PeatWatch

# Multi-source Remote Sensing Peatland Monitoring System A deep learning-based system for monitoring peatland degradation and assessing restoration effectiveness using multi-source remote sensing data. This project integrates spectral, spatial, and temporal information with explainable AI components to support ecological monitoring and decision-making. ## 🛰️ Overview Peatlands are vital for climate regulation, carbon storage, and biodiversity conservation. However, they face rapid degradation due to human activities. This repository presents **GeoSpectral Adaptive Network (GeoSAN)** and the **Contextual Earth-Aware Reasoning (CEAR)** framework, offering a scalable and interpretable solution to monitor and evaluate peatland degradation and restoration using multi-modal Earth observation data. ## 🚀 Key Features - **GeoSAN Architecture** - Spectral reweighting mechanism - Deformable spatial learning - Auxiliary feature fusion - Multi-resolution context aggregation - **CEAR Inference Framework** - Graph-based contextual reasoning - Integration of physics-guided priors (e.g., NDVI) - Spatio-temporal alignment and geolocation awareness - **Multi-source Data Support** - Optical (e.g., Sentinel-2) - SAR (e.g., Sentinel-1) - LiDAR-derived elevation - Auxiliary geophysical priors (e.g., slope, land cover history) ## 📊 Datasets - **BigEarthNet** Multi-spectral imagery across Europe with CORINE land cover labels. - **UC Merced Land Use** High-resolution RGB aerial scenes across 21 land-use categories. - **SpaceNet** Urban satellite imagery with building footprints and road networks. - **RSI-CB256** Google Earth RGB tiles labeled for land-use classification. ## 📈 Performance Highlights | Dataset | Precision | Recall | NDCG | MAP ||----------------|-----------|--------|--------|--------|| BigEarthNet | 78.92% | 75.14% | 76.80% | 74.52% || UC Merced | 77.45% | 73.90% | 75.61% | 72.88% || SpaceNet | 76.88% | 72.67% | 73.91% | 70.84% || RSI-CB256 | 78.22% | 74.19% | 75.63% | 72.75% | *Outperforms baseline models such as SASRec, LightGCN, and ConvNCF across all benchmarks.* ## 🛠️ Installation ```bashgit clone https://github.com/your-username/peatland-monitoring.gitcd peatland-monitoringpip install -r requirements.txt