Granite Geospatial
IBMGranite Geospatial is IBM's family of Earth-observation models built on a Swin-B backbone pretrained with masked image modeling over Harmonized Landsat and Sentinel-2 imagery. The declared biomass member replaces the pretraining decoder with a UPerNet regression head and is fine-tuned against GEDI L4A lidar labels drawn from 15 biomes, predicting above-ground biomass from HLS L30 optical composites through TerraTorch.
The companion GitHub repository is archived and carries getting-started notebooks rather than training code, so the checkpoint and its TerraTorch configuration are the whole distributed surface.
Openness
3 high confidence- license
- Apache-2.0(OSI)
- weights
- open(Apache-2.0 ungated checkpoint files)
- data
- documented-not-released(HLS L30 and GEDI L4A sources and the leaf-on compositing method described, assembled training set not published)
- code
- partial(getting-started and few-shot notebooks over TerraTorch, no pretraining or fine-tuning pipeline)
The checkpoint and its configuration are published under Apache-2.0 and the card names its NASA source products in detail, but the model repository is archived notebooks and the assembled corpus never shipped.
- https://huggingface.co/api/models/ibm-granite/granite-geospatial-biomass recorded 2026-10-21
gated: false; private: false; license apache-2.0; biomass_model.ckpt, biomass_model_lightning.ckpt, and config.yaml present.
- https://huggingface.co/ibm-granite/granite-geospatial-biomass/raw/main/README.md recorded 2026-10-21
Apache-2.0 card stating that the weights and the configuration file that defines them are made available, naming HLS L30 and GEDI L4A as the training data with the cloud-free compositing method, and pointing to getting-started notebooks rather than training code.
- https://raw.githubusercontent.com/ibm-granite/granite-geospatial-biomass/main/README.md recorded 2026-10-21
Describes the repository as public notebooks and utilities for working with the fine-tuned model, with no training or fine-tuning pipeline.
Adoption
1 high confidenceDownloads of the declared biomass checkpoint on Hugging Face are the adoption signal here, the model's only counted distribution channel.
- https://huggingface.co/api/models/ibm-granite/granite-geospatial-biomass recorded 2026-09-10
downloads: 127; gated: false; private: false.
Capability
2 medium confidenceGranite Geospatial documents one capability: a TerraTorch fine-tuning workflow, demonstrated in zero- and few-shot notebooks. Its only input is HLS L30, the Landsat-only half of the Harmonized Landsat/Sentinel-2 product; GEDI lidar is a training label rather than a model input, the pretraining objective is masked reconstruction rather than generation, and the release covers one spatial scale.
- https://huggingface.co/ibm-granite/granite-geospatial-biomass/raw/main/README.md recorded 2026-09-10
Documents TerraTorch inference and zero-shot/few-shot fine-tuning notebooks, with HLS L30 optical imagery as the model input and above-ground biomass regression as the output; the README's own Training Data link, however, points to the S30 product page.
Verified 2026-09-10