Establish whether selected geospatial foundation model parameters are optimal

Establish whether the parameter choices used for the geospatial foundation models and their predictive analyses provide optimal performance for population-health applications.

Background

The study evaluates multiple geospatial foundation model families and makes specific choices about model versions, embedding dimensions, and analysis settings. Because these models provide many researcher degrees of freedom, the authors state that their design choices are defensible and reproducible but do not establish that they are optimal. Determining the best parameters and configurations therefore remains unresolved.

References

Second, GFMs offer more researcher degrees of freedom than LLMs or other machine learning-based methods. This is a major challenge for benchmarking these models and means that we cannot confirm that the parameters chosen provide optimal performance.

— Geospatial Foundation Models Capture Health-Relevant Dimensions of Place Beyond Conventional Social Risk Indices  (2609.11689 - Hendrix et al., 10 Sep 2026) in Discussion, paragraph beginning “This study has several limitations”