Satellite-Ground Reconciliation & Crop Field Verification System for Maharashtra
"A satellite sees, The System suggests, An officer confirms"
Around 55–60% of Maharashtra's geography consists of farmland, roughly 30 million football fields. Surveying this land is often a manual and time-consuming process. Today, the responsibility of verifying what's growing on those fields is on the farmers who need to self-declare what they are growing. This still doesn't offer statewide coverage.
This leads to challenges in crop verification, planning, subsidy distribution, insurance verification, and disaster management and assessment.
Without Bhu-Drushti
A Krishi Sahayak in Baramati taluka, post-Kharif, October
Result: physically impossible to complete verification within the seasonal window.
With Bhu-Drushti
Same Krishi Sahayak, same October morning
Entire Circle verified within the window. She is no longer the bottleneck.
Satellite crop data from ESA Sentinel-2 (free, 10m resolution, every 5 days) already covers every plot in Maharashtra. Bhu-Drushti is the missing reconciliation layer.
None of which exist in the workflow today.
Satellite imagery to plot boundaries. Same code processes 10 prototype plots or all 2.3 crore Maharashtra plots without change. Cloud cover > 40%: auto-flagged for mandatory Krishi Sahayak visit.
Range matching (Phase 1), pixel-count area check (Phase 2), and DTW time-series classification (>84% accuracy for Indian Kharif crops).
Offline first, full field day with zero internet. Works on budget Android phones. React Native with expo-sqlite.
Progressive Web App on any modern browser, no install. Role-based access per officer scope. Module based activation for different features.
Working prototype, live demo using real Sentinel-2 data, Malegaon Khurd village, Baramati taluka, Kharif 2024.
GEE Python pipeline live. All 10 plots returned NDVI 0.64–0.79, consistent with healthy Kharif vegetation. Plot boundaries digitised from Mahabhunaksha cadastral data.
NDVI range matching produces 0–100 scores. Demo output: 6 auto-confirmed, 3 flagged, 1 escalated. DTW time-series classification staged for Phase 2.
Risk-sorted queue of 4 plots, offline-first SQLite, GPS accuracy gate ≤ 15m, mandatory geo-tagged photo before verdict. Demo Mode at hackathon venue.
Leaflet map with plot polygons coloured by satellite crop classification. Risk summary panel. NDVI disaster change detection demonstrated with simulated pre/post event data.
Node.js (Express) + PostGIS on Docker. Three REST endpoints: plot queue, verdict upload, NDVI delta heatmap. Zero government API dependency in Phase 1.
10 plots, Malegaon Khurd. Signal 1 risk scoring with real data. Self-contained, no govt. API needed.
Full taluka, DILRMP cadastral MoU, Signal 2 & 3, Mode A sowing capture.
All 36 districts, 358 talukas, 2.3 crore plots, both seasons. AgriStack API integration.
Technology Advisors: Anand Hatwalne · Anand Sathe