⛈ AI Cloudburst Early Warning System

Physics-Informed Spatio-Temporal Deep Learning for Cloudburst Prediction and Impact Assessment in the Indian Himalayas

B.Tech Major Project — SRM Institute of Science and Technology, May 2026
Authors: Aaditya Goshike & Maddili Benarjee
Guide: Dr. Gowtham P In collaboration with: CSIR-Fourth Paradigm Institute (CSIR-4PI), Bengaluru

---

What it does

Predicts cloudbursts 1–3 hours in advance using a hybrid AI + physics approach, and assesses secondary impacts like flash floods, landslides, and runoff. Built for the Indian Himalayan region.

Core pipeline:
Multi-modal data → ConvLSTM + Transformer + PINN → Ensemble prediction → Impact assessment → Streamlit dashboard

---

Key Features

• Cloudburst probability with early warning (1–3 hour lead time)
• Flash flood, landslide, and runoff risk estimation
• Spatial risk heatmaps (12×12 grid)
• Physics-informed constraints (moisture conservation, orographic lifting, CAPE)
• Explainable AI outputs with feature importance
• Excel-driven input — no coding needed to change inputs
• Auto-detects file changes every 30 seconds

---

Performance

Metric	Value
Accuracy	85%
Precision	92.8%
Recall	86.7%
F1-Score	89.6%
ROC-AUC	0.913
PR-AUC	0.941

---

Screenshots

Dashboard

Heatmaps

Impact Analysis

Physics Indicators

Explainable AI

Alerts

Excel Data

---

Quick Start

# 1. Clone the repo
git clone https://github.com/YOUR_USERNAME/cloudburst-early-warning.git
cd cloudburst-early-warning

# 2. Install dependencies
pip install -r requirements.txt

# 3. Run
streamlit run app.py

Opens at http://localhost:8501

---

How the Excel Integration Works

data/cloudburst_data.xlsx  ←→  app.py
       │                            │
       │  Edit Weather_Input sheet   │  Reads on load / auto-detects changes
       │  (blue cells = inputs)      │  Runs ML + Physics engine
       │                             │  Writes results to Prediction_Log sheet
       └─────────────────────────────┘

Steps

1. Open data/cloudburst_data.xlsx
2. Go to Weather_Input sheet → edit the blue cells
3. Save the file (Ctrl+S / Cmd+S)
4. In Streamlit: click 🔄 Reload or wait ~30 sec for auto-detect
5. All 7 tabs update — Prediction, Heatmaps, Impact, Physics, XAI, Alerts, Excel Data

---

Project Structure

cloudburst-early-warning/
├── app.py                      ← Streamlit app (run this)
├── requirements.txt
├── data/
│   └── cloudburst_data.xlsx    ← Input/output data file
└── backend/
    ├── __init__.py
    ├── prediction_engine.py    ← ML + Physics engine
    └── excel_loader.py         ← Excel read/write module

---

Tech Stack

• ML/DL: ConvLSTM, Transformer, Attention Mechanism, PINN
• Physics: Moisture conservation, orographic lifting, CAPE, instability index
• Frontend: Streamlit, Plotly
• Data: NumPy, Pandas, openpyxl
• Deployment: Streamlit (local / cloud)

---

Publication

Paper submitted and abstract accepted at conference (2026).
Title: Physics-Informed Spatio-Temporal Deep Learning for Cloudburst Prediction and Impact Assessment in the Indian Himalayas