Hi everyone!
I’m Soham Roy, a CS student from India with a background in ML, geospatial analysis, and full-stack development.
I’ve been exploring two ideas for GSoC 2027:
Idea #2 — Improve GRASS in Jupyter Notebook
I spend a lot of time in Jupyter working with geospatial and climate data, and I’ve felt firsthand how rough the visualization experience can be. I’d love to improve legend support, symbology handling, and matplotlib integration in grass.jupyter. I’m currently reading through the codebase and plan to start with writing pytest tests for the Map and InteractiveMap classes.
Idea #9 — ML-based Urban Heat Island Module for GRASS
I’ve been building ShadowMap — a geo-ML system that predicts urban heat island intensity across Delhi using land surface temperature, NDVI, and impervious surface rasters. It uses gradient boosting with spatial cross-validation (R² = 0.906). The whole time I was building it, I kept thinking: “this should just be a GRASS module.” So that’s what I want to propose — something like r.uhi that lets anyone run this kind of analysis natively inside GRASS without duct-taping Python scripts together.
I’ve cloned the repo and am getting my dev environment up. Would love any pointers on where to start contributing and whether the r.uhi direction makes sense to the community.
GitHub: sohamroy06 (soham roy) · GitHub
Email: sohamroy20sr@gmail.com