Pakistan Climate Prediction System: Advanced ML for Flood Risk Analysis

A production-grade ML system predicting rainfall and temperature patterns in Pakistan using 116 years of historical data (1901-2016) to support flood risk assessment during critical monsoon periods.

Key Achievement: R² = 0.992 for temperature, R² = 0.716 for rainfall prediction.

Quick Start

Docker (Recommended)

docker build -t pakistan-climate-engine .
docker run -p 8000:8000 pakistan-climate-engine

Local Python

pip install -r requirements.txt
uvicorn fastapi_app:app --reload

API Usage

Access interactive docs at: http://localhost:8000/docs

Endpoint: /predict (POST)

Request:

{
  "year": 2025,
  "month": 7
}

Response:

{
  "year": 2025,
  "month": "July",
  "rainfall": 56.6,
  "temperature": 28.9,
  "season": "Summer",
  "is_monsoon": true,
  "success": true,
  "method": "pipeline"
}

Project Context & Significance

Pakistan's Flood Challenge

Pakistan faces devastating floods during July-August monsoon seasons. This system provides:

• Early warning for flood-prone regions
• Agricultural planning support
• Water resource management
• Emergency preparedness

Picture Reference: CNN World Article (https://edition.cnn.com/2025/07/17/asia/pakistan-flood-deaths-climate-intl-hnk)

Technical Architecture

Production Features

• FastAPI: Asynchronous inference with auto-generated docs
• Docker: Fully containerized for reproducibility
• CI/CD: GitHub Actions for automated testing
• Error Handling: Smart fallback to heuristic models
• Type Safety: Pydantic validation

Hybrid Modeling Strategy

• Temperature: Advanced 67-feature engineering (R² = 0.992)
• Rainfall: Optimized simple approach (R² = 0.716)

Algorithm Portfolio

• Linear/Ridge Regression, Random Forest
• Gradient Boosting, XGBoost
• Support Vector Regression

Model Performance

Target Variable	Model Architecture	Performance
Rainfall	Gradient Boosting Regressor	See Notebook
Temperature	LightGBM / Random Forest	See Notebook

Data Visualization & Analysis

Flood Risk Analysis

Model Evaluation

Data Visualization

Deployment Options

1. FastAPI (Recommended)

Production-ready API with MLOps pipeline

2. Streamlit Application

Interactive web interface for real-time predictions 📹 Watch Demo Video

3. Flask Web Service

RESTful API with batch processing capabilities

Project Structure

pakistan_temp_rainfall_predictive_modelling/
├── .github/workflows/           # CI/CD Pipeline
├── Flask_app/                   # Flask API service
├── Streamlit_app/              # Streamlit interface
├── media-assets/               # Images and videos
├── research/                   # Jupyter notebooks & analysis
├── fastapi_app.py             # Main API entry point
├── Dockerfile                 # Container configuration
├── requirements.txt           # Dependencies
└── *.joblib                   # Trained models

System Limitations

• Temporal Range: Data limited to 1901-2016
• Spatial Resolution: National averages only
• Climate Change: Recent shifts may not be captured

Future Enhancements

• Deep Learning integration (LSTM networks)
• Regional province-level models
• Real-time weather API integration
• Mobile-responsive interface

Data Source & Contact

Data: CHISEL @ LUMS Center for Climate Research Analysis Period: 1901-2016 Pakistan Climate Dataset

Author: Ahsan Javed - Data Scientist & ML Engineer

• LinkedIn: Ahsan Javed
• Email: [email protected]

---

Disclaimer: For critical flood management decisions, integrate with official meteorological services and consider model limitations.