Application of Geospatial Techniques for visualization and analysis to interpret Geo-Coded lmages to enhance watershed Development Outcomes.

1. Executive Snapshot

Summary

The Department of Land Resources (DoLR), Ministry of Rural Development, oversees the Watershed Development Component of Pradhan Mantri Krishi Sinchayee Yojana (WDC-PMKSY 2.0) [KNOWN]. While millions of field-level geo-tagged photographs are collected via mobile applications like DRISHTI, they are currently treated primarily as static administrative proof rather than dynamic spatial intelligence [PS]. This dossier proposes GeoSrishti-AI, an integrated geospatial visualization and computer vision analytics platform designed to bridge ground-truth geo-coded imagery with 30m multi-spectral satellite observations (SRISHTI platform / Sentinel-2 / Landsat) [INFERENCE]. By combining spatial verification algorithms, automated land-use/cover change detection (NDVI, NDWI, SAVI), hydrological terrain analysis (DEM, slope, drainage density), and lightweight computer vision (YOLOv8 asset detection), the system transforms raw ground photos into verified, actionable watershed intelligence for planners, district officers, and national administrators [INFERENCE].

Metadata Table

Fundamental System Objective

To build a scalable, web-based GIS and AI analytics platform that ingests ground-level geo-coded photographs, correlates them with multi-temporal 30m satellite data and hydrological models, and automatically generates thematic maps, change analytics, and verified intervention impact reports for WDC-PMKSY watershed management [INFERENCE].

2. Problem Deconstruction

The problem description provided by DoLR explicitly lays out seven expected solution deliverables (labeled 'a' through 'g') [PS]. Below is the atomic extraction and implementation breakdown:

Explicit Requirement Extraction Table

3. Stakeholders & User Needs

4. Requirements Traceability

5. Constraints & Environment Realities

1. Connectivity & Network Dead Zones: Watershed intervention sites are located in remote rural, semi-arid, and hilly terrains where 4G/5G signal is frequently absent [KNOWN]. Mobile apps must store photo EXIF data and geospatial coordinates locally and sync lazily when connected [INFERENCE].
2. Satellite Resolution Limits (30m Spatial Resolution): The problem statement explicitly anchors analysis to 30m spatial resolution satellite data (e.g., Landsat-8/9, Sentinel-2 resampled/downscaled, or ResourceSat-2 AWiFS) [PS]. Small watershed structures (e.g., a 5-meter wide check dam or farm pond) cannot be directly resolved in a single 30m pixel [INFERENCE]. The system must analyze local contextual signatures (e.g., surrounding NDWI wetness bloom, localized vegetation index uptick around the structure site) rather than expecting crisp pixel boundary detection of small concrete walls [INFERENCE].
3. Device Constraints of Field Workers: Field workers utilize low-to-mid tier Android smartphones (2GB-4GB RAM) [KNOWN]. Heavily complex AI inference or GIS rendering must be offloaded to cloud backends or executed via lightweight mobile quantized models (TensorFlow Lite / ONNX) [INFERENCE].
4. Geographic Diversity: India encompasses diverse hydro-geological zones (Deccan trap, Indo-Gangetic alluvium, Himalayan slopes, Rajasthan arid basins) [KNOWN]. Thematic map generation must accept regionally calibrated threshold parameters for soil and vegetation indices [INFERENCE].
5. Data Standards Compliance: Must comply with NDIS / NNRMS standards and ISRO Bhuvan spatial reference systems (EPSG:4326 / EPSG:3857) [RESEARCHED].

6. Research Questions

1. How do ISRO's Bhuvan Srishti and Drishti applications format geo-tagged photo metadata and spatial WMS layers [RESEARCHED]?
2. Can a 30m multi-spectral pixel adequately reflect environmental changes caused by micro-watershed structures like gully plugs or boulder bunds [INFERENCE]?
3. What open-source satellite datasets best simulate or complement the 30m SRISHTI platform data for hackathon demonstration [INFERENCE]?
4. How can computer vision on field photos detect structural integrity or water storage level, and how can this be spatially cross-checked against digital elevation models (DEM) [INFERENCE]?
5. What spatial tolerance (buffer distance in meters) should be applied when validating whether a field photo location corresponds to a hydrologically valid stream order [INFERENCE]?

7. Data Requirements & Sources

8. Existing Solutions & Gap Analysis

9. Candidate Solution Approaches

Approach 1: Classical GIS Overlay & Rule-Based Heuristics (Traditional)

• Mechanism: Ingests geo-tagged photos into PostGIS. Generates standard buffer zones (e.g., 50m radius). Computes mean Sentinel-2 NDVI and NDWI within the buffer before and after the photo upload date using GDAL scripts.
• Complexity: Low.
• Explainability: 100% transparent math.
• Cost: Extremely low.
• Limitations: Fails to analyze the actual image contents (a photo of a rock could be uploaded at a check dam site and passed as valid).

Approach 2: Deep Learning Image Vision + Cloud Satellite Processing (CV-Heavy)

• Mechanism: Uses a fine-tuned Vision Transformer (ViT) or YOLOv8 model trained on field assets to classify photos (Check Dam, Farm Pond, Gully Plug, Plantation, Terrace). Uses Google Earth Engine API to run cloud-based spatio-temporal change detection around verified GPS locations.
• Complexity: Medium-High.
• Explainability: High for vision (bounding boxes), Moderate for cloud ML models.
• Cost: Low to Moderate.
• Limitations: Dependent on GEE API rate limits during live presentation if cloud connection stutters.

Approach 3: Hybrid Geo-AI Spatial Validation Engine (Recommended)

• Mechanism:
  1. Ground Tier: Lightweight YOLOv8 model detects asset types and condition flags in field photos.
  2. Spatial Verification Tier: PostGIS checks whether the photo's GPS coordinates and compass azimuth fall within a hydrologically valid zone (e.g., check dam within 30m of derived D8 stream network; farm pond within low-slope depression).
  3. Satellite Analytics Tier: Cloud-Optimized GeoTIFF (COG) pipeline computes 30m multi-spectral indices (NDVI, NDWI, SAVI, NDMI) across pre/post intervention dates.
  4. Integrated Scoring: Combines vision confidence + spatial validity score + satellite index delta into a single Watershed Health & Verification Index (WHVI).
• Complexity: Moderate-High.
• Explainability: High (decomposed into visual score, spatial score, and spectral delta score).
• Cost: Low (open-source, self-hosted stack).
• Hackathon Feasibility: High (can run locally with pre-indexed COG tiles and FastAPI backend).
• Production Feasibility: High (scales horizontally via Docker/Kubernetes).

10. Recommended Approach & Why

Selection

Approach 3: Hybrid Geo-AI Spatial Validation Engine is selected [INFERENCE].

Justification

1. Direct Problem Alignment: The PS explicitly calls for interpreting geo-coded images in relation to watershed characteristics and satellite-derived information [PS]. Pure GIS overlay (Approach 1) ignores photo content, while pure CV (Approach 2) ignores terrain hydrology rules [INFERENCE].
2. Robustness Against Fraud/Errors: Field geo-tags can suffer from GPS drift or deliberate spoofing. By requiring the photo content (YOLOv8) to match the spatial hydrological location (D8 Stream Network) and satellite spectral response (NDWI/NDVI delta), the system ensures multi-tier validation [INFERENCE].
3. Hackathon Demonstrability: A student team can easily run YOLOv8 inference locally on sample photos, compute PostGIS spatial queries instantly, and render satellite change maps via GeoServer/Leaflet in real time without external API latency [INFERENCE].

Why Alternatives Were Rejected

• Approach 1 Rejected: Does not perform true "interpretation of geo-coded images" as demanded by Requirement (b) [PS].
• Approach 2 Rejected: Over-relies on continuous cloud GEE API access, introducing external failure risks during hackathon judging and lacking strict hydrological spatial rules [INFERENCE].

11. Technical Architecture

+-----------------------------------------------------------------------------------+
|                                 USER INTERFACE LAYER                              |
|   +------------------------------------+     +--------------------------------+   |
|   | Field Worker PWA / Mobile App      |     | Web GIS Admin Dashboard        |   |
|   | (Offline Capture, GPS, Exif, Sync) |     | (React.js + Leaflet/OpenLayers) |   |
|   +-----------------+------------------+     +---------------+----------------+   |
+---------------------|----------------------------------------|--------------------+
                      | HTTPS / REST / OGC                     |
+---------------------v----------------------------------------v--------------------+
|                                  API GATEWAY (FastAPI)                            |
|                  Authentication (JWT) | Rate Limiting | CORS Router               |
+-------------------------------------+---------------------------------------------+
                                      |
+-------------------------------------v---------------------------------------------+
|                             BACKEND MICROSERVICES LAYER                           |
|  +------------------------+  +-----------------------+  +-----------------------+ |
|  | Geo-Photo AI Service   |  | Spatial Analysis      |  | Satellite Index       | |
|  | (YOLOv8 Asset Classifier|  | Engine (PostGIS /     |  | Pipeline (GDAL /      | |
|  |  & Structural Health)  |  |  Network / DEM D8)    |  |  Rasterio / Xarray)   | |
|  +-----------+------------+  +-----------+-----------+  +-----------+-----------+ |
+--------------|---------------------------|--------------------------|-------------+
               |                           |                          |
+--------------v---------------------------v--------------------------v-------------+
|                                  DATA STORAGE LAYER                               |
|  +------------------------------------+  +-------------------------------------+  |
|  | PostgreSQL 16 + PostGIS 3.4        |  | GeoServer / COG Tile Server         |  |
|  | (Vectors, Photos Metadata, Scores) |  | (WMS / WFS / WMTS Tile Delivery)    |  |
|  +------------------------------------+  +-------------------------------------+  |
|  +------------------------------------------------------------------------------+  |
|  | Object Storage (MinIO / S3) - Geo-tagged JPEGs, COG Sat Rasters, DEM GeoTIFF|  |
|  +------------------------------------------------------------------------------+  |
+-----------------------------------------------------------------------------------+

Step-by-Step Data Flow Narrative

1. Photo Ingestion: A field officer captures an asset photo using the app. The app attaches GPS metadata (latitude, longitude, elevation, timestamp, compass bearing) and pushes the JSON + JPEG payload to FastAPI backend [INFERENCE].
2. AI Vision Classification: The Geo-Photo AI Service runs YOLOv8 on the uploaded photo to identify structure type (e.g., check_dam, farm_pond, plantation) and structural state (intact, silted, dry, overflowing) with confidence scores [INFERENCE].
3. Spatial Hydrological Validation: The Spatial Analysis Engine queries PostGIS to cross-reference the photo coordinates with DEM-derived drainage layers. It computes:
   • Distance to nearest D8 stream line ($D_{stream}$)
   • Terrain slope angle ($\theta$)
   • Micro-watershed catchment ID [INFERENCE].
4. Satellite Index Delta Extraction: The Satellite Index Pipeline streams multi-temporal 30m Cloud-Optimized GeoTIFFs (Sentinel-2 / Landsat) for the watershed. It computes 50m-buffer zonal statistics before and after the photo timestamp:
   • $\Delta \text{NDVI} = \text{NDVI}{\text{post}} - \text{NDVI}{\text{pre}}$ (Vegetation growth)
   • $\Delta \text{NDWI} = \text{NDWI}{\text{post}} - \text{NDWI}{\text{pre}}$ (Surface water retention) [INFERENCE].
5. Score Fusion & Visualization: The backend calculates the Watershed Health & Verification Index (WHVI) and updates GeoServer. The web dashboard dynamically renders vector pins, satellite overlay layers, and interactive before/after swipe sliders [INFERENCE].

12. Technology Stack

13. AI/ML Justification

Component 1: Field Asset & Health Classifier (YOLOv8)

• Purpose: Automatically recognize physical watershed structures in field photos and detect damage or silting [INFERENCE].
• Model Type: Object Detection & Instance Segmentation (YOLOv8 medium/small) [INFERENCE].
• Input: Raw field photograph JPEG ($1024 \times 768$ or higher) [INFERENCE].
• Output: Bounding boxes, class labels (check_dam, farm_pond, trench, gully_plug, afforestation), confidence score, structural condition flag (intact, silted, eroded) [INFERENCE].
• Data Required: ~1,500 annotated field images across Indian watershed topographies [INFERENCE].
• Why AI is Preferable: Human manual audit of 1,000,000+ uploaded Drishti photos is impossible; simple color/edge algorithms fail under varying field lighting and vegetation camouflage [INFERENCE].
• Evaluation: Mean Average Precision ([email protected]) $> 85%$ [INFERENCE].
• Failure Mode: Poor lighting or extreme close-up photos causing misclassification. System flags confidence $<0.60$ for human review [INFERENCE].
• Human Oversight: District Project Manager gets an "Unverified/Flagged" queue to approve low-confidence AI detections [INFERENCE].

Component 2: Spatio-Temporal Anomaly Detector (Isolated Forest / Z-score Heuristic)

• Purpose: Flag fraudulent or mislocated photo uploads where claimed structure impact contradicts satellite spectral physics [INFERENCE].
• Model Type: Classical Anomaly Detection / Rule-Based Multi-Criteria Scoring [INFERENCE].
• Input: GPS location, photo azimuth, $\Delta\text{NDWI}$, $\Delta\text{NDVI}$, distance to DEM stream line [INFERENCE].
• Output: Anomaly Score ($0-100%$) and Flag (Location Mismatch, Implausible Spectral Response) [INFERENCE].
• Data Required: Historical Sentinel-2 time series + PostGIS stream vectors [INFERENCE].
• Why AI/ML is Preferable: Captures non-linear relationships between slope, stream order, and satellite index change [INFERENCE].
• Evaluation: False Positive Rate $< 5%$ on test benchmark [INFERENCE].
• Failure Mode: Seasonal drought causing low NDWI despite valid check dam presence. System incorporates rainfall data (TRMM/CHIRPS) to prevent false alerts [INFERENCE].
• Human Oversight: Mandatory manual review before rejecting any field worker submission [INFERENCE].

Where AI Should NOT Be Used

• Stream Network Delineation & Catchment Masking: AI/Deep Learning should NOT be used for basic hydro-delineation. Classical, deterministic hydrologic models (D8 flow direction, Deterministic 8-neighbor flow accumulation on DEM via WhiteboxTools/GDAL) are 100% mathematically precise, fast, and physically consistent [INFERENCE].

14. Data Architecture & Offline/Low-Network Design

Database Schema (PostgreSQL/PostGIS SQL Excerpt)

-- Micro-Watershed Boundaries
CREATE TABLE micro_watersheds (
    watershed_id VARCHAR(32) PRIMARY KEY,
    name VARCHAR(100),
    district VARCHAR(100),
    state VARCHAR(100),
    area_ha NUMERIC(10,2),
    geom GEOMETRY(MultiPolygon, 4326)
);

-- Geo-Coded Field Photos
CREATE TABLE geo_photos (
    photo_id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    watershed_id VARCHAR(32) REFERENCES micro_watersheds(watershed_id),
    worker_id VARCHAR(50),
    captured_at TIMESTAMP WITH TIME ZONE,
    gps_lat NUMERIC(10,7),
    gps_lng NUMERIC(10,7),
    azimuth_deg NUMERIC(5,2),
    altitude_m NUMERIC(7,2),
    location GEOMETRY(Point, 4326),
    image_url TEXT,
    ai_asset_type VARCHAR(50),
    ai_confidence NUMERIC(4,3),
    ai_condition VARCHAR(50),
    is_spatially_valid BOOLEAN,
    whvi_score NUMERIC(5,2),
    sync_status VARCHAR(20) DEFAULT 'SYNCED'
);

-- Spatial Indexing
CREATE INDEX idx_geo_photos_loc ON geo_photos USING GIST(location);
CREATE INDEX idx_watershed_geom ON micro_watersheds USING GIST(geom);

Offline & Synchronization Strategy

1. PWA Local Storage: Field app uses IndexedDB / SQLite (via Capacitors/PWA) to store offline map tiles (Vector PBFs for target micro-watershed) and pending photo submissions [INFERENCE].
2. Compress-On-Capture: High-res camera photos ($12\text{MP}+$) are locally downsampled to $1920 \times 1080$ JPEG ($80%$ quality) with EXIF metadata preserved, reducing upload size from $\sim 5\text{MB}$ to $\sim 400\text{KB}$ [INFERENCE].
3. Background Sync & Retry: Web API client registers a Service Worker with Exponential Backoff retry strategy. When 2G/3G connectivity is restored, pending queue uploads automatically [INFERENCE].
4. Conflict Resolution: Last-Write-Wins (LWW) based on client timestamp, with cryptographic hash comparison to prevent duplicate photo uploads [INFERENCE].

15. Security, Privacy & Governance

1. Personal & Location Data Protection: Field worker identity and exact capture coordinates are stored with Role-Based Access Control (RBAC). Location coordinates are obfuscated/rounded on public-facing citizen views to protect field worker privacy [INFERENCE].
2. Image Integrity & Anti-Spoofing: EXIF timestamp and GPS tags are validated against device hardware sensors. Uploaded images undergo perceptual hashing (pHash) to detect duplicated/re-uploaded web photos [INFERENCE].
3. Authentication & Authorization: JWT-based OAuth2 authentication. Roles: Field_Worker, PIA_Officer, SLNA_Admin, Public_Auditor [INFERENCE].
4. Data Ownership & Regulatory Compliance: All spatial data and derived thematic maps belong to Department of Land Resources (DoLR), MoRD [PS]. Outputs produced during hackathon are strictly tagged as Prototype Demonstrations and require independent field validation prior to policy enforcement [INFERENCE].

16. MVP Scope (Hackathon Build)

In Scope

• Interactive Dual-Map Dashboard: Web GIS interface displaying micro-watershed boundary, 30m satellite basemap (Sentinel-2/Landsat), and geo-tagged photo pins [INFERENCE].
• Geo-Photo AI Classifier: YOLOv8 model running on FastAPI server to classify 5 watershed structure types from uploaded photos [INFERENCE].
• Hydrological Overlay: DEM-derived D8 drainage network overlay with automatic proximity check ($\le 30\text{m}$ to stream) for photo locations [INFERENCE].
• Spatio-Temporal Delta Engine: Dynamic calculation of pre/post NDVI and NDWI zonal statistics in a 50m buffer around the photo pin [INFERENCE].
• Automated Report Generator: Exportable PDF/GeoJSON summary report for a selected micro-watershed showing intervention density, vegetation growth delta, and anomaly alerts [INFERENCE].
• Offline Field App Simulation: Responsive PWA interface showcasing offline photo capture, EXIF parsing, and sync queue [INFERENCE].

Described Only / Out of Scope

• Direct live production database sync with official ISRO Bhuvan / DoLR servers (requires official government API credentials) [NEEDS VALIDATION].
• Multi-terabyte nationwide satellite processing (restricted to sample district/micro-watersheds for demonstration) [ASSUMPTION].
• Custom satellite sensor calibration or hardware drone imagery processing [INFERENCE].

17. Live Demo Simulation Plan

To present a compelling, failure-proof live demo without relying on external government network availability:

1. Simulated Bhuvan WMS / Satellite Stream:
   • Host pre-downloaded, Cloud-Optimized GeoTIFFs (COGs) of a real Indian watershed district (e.g., Ananthapuramu, AP or Solapur, MH) on a local MinIO/GeoServer instance [ASSUMPTION].
   • Deliver instant WMS/WMTS map tiles locally at $60\text{fps}$ [INFERENCE].
2. Field Photo Capture Simulation:
   • Provide a live Android phone / browser device connected to the local server.
   • Capture a photo of a test check dam image (or select from a pre-loaded suite of sample field images from the provided PS Drive link https://drive.google.com/drive/folders/1ibmzWpl_nK7aBhQurs22R9kqh9fPAQwC) [PS].
   • App parses GPS EXIF data in real-time, displays metadata, and queues upload [INFERENCE].
3. Live AI & Spatial Analysis Execution:
   • Trigger the backend API live during the presentation.
   • Show YOLOv8 bounding box detection (Check Dam - 92% confidence), PostGIS stream proximity check (Passed: 12m from 2nd-order stream), and pre/post satellite NDVI swipe comparison [INFERENCE].
4. Interactive Anomaly Trigger:
   • Deliberately upload a photo tagged in an invalid hilltop location.
   • System flags: CRITICAL ANOMALY: Check dam photo located 450m away from drainage network. Flagged for review. proving system intelligence [INFERENCE].

18. Implementation Roadmap

Phase 1: Environment & Data Prep (Days 1 - 3)
├── Setup PostgreSQL/PostGIS database schema
├── Download Sentinel-2/Landsat 30m rasters & CartoDEM for target test region
├── Run WhiteboxTools script to extract D8 drainage vectors
└── Set up MinIO object storage and FastAPI skeleton

Phase 2: AI Vision Model & GIS Engine (Days 4 - 8)
├── Train/fine-tune YOLOv8 on watershed asset dataset
├── Build Python GDAL/Rasterio pipeline for NDVI/NDWI zonal stats
├── Implement PostGIS spatial validation queries (ST_DWithin to streams)
└── Develop FastAPI REST endpoints for photo upload & spatial analytics

Phase 3: Web GIS Frontend & PWA Development (Days 9 - 13)
├── Build React + Leaflet/MapLibre map viewer interface
├── Integrate layer toggle (Satellite, DEM, Drainage, Photos, Heatmaps)
├── Implement Before/After image comparison slider
├── Create Offline PWA mobile submission view
└── Design Executive Dashboard analytics & PDF exporter

Phase 4: Integration, Testing & Demo Polish (Days 14 - 15)
├── Conduct end-to-end simulation test (Field app -> AI -> PostGIS -> Map UI)
├── Run false-positive & offline sync boundary testing
├── Pre-cache map tiles for offline demo stability
└── Finalize presentation deck, architecture diagrams, and video walkthrough

Target Milestone Completion Date: Well ahead of 30 September 2026 [PS]

19. Testing & Validation Plan

20. Risk Register

21. Cost & Scalability

Hackathon Prototype

• Infrastructure Cost: ₹0 (Runs entirely on student laptops using Docker, PostgreSQL/PostGIS, GeoServer, and Python open-source stack) [INFERENCE].
• Data Cost: ₹0 (Utilizes free public Copernicus Sentinel-2 30m resampled data, CartoDEM/SRTM 30m, and sample photos from PS Drive folder) [INFERENCE].

Production Deployment (State / National Scale - e.g., 500 Districts)

• Cloud Compute: ~₹1,500,000 / year (Kubernetes cluster on AWS/NIC Cloud with autoscaling FastAPI worker nodes and GPU instances for batch YOLO inference) [NEEDS VALIDATION].
• Storage: ~₹500,000 / year (S3/MinIO storage for multi-terabyte satellite rasters and millions of JPEG field photos) [NEEDS VALIDATION].
• Database: ~₹600,000 / year (Managed PostgreSQL High-Availability cluster with PostGIS extension) [NEEDS VALIDATION].
• Total Estimated Operating Cost: ~₹2,600,000 / year (Extremely cost-effective compared to manual field inspection overhead across India) [INFERENCE].

22. Differentiation

23. Future Scope

1. Sub-Meter Planet / Cartosat Integration: Upgrade satellite pipeline to ingest high-resolution (1m-2.5m) Cartosat/PlanetScope imagery when available for micro-structure edge detection [INFERENCE].
2. SAR Satellite Moisture Monitoring: Ingest Sentinel-1 Synthetic Aperture Radar (SAR) C-band data to measure subsurface soil moisture changes independent of cloud cover [INFERENCE].
3. Crowdsourced Citizen Audit Portal: Allow local farmers to view verified watershed interventions in their Gram Panchayat and submit feedback/ratings on water availability [INFERENCE].
4. Automated Budget Disbursement Triggers: Integrate with Public Financial Management System (PFMS) to automatically trigger stage-wise financial releases upon AI spatial verification of completed check dams [INFERENCE].

24. Assumptions & Open Items Log

Consolidated Evidence & Assumption Tags

• [PS]: Requirements directly sourced from SIH26015 JSON statement [PS].
• [KNOWN]: Established facts regarding WDC-PMKSY, DoLR, ISRO Bhuvan, and remote sensing standards [KNOWN].
• [RESEARCHED]: Facts verified via PIB, NRSC publications, and official government releases [RESEARCHED].
• [INFERENCE]: Architectural, technical, and algorithmic deductions derived to meet problem requirements [INFERENCE].
• [ASSUMPTION]: Operational decisions made for hackathon buildability (e.g., using Sentinel-2/Landsat as 30m proxy, MinIO local S3) [ASSUMPTION].
• [NEEDS VALIDATION]: Items requiring official DoLR/ISRO API confirmation or field validation [NEEDS VALIDATION].

Open Items Needing Field Validation

1. Official Bhuvan API Access: Confirm whether DoLR/NRSC will grant direct REST/WFS API endpoints for Drishti database sync during later implementation phases [NEEDS VALIDATION].
2. Local Hydrological Calibrations: Verify local soil run-off coefficients (Curve Number method) for specific semi-arid micro-watersheds [NEEDS VALIDATION].

Final Recommendation

Adopt Approach 3: Hybrid Geo-AI Spatial Validation Engine. It provides the exact balance of computer vision image interpretation, hydrological spatial verification, and satellite time-series analytics demanded by the Ministry of Rural Development [INFERENCE].

Recommended Stack

• Frontend: React.js, Tailwind CSS, Leaflet.js, Leaflet-Geoman
• Backend: Python FastAPI, GDAL, Rasterio, WhiteboxTools, PyTorch (YOLOv8)
• Database & Server: PostgreSQL 16 + PostGIS 3.4, GeoServer, MinIO Object Storage
• DevOps: Docker, Docker Compose, Nginx

Biggest Dependency

Availability of high-quality sample field photos with valid GPS EXIF metadata from the provided DoLR Drive repository (https://drive.google.com/drive/folders/1ibmzWpl_nK7aBhQurs22R9kqh9fPAQwC) [PS].

Biggest Risk

30m satellite spatial resolution blur failing to show structural boundaries of small check dams. (Mitigated by evaluating 50m surrounding zonal spectral index deltas ($\Delta\text{NDWI}/\Delta\text{NDVI}$) rather than pixel boundaries) [INFERENCE].

Biggest Technical Challenge

Performing fast, on-the-fly raster zonal statistics across multi-temporal satellite images when a user clicks on a photo pin on the web map [INFERENCE].

First 5 Actions

1. Download sample data folder from provided PS Drive link and extract JPEG EXIF metadata [PS].
2. Set up PostgreSQL + PostGIS Docker container and ingest sample micro-watershed GeoJSON vectors [INFERENCE].
3. Download 30m Sentinel-2 / Landsat-8 geotiffs and CartoDEM for a target district [INFERENCE].
4. Run WhiteboxTools script in Python to generate D8 drainage network vectors and store in PostGIS [INFERENCE].
5. Annotate 100 field photos and train baseline YOLOv8 asset detection model [INFERENCE].

Validation Before Coding

• Verify that GDAL and PostGIS Docker containers communicate seamlessly.
• Verify that sample JPEGs contain valid GPSLatitude and GPSLongitude EXIF header tags.
• Confirm that WhiteboxTools accurately extracts stream orders matching known district topographic maps.

Buildability Score / 100

92 / 100 (Highly feasible for a 4-6 student team using standard open-source geospatial and AI libraries within the hackathon timeframe) [INFERENCE].

Confidence

HIGH (The solution directly answers every explicit PS requirement and avoids reliance on unverified proprietary software) [INFERENCE].

Research Sources

1. Press Information Bureau (PIB), Govt of India (2023): MoU with NRSC for Watershed Development under WDC-PMKSY 2.0. https://www.pib.gov.in/PressReleaseIframePage.aspx?PRID=1968491

   • Source Type: Official Government Release
   • Supported Claim: DoLR partnership with ISRO NRSC for Bhuvan Srishti web portal and Drishti mobile app monitoring of WDC-PMKSY 2.0.
   • Reliability: High (Official Government Portal).

2. ISRO Bhuvan Geoportal - IWMP Srishti: Bhuvan IWMP Application Portal. https://bhuvan-app1.nrsc.gov.in/iwmp/

   • Source Type: Official Spatial Portal
   • Supported Claim: Existing functionality of Bhuvan IWMP/Srishti portal and Drishti geotagging workflow.
   • Reliability: High (Official ISRO Portal).

3. Andhra Pradesh Space Applications Centre (APSAC) (2023): Monitoring of IWMP Watershed Projects Using Geo-Information. https://apsac.ap.gov.in/wp-content/uploads/2023/05/GUNTUR_IWMP_04_GURAJALA.pdf

   • Source Type: Government Technical Report
   • Supported Claim: Use of multi-temporal satellite imagery (Cartosat, LISS-IV) alongside Drishti geotagged points for T0-T5 baseline and impact monitoring.
   • Reliability: High (State Space Application Centre Report).

4. ISRO Annual Report (2018-2019): Geospatial Applications in Watershed Monitoring. https://www.vssc.gov.in/pdf/AnnualReport/annual_report_2018-19.pdf

   • Source Type: Official Agency Annual Report
   • Supported Claim: Over 9.68 lakh ground activity geo-tags uploaded via Bhuvan-Drishti onto Bhuvan-Srishti platform across 6,850 IWMP projects nationwide.
   • Reliability: High (Official ISRO Publication).