Groundwater_Risk_prediction_system

Groundwater Risk Prediction System using Graph Attention Networks

An AI-driven Integrated Water Intelligence Framework for district-level groundwater risk prediction by combining groundwater stress assessment, water quality analysis, water pipeline leakage detection, and graph neural networks.

---

Project Overview

Water resource management is one of the major sustainability challenges in India. Existing systems generally analyze groundwater quantity, water quality, and infrastructure failures independently, making it difficult to obtain a comprehensive understanding of district-level water security.

This project proposes an Integrated Water Intelligence Framework that combines:

• Groundwater Stress Assessment
• Water Quality Assessment
• Water Pipeline Leakage Detection
• Graph Attention Networks (GAT)

to generate an Integrated Water Risk Index (IWRI) for each district.

---

Key Features

• Groundwater Stress Index (GSI)
• Water Quality Score (WQS)
• Water Pipeline Leakage Detection using YOLOv8
• Integrated Water Risk Index (IWRI)
• District-wise Graph Construction
• Graph Attention Network for spatial learning
• Explainable AI using attention weights
• District-level risk ranking

---

System Architecture

Groundwater Dataset
        │
        ▼
Groundwater Stress Assessment
        │
        ▼
Groundwater Stress Score
──────────────────────────────────

Water Quality Dataset
        │
        ▼
Water Quality Assessment
        │
        ▼
Water Quality Score
──────────────────────────────────

Leakage Images
        │
        ▼
YOLOv8 Leakage Detection
        │
        ▼
Leakage Score
──────────────────────────────────

Population + Rainfall
        │
        ▼
Feature Integration
        │
        ▼
Integrated Water Risk Index
        │
        ▼
District Graph Construction
        │
        ▼
Graph Attention Network
        │
        ▼
District Risk Prediction

---

Repository Structure

Groundwater_Risk_Prediction_System/

│
├── README.md
│
├── stress_assessment.ipynb
│      Groundwater stress assessment pipeline
│      • Data preprocessing
│      • Feature engineering
│      • Groundwater Stress Index calculation
│
├── water_potability.csv
│      Water quality dataset
│
├── rs.csv
│      Groundwater resource dataset
│
├── water_leak_detection_1000_rows.csv
│      Leakage detection annotations
│
├── water-leakage-detection.ipynb
│      YOLOv8 training pipeline
│      • Dataset loading
│      • Model training
│      • Prediction
│
└── outputs/
       Model predictions
       Risk maps
       Trained models

---

Datasets Used

1. Groundwater Resource Dataset

Source: Central Ground Water Board (CGWB)

Features:

• Annual Recharge
• Net Groundwater Availability
• Annual Draft
• Stage of Groundwater Development
• Future Irrigation Availability
• Domestic Water Demand

---

2. Water Quality Dataset

Source:

Kaggle Water Potability Dataset

Features

• pH
• Hardness
• Solids
• Chloramines
• Sulfate
• Conductivity
• Organic Carbon
• Trihalomethanes
• Turbidity

---

3. Leakage Detection Dataset

Source

Water Pipes Dataset (Kaggle)

Contains annotated images of water pipelines for leakage detection using YOLOv8.

---

Methodology

Step 1

Groundwater Stress Assessment

Calculate

• Recharge
• Extraction
• Stress Index
• Irrigation Dependency
• Future Sustainability

↓

Groundwater Stress Score

---

Step 2

Water Quality Assessment

Preprocess

↓

Normalize

↓

Train ML Model

↓

Generate Water Quality Score

---

Step 3

Leakage Detection

YOLOv8

↓

Leak Detection

↓

Leakage Severity Score

---

Step 4

Integrated Water Risk Index

Combine

Groundwater Stress

Water Quality

Leakage

↓

IWRI

---

Step 5

Graph Construction

Each district

↓

Node

Neighbouring districts (<100 km)

↓

Edges

↓

Weighted Graph

---

Step 6

Graph Attention Network

Input Features

↓

GAT Layer

↓

Attention Learning

↓

District Risk Prediction

---

Machine Learning Models

Module	Algorithm
Water Quality	Random Forest
Leakage Detection	YOLOv8
Spatial Learning	Graph Attention Network

---

Evaluation Metrics

Regression

• MAE
• RMSE
• R² Score

Detection

• Precision
• Recall
• F1 Score
• mAP

---

Results

Example district prediction

District	Risk Score
Lucknow	82
Kanpur	74
Agra	45

---

Technologies Used

• Python
• Pandas
• NumPy
• Scikit-learn
• PyTorch
• PyTorch Geometric
• Ultralytics YOLOv8
• Matplotlib
• OpenCV

---

Future Improvements

• IoT sensor integration
• Real-time groundwater monitoring
• Satellite imagery integration
• Aquifer-level graph modeling
• Interactive web dashboard
• State-wide and national-scale deployment

---

Research Paper

This repository accompanies the research work:

"An AI-driven Integrated Water Intelligence Framework for Explainable District-Level Groundwater Risk Prediction using Graph Attention Networks."

---

Citation

@misc{groundwaterrisk2026,
  title={Groundwater Risk Prediction System using Graph Attention Networks},
  author={Sejal Pandey},
  year={2026},
  howpublished={GitHub Repository}
}

---

Author

Sejal Pandey

• LinkedIn: https://www.linkedin.com/in/sejal-pandey-2aa018270/
• GitHub: https://github.com/sejalpandey30