Automated Multi-Source Carbon Stock Estimation with Spatial Validation and Uncertainty-Aware Adjustment for Blue Carbon MRV

Abstract

Blue carbon ecosystems can play a vital role in mitigating climate change, but precise and efficient Monitoring, Reporting, and Verification (MRV) is complicated by spatial variability and uncertainty. This paper presents an automated multi-source carbon stock estimation system that includes IoT sensor telemetry, LiDAR-derived canopy properties and multi-spectral remote sensing to quantify Soil Organic Carbon (SOC) and Above-Ground Biomass (AGB). The models (Support Vector Regression, XGBoost and ensemble) are trained with laboratory-based ground truth. Spatial-block cross-validation is used to ensure spatial independence and prevent over fitting. Uncertainty is estimated via ensemble variance and confidence intervals and conservative adjustment of carbon estimates to comply with MRV and avoid overestimates. The carbon pool is obtained by combining SOC and AGB. All results (predictions, validation scores, and uncertainty) are securely hashed and stored in the InterPlanetary File System** to guarantee data integrity and traceability. Evaluation shows better prediction and spatial transferability, as well as system-level efficiency improvements, such as 2.4× faster throughput and 93.6% anomaly detection. The system facilitates transparent, scalable and reliable blue carbon estimation for the next-generation MRV systems.