# BioMassters: Forest Aboveground Biomass from Sentinel-1 and Sentinel-2 Time Series

## Facts

- Provider: Nascetti et al. (BioMassters); re-packaged by IBM-NASA Geospatial and Clark University for Prithvi
- License: CC-BY-4.0 (https://proceedings.neurips.cc/paper_files/paper/2023/file/40daf2a00278c4bea1b26cd4c8a654f8-Paper-Datasets_and_Benchmarks.pdf)
- Size: 27 files
- Format: PNG, CSV (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/ibm-nasa-geospatial/BioMassters
- Access: public, readable without an account
- Category: Earth Observation & Geospatial
- Source: https://proceedings.neurips.cc/paper_files/paper/2023/file/40daf2a00278c4bea1b26cd4c8a654f8-Paper-Datasets_and_Benchmarks.pdf
- Page: https://scigantic.com/archive/4ddb2e42-a07b-5c80-8c8b-9f50a61fe148
- Keywords: Earth Observation & Geospatial, satellite imagery, LiDAR, time series, PNG, CSV, Hugging Face

## About

Sentinel-1 synthetic aperture radar (SAR) and Sentinel-2 multispectral (MSI) imagery of Finnish forests for 2016 to 2021, paired with 11,462 reference maps of aboveground biomass (AGB) derived from airborne LiDAR calibrated with in-situ forest inventory. Each reference patch covers 2,560 by 2,560 m at 256 by 256 pixels and comes with one image per month for the 12 months before the AGB observation: Sentinel-1 for every month, Sentinel-2 (10 bands plus cloud probability) where available. This copy keeps the original BioMassters (NeurIPS 2023 Datasets and Benchmarks) layout and adds biomassters_chip_tracker.csv with cloud_percentage, corrupt_values, red_mean and an 80/20 train/validation split, used to train the Prithvi geospatial foundation model. Listing the root (measured 2026-11-13) shows train_features as a 14-part split zip (thirteen 10.7 GB .z01 to .z13 parts plus train_features.zip), test_features_splits as a 5-part split zip, train_agbm.zip (0.58 GB), test_agbm.tar (0.73 GB) and feature and AGB metadata CSVs; about 201 GB in all. Rejoin the split zips before extracting, then read the image chips with rasterio or xarray. Released under CC-BY-4.0; cite the BioMassters paper.

## Load the data

```python
import pandas as pd
df = pd.read_csv("hf://datasets/ibm-nasa-geospatial/BioMassters/The_BioMassters_-_features_metadata.csv.csv")
df.head()
```

## Data structure

File types in sampled files: PNG, CSV

Example file: `The_BioMassters_-_features_metadata.csv.csv`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/4ddb2e42-a07b-5c80-8c8b-9f50a61fe148
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/4ddb2e42-a07b-5c80-8c8b-9f50a61fe148
- Over MCP: `get_schema_card` with id `4ddb2e42-a07b-5c80-8c8b-9f50a61fe148` at https://api.scigantic.com/mcp

## Related topics

- [Earth Observation & Geospatial datasets](https://scigantic.com/topics/earth-observation-and-geospatial)
- [CSV datasets](https://scigantic.com/topics/csv)
- [Satellite imagery datasets](https://scigantic.com/topics/satellite-imagery)
- [Time series datasets](https://scigantic.com/topics/time-series)
- [PNG datasets](https://scigantic.com/topics/png)
- [LiDAR datasets](https://scigantic.com/topics/lidar)
