• An extensive soil survey were conducted across the Taklimakan Desert, China’s largest desert. • A high-resolution map of soil inorganic carbon (SIC) was produced for the Taklimakan Desert. • Integrating VIP and RF provides the most accurate estimation of SIC. • Remote sensing predictors contributed 81.4% of the total explanatory power. • Moisture relevant spectral indices and pH were key contributors to desert SIC estimation. Deserts cover ∼10% of Earth’s land area and store substantial soil inorganic carbon (SIC). The Taklimakan Desert (3.28 × 10 5 km 2 ) is China’s largest desert, yet the lack of SIC maps limits understanding of its role in the global carbon pool and carbon cycle. To address this gap, we collected 347 surface soil samples (0–30 cm) across the Taklimakan Desert during August–November 2024. Remote-sensing variables derived from Landsat 8 imagery, together with environmental covariates, were used as predictors. Six feature-selection algorithms (Variable Importance in Projection, Mutual Information, Recursive Feature Elimination, Particle Swarm Optimization, Least Absolute Shrinkage and Selection Operator, and Elastic Net) were applied to identify sensitive factors, and three models (Random Forest, Partial Least Squares Regression, and Convolutional Neural Network) were evaluated to produce the first 30 m SIC map for the Taklimakan Desert. Across 18 model–feature-selection schemes (6 feature selectors × 3 models), VIP + RF achieved the best performance (R 2 = 0.62, RMSE = 0.79 g kg −1 , MAE = 0.43 g kg −1 ). Remote-sensing variables contributed 81.40% of cumulative importance, while environmental variables accounted for 18.6%; Normalized Difference Drought Index, Normalized Difference Moisture Index, and pH were the most influential predictors. The estimated mean SIC across the desert was 13.39 g kg −1 . The mean pixel-wise standard deviation across repeated RF predictions was 0.27 g kg −1 . This study provides a high-resolution baseline for SIC spatial patterns in the Taklimakan Desert and supports improved assessments of regional-to-global carbon stocks.
Yang et al. (Thu,) studied this question.