A new study presents a method for estimating above-ground carbon at an individual tree level, especially in semi-arid regions, by utilizing very high-resolution satellite imagery coupled with machine learning algorithms. This approach offers a more precise tool for measuring carbon sequestration, which is critical for climate adaptation and land management strategies on a global scale. The development, detailed in a study published on November 21, 2024, in Journal of Remote Sensing, promises to revolutionize land management and climate adaptation strategies.
The core innovation of this study lies in the development of an Artificial Neural Network model trained on over 400 individual tree crowns, incorporating both spectral signatures and crown area extracted from Pléiades high-resolution satellite imagery. By doing so, the researchers achieved more precise above-ground carbon estimates, delivering an R² of 0.66 and a relative RMSE of 78.6%. This method significantly reduces the biases seen in previous technologies, particularly those that underestimated carbon stocks in dryland regions.
To create this model, the researchers constructed a comprehensive above-ground carbon reference database from on-the-ground tree measurements, converting them into biomass using species-specific allometric equations. Through deep learning models, they were able to segment individual tree crowns and extract spectral information from very high-resolution imagery, which was then used to train and validate the ANN model. The result was a highly accurate model, with a tree-level RMSE of just 373.85 kg, confirming its robustness in predicting above-ground carbon from remote sensing data.
The study utilized Pléiades Neo satellite imagery, known for its exceptional 30cm native resolution, which enabled unprecedented precision in Earth observation. This precision, combined with deep learning algorithms for crown extraction and ANN models for above-ground carbon prediction, allowed for the accurate geolocation of individual trees—addressing longstanding limitations in carbon stock estimation. The original research can be accessed at https://doi.org/10.34133/remotesensing.0359.
Looking ahead, the future applications of this technology are vast. It promises to improve global carbon cycle assessments, optimize land use, and enhance reforestation initiatives. Furthermore, it could provide essential data for climate change mitigation strategies, helping policymakers address pressing environmental challenges. As the method gains wider adoption, it has the potential to harmonize carbon estimation discrepancies, offering invaluable support for international climate agreements and global sustainability efforts.
