# Varuna — Water Intelligence Engine

> GoatAI's AI-native water intelligence platform for groundwater quality, coastal salinity risk, and floodplain intelligence across India.

Built by GoatAI (goatai.io). Live at https://gws.vercel.app

---

## What Varuna does

Varuna analyses publicly available groundwater chemistry data from India's Central Ground Water Board (CGWB 2022) using physics-based diagnostics and AI-assisted forecasting. It is the water module of GoatAI's broader AI Systems Intelligence platform for coupled physical environments.

---

## Modules

### National Groundwater Overview (Landing page)

- 13,506 CGWB wells mapped across all Indian states
- Geography tab: SVG choropleth coloured by WQI risk, data quality (CBE pass rate), or primary contaminant per state
- Quality tab: all 13,506 wells as individual MapLibre dots coloured by BIS failure count (Good / Moderate / Warning / Poor)
- Hover on any well → quality card: EC, pH, TDS, fluoride, nitrate, chloride, CBE status, exceedance flags

### India Groundwater Audit

- Charge Balance Error (CBE) analysis for all 13,506 wells
- State-level pass rate choropleth (Pass Rate / Bias Pattern / Failure Mode views)
- Detects pre-filtered datasets (Rajasthan, J&K: 100% pass rate is statistically implausible)
- Detects systematic bias (Kerala: cation excess → likely missing silica)
- Detects physical violations (Gujarat: ion mass > TDS in some samples)

### Dholera AOI — Coastal Salinity Risk

- Area-of-Interest engine for Dholera Special Investment Region, Gujarat
- Band A (0–10 km from coast) and Band B (10–30 km) well classification
- Rule-based coastal salinity risk score: 0.50 × distance_score + 0.35 × EC_score + 0.15 × Cl_ratio
- Sampling plan: 3 perpendicular transects × 4 distance stations (2/5/10/20 km inland)
- Forecast-uncertainty field from IDW + advection-diffusion simulation with data assimilation
- DEM + Natural Earth land-mask validity filter (no sampling points placed in the sea)
- Full methodology page: §1 CBE, §2 Band assignment, §3 Risk scoring, §4 Confidence, §5 Sampling plan (incl. §5.6 validity filter), §6 Data sources, §7 Limitations

### River Floodplain Analysis

- Floodplain risk and inundation zone mapping

---

## Data Sources

| Source | Coverage | Notes |
|--------|----------|-------|
| CGWB 2022 Ground Water Quality | India · 13,506 wells | Unconfined aquifers only |
| Natural Earth ne_50m_land | Global | Validity mask for sampling plan |
| SRTM 30m DEM | Gujarat AOI | Elevation filter for sampling plan |
| India state boundaries | India | CC0 license |
| Gulf of Khambhat coastline | Dholera AOI | 7-vertex polyline, ±3 km accuracy |

---

## Scientific methodology

All methods are documented, auditable, and contain no black-box ML:

**Charge Balance Error (CBE)**
CBE = 100 × (Σcations − Σanions) / (Σcations + Σanions), in meq/L.
Threshold: |CBE| ≤ 5% (Hounslow, 1995).

**WQI**
WHO/BIS weighted scoring across EC, TDS, hardness, chloride, fluoride, nitrate.

**Coastal salinity risk**
risk = 0.50 × max(0, 1 − dist_km/30) + 0.35 × min(1, EC/3000) + 0.15 × Cl_meq/(Cl_meq + TH_meq)
Classes: HIGH ≥ 0.66 · MEDIUM 0.33–0.66 · LOW < 0.33.

**Sampling plan generation (test7.py)**
1. IDW interpolation → initial contamination field
2. Advection-diffusion simulation with data assimilation (nudging, g = 0.6)
3. Uncertainty field = smoothed forecast × spatial data-gap product
4. Greedy selection maximising (uncertainty + forecast) over valid cells
5. Validity filter: DEM ≥ 1.0 m AND centroid ∈ Natural Earth land polygon

---

## Technology

- Frontend: React 19, MapLibre GL, Vite, pure SVG charts (no charting library)
- Styling: CSS custom properties (design token file), inline React style objects
- Routing: client-side `useState` (no React Router)
- Data pipeline: Python (NumPy, SciPy, Shapely, Rasterio, GeoPandas)
- Deployment: Vercel (static SPA)

---

## Organisation

GoatAI is an AI systems intelligence company focused on coupled physical environments.
Domains: Water · Infrastructure · Industrial · Mobility · Agriculture.

- Website: https://goatai.io
- Platform: https://gws.vercel.app
- Contact: business@goatai.io
- Founders: Prabhat Tiwari (industrial automation, infrastructure), Aditya Tiwari (ML, geospatial systems)
- Partners: IWMI, IIT Delhi HydroSense Lab, Poornima College of Engineering, Tata Steel
- Research submitted: IWA SCED-2026, ICIMOD HKH Conference