🌊 From SAR Data to Actionable Maps: Building an Open-Source Flood Detection Pipeline with Python

Aditya Singh — Sat, 01 Aug 2026 07:20:38 +0000

Optical satellites are useless during floods because, well, it’s cloudy. Sentinel-1 SAR sees through rain and night, but its data is noisy, complex, and full of "permanent water" that isn't actually flooding.

I built an end-to-end pipeline to turn raw SNAP GeoTIFFs into clean, actionable flood maps. Here’s how I solved the three biggest headaches in SAR processing.

1. The "Padding" Trap

SNAP exports often include zero-value borders. If you don’t crop these, your histogram gets skewed by millions of 0s, breaking automatic thresholding.

The Fix: Treat 0.0 as NaN, convert to dB, and crop strictly to the valid data footprint before any analysis.

# Crop to valid data only
rows = np.where(~np.all(np.isnan(dB_full), axis=1))[0]
cols = np.where(~np.all(np.isnan(dB_full), axis=0))[0]
dB = dB_full[rows.min():rows.max()+1, cols.min():cols.max()+1]

2. Isolating New Floodwater

A simple -17 dB threshold detects all water. To find actual flood damage, you must subtract permanent rivers and lakes.

The Fix: I used OSMnx to fetch permanent water bodies from OpenStreetMap, rasterized them to match the SAR grid, and subtracted them from the detected mask.

# Fetch permanent water from OSM
osm_water = ox.features_from_bbox(bbox=bbox, tags={"natural": "water"})
# Subtract from detected mask
flood_only = mask_clean & (~binary_dilation(permanent_water, iterations=2))

3. The QGIS "Black Background" Bug

Exporting transparent GeoTIFFs is a nightmare. QGIS often renders "transparent" pixels as solid black if you rely on standard Alpha channels.

The Fix: Instead of RGBA, I use a single-band raster with a dedicated NoData value (255). QGIS treats NoData as transparent automatically.

# 1 = flooded, 255 = NoData (Transparent in QGIS)
overlay = np.where(final_flood_raster == 1, 1, 255).astype(np.uint8)

The Result

The pipeline outputs:

Vector Data: GeoJSON/Shapefiles with accurate area calculations (in km²).
Interactive Maps: A Folium HTML map with Satellite/Street toggles.
QGIS-Ready Rasters: Clean overlays that work immediately upon import.

🚀 Try It Out

The full notebook is open-source. It handles everything from speckle cleanup to final visualization.

🔗 [https://github.com/Adityas221b/QGIS-Flood]

Python #GeoAI #RemoteSensing #OpenSource #ClimateTech