CHM dataset to offer multi-temporal sub-meter CHM-aligned imagery specifically designed for training and evaluating tree height prediction models**.

![Comparison with Existing Datasets](./comparison.png)

---

## 📜 Citation
If you use this dataset, please cite:

```bibtex
@inproceedings{debary2025preditree,
  title={PrediTree: A Multi-Temporal Sub-meter Dataset of Multi-Spectral Imagery Aligned With Canopy Height Maps},
  author={Debary, Hiyam and Fiaz, Mustansar and Klein, Levente},
  booktitle={GAIA},
  year={2025},
  url={https://huggingface.co/datasets/hiyam-d/PrediTree}
}
```

---

## 🔖 Tags
`remote-sensing` · `multi-temporal` · `multi-spectral` · `canopy-height-prediction` · `infrared` · `rgb` · `model`