Empirical Validation: How Satellite Radar (Sentinel-1 SAR) Tracked the 2022 Nagaon & 2024 Lakhimpur Floods in Assam

In our previous article, we explored the microwave physics of why optical satellite data fails during overcast seasons. But how does this radar pipeline actually perform against recorded climate disasters on the ground?

To test this empirically, we deployed a multi-season Sentinel-1 dual-orbit radar pipeline across two of the most flood-vulnerable agricultural corridors in Assam: the Nagaon Agricultural Basin (2022) and the Lakhimpur Central Paddy Belt (2024). The empirical data reveals clear, objective microwave scattering signatures that track field submergence, plant survival, and post-flood vegetative recovery.

---

1. Detection Framework: Dual-Orbit SAR & Cropland Masking

A major challenge in satellite analytics is spatial aggregation bias: averaging raw radar backscatter across a broad district polygon inadvertently mixes settlements, tree canopies, and tea estates with rice paddies, artificially inflating backscatter values. To eliminate this noise, our workflow enforces two critical constraints:

• ESA WorldCover 10m Cropland Masking: Pixel filtering isolates strictly Class 40 (Cropland), ensuring radar statistics represent active agricultural fields only.
• Dual-Orbit Fusion: Combining ascending (evening) and descending (morning) passes provides high-frequency observations (every 6 to 12 days), closing mid-monsoon data gaps.
• Physical Metrics Tracked: Co-polarization (σ⁰_VV) for surface specular water reflection, cross-polarization (σ⁰_VH) for canopy volume scattering, and the standardized biomass anomaly (Z_SAR) against a 4-year rolling baseline.

---

2. Empirical Case Studies: Ground-Truth Disaster Verification

The dual-panel trajectory plot below demonstrates the empirical radar response across both historical extreme events:

---

3. Multi-Season Quantitative Comparison

---

4. Reproducible Earth Engine Pipeline Snippet

To extract dual-orbit Sentinel-1 backscatter masked to ESA WorldCover cropland in Python:

import ee
import pandas as pd

ee.Initialize(project='your-cloud-project-id')

# 1. Define Cropland Mask (Class 40)
worldcover = ee.ImageCollection("ESA/WorldCover/v100").first()
cropland_mask = worldcover.select('Map').eq(40)

# 2. Query Sentinel-1 GRD Collection across both orbits
s1 = (ee.ImageCollection('COPERNICUS/S1_GRD')
      .filterBounds(aoi_geometry)
      .filterDate('2024-06-01', '2024-11-15')
      .filter(ee.Filter.eq('instrumentMode', 'IW'))
      .filter(ee.Filter.listContains('transmitterReceiverPolarisation', 'VV'))
      .filter(ee.Filter.listContains('transmitterReceiverPolarisation', 'VH')))

def extract_cropland_stats(img):
    masked = img.updateMask(cropland_mask)
    vv = masked.select('VV').focal_mean(1.5, 'circle', 'pixels')
    vh = masked.select('VH').focal_mean(1.5, 'circle', 'pixels')
    
    stats = ee.Image([vv, vh]).reduceRegion(
        reducer=ee.Reducer.mean(),
        geometry=aoi_geometry,
        scale=30,
        maxPixels=1e9
    )
    return ee.Feature(None, {
        'date': img.date().format('YYYY-MM-dd'),
        'VV': stats.get('VV'),
        'VH': stats.get('VH')
    })

raw_data = s1.map(extract_cropland_stats).getInfo()['features']
df = pd.DataFrame([f['properties'] for f in raw_data if f['properties']['VV'] is not None])
print(f"Ingested {len(df)} cloud-free cropland observations.")

---