README.md

---
configs:
- config_name: labeled_all
  data_files: labeled_all/all.csv
  default: true
  description: Labeled spectra data with trait measurements for supervised learning.
- config_name: unlabeled
  data_files: unlabeled/*.csv
  description: Unlabeled spectra data for semi-supervised or self-supervised learning.
- config_name: labeled_splits
  data_files:
  - split: train
    path: labeled_splits/train.csv
  - split: test
    path: labeled_splits/test.csv
  description: A stratified splitting of the labeled data.
license: cc-by-nc-4.0
task_categories:
- feature-extraction
language:
- en
tags:
- hyperspectral
- plant-traits
- remote-sensing
- vegetation
- multi-regression
size_categories:
- 100K<n<1M
---
# 🌱 GreenHyperSpectra: A multi-source hyperspectral dataset for global vegetation trait prediction 🌱

GreenHySpectra is a collection of hyperspectral reflectance data of vegetation from different sources. It is intended for Regression machine learning task for plant trait prediction with self and semi-supervised learning.

<!-- ![DatasetIcon](./MultiSourceIcon.png) -->
<p align="center">
  <img src="./MultiSourceIcon.png" alt="DatasetIcon" width="40%"/>
</p>

# Spatial coverage
<!-- ![Dataset overview](./map_HF.png) -->
<p align="center">
  <img src="./map_HF.png" alt="Dataset overview" width="60%">
</p>

## 📁 Configurations
### 1. `GreenHyperSpectra: Unlabeled set`
- Files: all CSVs under `unlabeled/`
- Contains:
  - Sample ID
  - Spectral bands (400-2450 nm) >> 1721 bands

| Column         | Description                             |
|----------------|-----------------------------------------|
| 400            | Reflectance at 400nm                    |
| ...            | More spectral bands                     |
| 2450           | Reflectance at 2450nm                   |
<!-- --- -->
#### Check the data with Hugging Face datasets library
```
from datasets import load_dataset

### GreenHyperSpectra: unlabeled ###
ds_un = load_dataset("Avatarr05/GreenHyperSpectra", "unlabeled")
GreenHyperSpectra = ds_un['train'].to_pandas().drop(['Unnamed: 0'], axis=1)

display(GreenHyperSpectra.head())
```
---

### 2. `Labeled set`
- File: `labeled/all.csv`
- Contains:
  - Sample ID
  - Dataset ID
  - Spectral bands (400-2450 nm) >> 1721 bands
  - Trait measurements (e.g., leaf chlorophyll, nitrogen content etc.)

| Column         | Description                             |
|----------------|-----------------------------------------|
| dataset        | Reference to the source of the dataset  |
| 400            | Reflectance at 400nm                    |
| ...            | More spectral bands                     |
| 2450           | Reflectance at 2450nm                   |
| Cp             | Nitrogen content (g/cm²)                |
| Cm             | Leaf mass per area (g/cm²)              |
| Cw             | Leaf water content (cm)                 |
| LAI            | Leaf area index (m²/m²)                 |
| Cab            | Leaf chrolophyll content (µg/m²)        |
| Car            | Leaf carotenoids content (µg/m²)        |
| Anth           | Leaf anthocynins content (µg/m²)        |
| Cbc            | Carbon-based constituents (g/cm²)       |
<!-- --- -->
#### Check the data with Hugging Face datasets library
```
from datasets import load_dataset

### Labeled data: labeled_all ###
ds = load_dataset("Avatarr05/GreenHyperSpectra", "labeled_all")
df = ds['train'].to_pandas().drop(['Unnamed: 0'], axis=1)

display(df.head())
```
---
### 3. `Split labeled set`
- Files: all CSVs under `labeled_splits/`
These files follow the same format as the previous set but are pre-split for machine learning purposes. The split is stratified based on the dataset ID, with 20% of the data reserved for testing.

#### Check the data with Hugging Face datasets library
```
from datasets import load_dataset

### Labeled splits: labeled_splits ###
annotated_ds_train = load_dataset("Avatarr05/GreenHyperSpectra", 'labeled_splits', split="train")
annotated_ds_train = annotated_ds_train['train'].to_pandas().drop(['Unnamed: 0'], axis=1)

annotated_ds_test = load_dataset("Avatarr05/GreenHyperSpectra", 'labeled_splits', split="test")
annotated_ds_test = annotated_ds_test['train'].to_pandas().drop(['Unnamed: 0'], axis=1)

display(annotated_ds_train.head())
display(annotated_ds_test.head())
```
> ⚠️ **Note:** Due to the high dimensionality of spectral datasets—often containing hundreds or thousands of columns—**Hugging Face Data Studio may not render these files properly**. This is a known limitation, as the Studio interface is not optimized for wide tabular data.  
>
> To work effectively with this dataset, we recommend using the **Hugging Face `datasets` library** or the **MLCroissant Python library** for programmatic access and exploration.

## Citation
If you use the **GreenHyperSpectra** dataset, please cite the following paper:

```bibtex
@article{cherif2025greenhyperspectra,
  title={GreenHyperSpectra: A multi-source hyperspectral dataset for global vegetation trait prediction},
  author={Cherif, Eya and Ouaknine, Arthur and Brown, Luke A and Dao, Phuong D and Kovach, Kyle R and Lu, Bing and Mederer, Daniel and Feilhauer, Hannes and Kattenborn, Teja and Rolnick, David},
  journal={arXiv preprint arXiv:2507.06806},
  year={2025}
}
```

If you use the labeled data included in this repository, please also cite the following study for more details about the compiled datasets:

```bibtex
@article{cherif2023spectra,
  title={From spectra to plant functional traits: Transferable multi-trait models from heterogeneous and sparse data},
  author={Cherif, Eya and Feilhauer, Hannes and Berger, Katja and Dao, Phuong D and Ewald, Michael and Hank, Tobias B and He, Yuhong and Kovach, Kyle R and Lu, Bing and Townsend, Philip A and others},
  journal={Remote Sensing of Environment},
  volume={292},
  pages={113580},
  year={2023},
  publisher={Elsevier}
}
```
license: cc-by-nc-4.0
---