Using satellite-derived attributes as proxies for soil carbon cycling to map carbon stocks in alpine grassland soils

• 3D mapping of SOC stocks from remote sensing. • The interacting roles of satellite-derived proxies on SOC were verified. • QRF showed better performance than PLS-SEM. • The capability of satellite-derived proxies to capture shadow subsurface variations in SOC was found. Alpine grassland ecosystems play a crucial role in the global carbon (C) balance by contributing to the soil organic carbon (SOC) pool; thus, quantifying SOC stocks in these ecosystems is essential for understanding potential gains or losses in soil C under the threat of climate change and anthropogenic activities. Remote sensing plays a vital role in estimating SOC stocks; however, identifying reliable remote sensing proxies to enhance SOC prediction remains a challenge. Information on soil C cycling proxies can reveal how the balance between C inputs and outputs affects SOC. Therefore, these proxies could be effective indicators of SOC variations. In this study, we explored the potential of satellite-derived attributes related to soil C cycling proxies for predicting SOC stocks. We derived remote sensing indices such as gross primary production, soil respiration, soil moisture, land surface temperature, radiation, and soil disturbance and assessed the relationships between these indices and SOC stocks via partial least squares structural equation modeling (PLS-SEM). We evaluated the effectiveness of these indices in predicting SOC stocks, we compared PLS-SEM and quantile regression forest (QRF) models across different variable combinations, including static, intra-annual, and inter-annual information. The PLS-SEM results demonstrated the suitability of the derived remote sensing indices and their interactions in reflecting processes related to soil C balance. The QRF models, using these indices, achieved promising prediction accuracies, with a coefficient of determination ( R 2 ) of 0.54 and a root mean square error (RMSE) of 0.79 kg m −2 at the topmost 10 cm of soil. However, the prediction performance generally decreased with increasing soil depth, up to 30 cm. The results also revealed that adding intra- and inter-annual information to remotely sensed proxies did not increase the prediction accuracy. Our study revealed that gross primary production, soil respiration, soil moisture, land surface temperature, radiation, and soil disturbance are effective proxies for representing factors influencing soil C balance and mapping SOC stocks in alpine grasslands.