Using Sentinel 2A and Landsat 8 imagery to assess changes in forest carbon storage.

Estimating carbon storage using high-resolution imagery of dominant species and types is often constrained by the availability of data. Herein, we developed a carbon storage estimation model for dominant species and types using high-resolution Sentinel 2A imagery and compared the two approaches using lower-resolution Landsat 8 imagery for whole-forest estimation. Approach 1 employs a traditional method using in-situ carbon storage measurements with Landsat 8 vegetation indices, whereas Approach 2 uses Sentinel 2A carbon storage estimates as a reference. Using Random Forest, Decision Tree, and Multiple Linear Regression models, we compared both approaches and found that Approach 2 estimates matched the Sentinel 2A results for different species and types more accurately, including Populus, Salix, Pinus tabuliformis, and shrub types. At the same time, our research results show that machine learning models effectively estimated carbon storage using Sentinel 2A imagery and dominant species classification. For the whole forest assessment with Landsat 8 imagery, Approach 2 yielded superior accuracy over Approach 1. This method enabled the calculation of historical carbon storage, showing that the Ordos Forest carbon storage increased by 27 Mt (89%) from 2013 to 2023, demonstrating the feasibility of long-term carbon monitoring using lower-resolution imagery.