QRS-Trs: Style Transfer-Based Image-to-Image Translation for Carbon Stock Estimation in Quantitative Remote Sensing

Forests serve as vital carbon reservoirs, reducing atmospheric CO2 and mitigating climate change. Monitoring carbon stocks typically combines ground-based data with satellite remote sensing, yet accuracy remains a challenge. This study analyzes Huize County, China, using GF-1 WFV and Landsat TM images and introduces the Quantitative Remote Sensing Transformer (QRS-Trs), which leverages style transfer and attention mechanisms to enhance carbon stock estimation as an image-to-image translation task. QRS-Trs demonstrates three advantages: 1) Swin-Pix2Pix effectively reduces inter-domain discrepancies caused by sensor and lighting variations while excelling in de-clouding, outperforming Pix2Pix. 2) It incorporates a median filter to eliminate anomalies and a mask module to exclude non-target areas, achieving MAE =16.29 Mg/ha, RMSE =29.38 Mg/ha, <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$R^{2} =0.71$ </tex-math></inline-formula>, and SSIM =0.75. 3) Applied to multi-year data, from 2005 to 2020, 44.04% of the area showed increased carbon stock, 10.22% decreased, and 45.74% remained unchanged. While QRS-Trs performs well, its generalization to diverse ecological conditions depends on high-quality training data. Nevertheless, this study provides a robust approach for high-resolution carbon stock estimation, contributing to improved forest carbon sink management.