🏔️ PARU: Ecological Surveillance AI

Protect. Analyze. Restore. Uttarakhand.

PARU is an end-to-end geospatial intelligence system designed to monitor forest health in the Himalayan region. Unlike standard vegetation indices that yield high false positives during dry seasons, PARU employs a multi-factor forensic engine to identify specific signatures of structural deforestation.

---

🚨 The Challenge

Satellite monitoring of the Himalayas is difficult due to extreme seasonality. A drop in "Greenness" (NDVI) often indicates winter dormancy, not deforestation. Distinguishing between Drought Stress (natural) and Biomass Loss (anthropogenic/fire) is critical for effective conservation.

🧠 The Solution: "Triad Detection Engine"

PARU implements a custom forensic algorithm that scans for the "Deforestation Triad" signature. An alert is triggered only if three specific conditions align simultaneously:

1. Vegetation Collapse: A statistically significant drop in NDVI Z-Scores ($< -2\sigma$).
2. Thermal Shock: A sudden spike in Land Surface Temperature (LST), indicating bare soil exposure.
3. Rainfall Decoupling: Rainfall levels remain normal (ruling out drought as a cause).

---

⚡ Key Features

1. 🌲 Structural Loss Detector (The Core)

• Forensic Scan: Automatically scans the entire 25-year dataset (2000–2025) to identify active loss zones.
• Alert System: Flags specific districts in Red on a 3D Map where the "Triad" signature is detected.
• Drill-Down: Click on any alert to see a historical graph of the event.

2. 🔮 Predictive Climate Simulation

• AI Forecasting: Uses HistGradientBoostingRegressor to predict future vegetation health based on climatic variables.
• Interactive Sandbox: Users can modify rainfall/temperature sliders to simulate "What-If" scenarios (e.g., How will a 50% rainfall deficit impact Nainital next month?).
• Smart Logic: Includes "Synthetic Injection" logic to simulate unseasonal rain effects even during dry months.

3. ⏳ Time-Machine 3D Analysis

• Delta Mapping: Compare any two dates in history (e.g., May 2005 vs. May 2025).
• Visuals: Renders a 3D Difference Map (Green = Growth, Red = Loss, Height = Magnitude of change).
• AI Comparative Reports: Automatically generates a text report explaining why the change occurred.

4. 🛰️ Autonomous Swarm Monitor

• Standby Agent: Automatically flags any district with a negative growth trajectory in the current month without manual input, acting as an "Always-On" watchdog.

---

🛠️ Technical Architecture

Data Pipeline

• Source: Curated dataset of 25 years of monthly granular data (2000-2025).
• Variables: NDVI (Vegetation), LST (Temperature), Rain_Sum (Precipitation), Soil_Moisture.
• Processing: Moving averages (3-month rolling), Z-Score calculation, and Savgol Filtering for signal smoothing.

Tech Stack

• Core Logic: Python, Pandas, NumPy
• Machine Learning: Scikit-Learn (HistGradientBoostingRegressor)
• Geospatial: PyDeck (3D Rendering), GeoJSON
• Visualization: Plotly Express, Plotly Graph Objects
• GenAI Integration: Groq API (Llama-3-70b & Llama-3.1-8b Fallback) for automated reporting.

---

🚀 Installation & Setup

1. Clone the Repository

   git clone [https://github.com/YOUR_USERNAME/PARU-Ecological-AI.git](https://github.com/YOUR_USERNAME/PARU-Ecological-AI.git)
   cd PARU-Ecological-AI

2. Install Dependencies

   pip install -r requirements.txt

3. Run the Application

   streamlit run app.py

4. Configure AI To enable the automated text reports, add your Groq API Key to .streamlit/secrets.toml

   GROQ_API_KEY = "your_key_here"
   
   

---

📂 Directory Structure ```bash

PARU-Ecological-AI/ ├── app.py # 🚀 Main Application (Streamlit UI & Maps) ├── core_logic.py # 🧠 Scientific Feature Engine & Triad Logic ├── llm_engine.py # 🤖 Groq AI Handler (with Llama-3 Fallback) ├── update_utils.py # 🔄 Pipeline Execution & Retraining Utilities ├── knowledge.py # 📚 Domain Knowledge Base & System Prompts ├── benchmark.py # 📉 Model Performance Evaluation Script ├── data_checker.py # 🛡️ Data Quality & Integrity Validation ├── requirements.txt # 📦 Python Dependencies ├── Ultra_Forest_Model.joblib # 🔮 Pre-trained HistGradientBoosting Model ├── uttarakhand.geojson # 🗺️ Geospatial Polygon Data for 3D Maps └── Uttarakhand_Forest_Data_Corrected (2).csv # 📊 25-Year Curated Climate Dataset

🔮 Future Roadmap [ ] Computer Vision: Integrate Satellite Imagery (Sentinel-2) for real-time visual validation.

[ ] Edge Deployment: Optimization for running on low-resource IoT devices in remote forest offices.

[ ] SMS Alerts: Integration with Twilio to send SMS alerts to rangers when the "Triad" is detected.

👨‍💻 Author [Mohit Nautiyal] B.Tech Computer Science Engineering