GeoClass: Browser-Based Land Cover Analysis
Problem
Traditional satellite land-use and land-cover (LULC) classification requires heavy desktop software (QGIS, ArcGIS, ENVI), gigabytes of raw raster downloads, and complex cloud-masking scripts. Environmental researchers face steep friction when attempting rapid assessments of regional environmental change.
Approach
Built a lightweight, browser-first geospatial platform that streams Google Earth Engine computation on demand. Geologists and environmental scientists can draw custom vector AOI boundaries directly onto an interactive Leaflet map, query multi-temporal Sentinel-2 imagery, and execute Deep Learning Spatial U-Net (GeoAI) and Dynamic World land cover models with automated cloud thresholding.
Tech Stack
Results
Eliminated desktop software dependencies by enabling instant, in-browser classification across four primary land-cover categories. Users can inspect temporal land transitions, evaluate vegetation health (NDVI), and download vector boundaries with zero local setup.
What I Learned
Engineered responsive dual-layer map rendering connecting Leaflet with Earth Engine tile endpoints; optimized asynchronous REST API communication between Next.js and FastAPI to stream geospatial compute without blocking client UI.
Enjoyed this piece?
Leave a reaction to let the author know what resonated with you.