AI Flood Detection & Water Inundation Extent
Reproducible satellite change detection algorithms, backscatter histograms, and georeferenced boundary polygons.
Calculated via geodesic polygon projection in WGS84 EPSG:4326 scaled to 26.3°N latitude. Represents 34% of monitored lowland territory.
Dual-polarization (VV+VH) radar intensity thresholded at -14.5 dB with 5x5 Lee Sigma filtering to eliminate speckle artifacts while preserving riverine char boundaries.
Pre-flood baseline scene from 2022-05-28 subtracted from 2022-06-20 inundation to prevent normal riverbeds from inflating emergency deluge figures.
Remote Sensing Methodology & Signal Physics
Why Optical Satellites Fail in Monsoon Floods
Traditional optical imaging (e.g. Landsat, Sentinel-2) relies on visible and near-infrared reflectance. During catastrophic monsoon events like the Assam Brahmaputra floods, cloud cover consistently exceeds 90-95%, blinding optical sensors completely.
FloodGuard AI solves this critical bottleneck by prioritizing Synthetic Aperture Radar (SAR), which operates at C-Band (5.405 GHz) microwave frequencies that travel unimpeded through clouds, torrential rain, and darkness.
Physics of Radar Specular Reflection
Rough dry terrain and vegetation backscatter microwave radar pulses in multiple directions, returning a strong signal (-6 dB to -10 dB) to the satellite sensor.
In contrast, smooth standing water acts like a mirror, reflecting incoming radar beams away from the satellite path. This creates a steep drop in backscatter intensity (< -14.5 dB), rendering flooded expanses jet black in radar imagery and enabling deterministic detection.