mwahunga01/ngong_land_degradation

# SoilGuard AI Report - Land ReGen Hackathon 2025 ## Project Overview SoilGuard AI is an AI-powered Minimum Viable Product (MVP) designed to monitor and detect soil degradation and vegetation loss in Ngong Forest, Kenya, aligning with SDG 15: Life on Land. Developed during the Land ReGen Hackathon (October 8-13, 2025), this project leverages Sentinel-2 Surface Reflectance imagery to generate annual NDVI composites and employs a simple machine learning model to identify degradation hotspots. ## Objectives - **Theme**: Early detection of soil degradation and vegetation loss. - **ROI**: Ngong Forest (approx. bounding box: [36.73, -1.33, 36.76, -1.30]). - **Timeframe**: 2018–2022 annual median composites. - **Goal**: Provide client-ready insights for reforestation and sustainable land management planning. ## Methodology ### Data Collection - **Source**: Sentinel-2 Level-2A Surface Reflectance (`COPERNICUS/S2_SR`) from Google Earth Engine. - **Processing**: - Filtered for <10% cloud cover. - Applied cloud and shadow masking using QA60 and SCL bands. - Computed median NDVI composites for each year (2018–2022) over the ROI. ### Analysis - **NDVI Calculation**: `(NIR - Red) / (NIR + Red)` using B8 (NIR) and B4 (Red) bands. - **ML Model**: A PyTorch-based DegradationClassifier with a sigmoid activation to flag areas with NDVI < 0.3 as degraded. - **Visualization**: Interactive map layers via `geemap` and sample NDVI plots. ### Exports - Annual NDVI GeoTIFFs (2018–2022) exported to `C:\Users\MWAHUNGAPC\Documents\Projects\PLP\hackerthon\Land degradation\code\`. ## Results - **Image Count**: Varies by year (e.g., 2018: 50 images, 2022: 70 images) due to cloud cover. - **Degradation Risk**: Sample 2022 analysis shows ~25.4% of the area at risk (NDVI < 0.3). - **Visuals**: NDVI timelapse and sample plots generated for presentation. ## Visualizations - **Map**: Interactive Ngong Forest NDVI layers (2018–2022) available in the `geemap` map. …