ABSTRACT Accurate estimation of the spatiotemporal variation of soil organic matter (SOM) is essential for effective soil management and carbon budget assessment. However, in perennially vegetated subtropical regions, frequent cloud and fog cover, along with the lack of remote sensing data during bare soil periods, pose significant challenges for SOM mapping. Taking Yingtan City, southeastern China, as a case study, this study proposes a novel framework to map the spatiotemporal distribution of SOM for the years 2003 and 2021 by integrating multi‐year vegetation phenological information and soil particle size fractions (sand, silt, clay) using convolutional neural network (CNN) models. Assuming that soil particle size fractions remain temporally stable, CNN‐based models were first developed to predict these fractions from soil samples and vegetation phenological features across different periods. The predicted particle size fractions, together with vegetation phenological features, were then incorporated into CNN models to predict SOM in each time period. To evaluate model performance, we compared three algorithms—CNN, random forest (RF), and partial least squares regression (PLSR)—as well as different input feature combinations (with and without vegetation phenology and predicted particle size fractions). Finally, the spatiotemporal variation of SOM was analysed across the study area. The results demonstrate the effectiveness of the proposed framework, with coefficient of determination (R 2 ) values exceeding 0.52 for SOM and soil particle size fractions (except clay content). CNN consistently outperformed RF and PLSR, with improvements in R 2 ranging from 5% to 88%. Incorporating predicted soil particle size and vegetation phenological features improved SOM prediction accuracy, increasing R 2 by 18% to 41%, with phenological features showing greater importance. Over the past two decades, SOM in paddy fields exhibited an increasing trend, while a slight decrease was observed in forest areas, particularly downslope near water bodies. This study provides a valuable approach for large‐scale SOM mapping and offers insights into SOM dynamics in perennially vegetated subtropical regions.
Ma et al. (Wed,) studied this question.