# StefaLand: An Efficient Geoscience Foundation Model That Improves Dynamic Land-Surface Predictions

arXiv: https://arxiv.org/abs/2509.17942 · HTML: https://mlsanity.com/paper/2509.17942/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 9 of its 82 references that are in the map.
- First posted on arXiv 22 Sep 2025.

## Builds on (most cited first, 9 of 9)

- [Reformer: The Efficient Transformer](https://mlsanity.com/paper/2001.04451/) (2020)
- [Are Transformers Effective for Time Series Forecasting?](https://mlsanity.com/paper/2205.13504/) (2022)
- [Recognition in Terra Incognita](https://mlsanity.com/paper/1807.04975/) (2018)
- [AlphaEarth Foundations: An embedding field model for accurate and efficient global mapping from sparse label data](https://mlsanity.com/paper/2507.22291/) (2025)
- [Galileo: Learning Global & Local Features of Many Remote Sensing Modalities](https://mlsanity.com/paper/2502.09356/) (2025)
- [When Geoscience Meets Foundation Models: Towards General Geoscience Artificial Intelligence System](https://mlsanity.com/paper/2309.06799/) (2023)
- [Geospatial foundation models for image analysis: evaluating and enhancing NASA-IBM Prithvi's domain adaptability](https://mlsanity.com/paper/2409.00489/) (2024)
- [Prediction in ungauged regions with sparse flow duration curves and input-selection ensemble modeling](https://mlsanity.com/paper/2011.13380/) (2020)
- [Probing the limit of hydrologic predictability with the Transformer network](https://mlsanity.com/paper/2306.12384/) (2023)

## Built on by (most cited first, 1 of 1)

- [Earth Observation Foundation Models for Terrestrial Ecohydrology: From Representation Learning to Process Inference](https://mlsanity.com/paper/2608.15282/) (2026)

## Abstract (arXiv)

Managing natural resources and mitigating risks from floods, droughts, wildfires, and landslides require models that can accurately predict climate-driven land-surface responses. Traditional models often struggle with spatial generalization because they are trained or calibrated on limited observations and can degrade under concept drift. Recently proposed vision foundation models trained on satellite imagery demand massive compute, and they are not designed for dynamic land surface prediction tasks. We introduce StefaLand, a generative spatiotemporal Earth representation learning model centered on learning cross-domain interactions to suppress overfitting. StefaLand demonstrates especially strong spatial generalization on five datasets across four important tasks: streamflow, soil moisture, soil composition and landslides, compared to previous state-of-the-art methods. The domain-inspired design choices include a location-aware masked autoencoder that fuses static and time-series inputs, an attribute-based rather than image-based representation that drastically reduces compute demands, and residual fine-tuning adapters that strengthen knowledge transfer across tasks. StefaLand can be pretrained and finetuned on commonly available academic compute resources, yet consistently outperforms state-of-the-art supervised learning baselines, fine-tuned vision foundation models and commercially available embeddings, highlighting the previously overlooked value of cross-domain interactions and providing assistance to data-poor regions of the world.