Geospatial deep learning

Multi-scale model for spatial biodiversity estimation.

A minimal JAX library for graph-based connectivity analysis at scale.

This project develops a graph neural network approach to accelerate ecological connectivity analysis by mapping fine-scale landscape features onto lower-dimensional graph representations, overcoming the computational bottlenecks of traditional resistance distance calculations. The research aims to enable high-resolution gene flow predictions across fragmented landscapes while maintaining computational tractability, with potential applications to real-world conservation planning for identifying critical ecological corridors and restoration priorities.