Project

Adaptation of Geospatial Foundation Models (GFMs) for Finer Scale Soil Moisture Variability

Objective

Soil moisture is a critical parameter that influences agricultural practices, climate conditions, hydrology, and various other phenomena. It also plays a significant role in nature-based carbon sequestration. However, different applications require soil moisture estimation at various spatio-temporal scales. As we move from coarser to finer spatial resolutions, physics-based models tend to become increasingly demanding on computational resources and are often limited by input variables. Therefore, as an alternative, we propose using a Geospatial Foundation Model (GFM) for finer spatial scale soil moisture estimation. Recently, geospatial foundation models have gained significant attention due to their adaptability in various Earth observation tasks related to classifications and segmentations. However, it remains to be seen how these models perform in estimating land surface variables such as soil moisture. In this project, we aim to achieve the following objectives: (1) Establish a regression framework suitable for any regression-specific task, serving as a validation step before adopting it for the actual use case. (2) Curate data for fine-tuning the GFM model for soil moisture estimation. (3) Fine-tune the model and validate its performance against a baseline. (4) Report our findings and prepare a paper for submission to arXiv.

Outcome

A research paper.