Why this matters
Accurate maps of rivers, lakes, reservoirs and coastal waters are essential for flood risk assessment, urban planning and climate studies, but traditional surveys are slow and sparse. This project explored how far we can push satellite imagery and simple algorithms to automatically pick out water bodies from complex urban scenes.
What I worked on
As part of a graduate remote-sensing course at IIT Bombay, I worked with Landsat 8 imagery over the Mumbai region to extract surface water from high-resolution multispectral scenes. The aim was to separate water from land, built-up areas and shadows using only information available in the optical bands.
How we detected water
I preprocessed the geoTIFF scenes in QGIS, then used MATLAB to convert digital numbers to radiance and top-of-atmosphere reflectance for key bands. On top of this, I implemented several spectral indices: the Normalized Difference Water Index (NDWI and MNDWI), Normalized Difference Vegetation Index (NDVI), and Automated Water Extraction Index with and without shadow terms (AWEIsh, AWEInsh). By analysing histograms for the near-infrared and shortwave-infrared bands and tuning index thresholds, I generated binary water masks that cleanly highlighted reservoirs, rivers and coastline while suppressing dark non-water pixels. The work gave me a practical feel for multispectral image physics, index design and the nuts and bolts of building end-to-end remote-sensing workflows in code.