Xu, Shengxian; Zhou, Wentao; Liu, Xiangjun; Qiao, Guanyu; Geer, Chaoli; Liu, Changhe; Nie, Zhujun; Tang, Huaizhi; Huang, Yuanfang
Abstract
Mollic epipedon thickness (MET) is a critical indicator of soil quality and degradation, and characterizing its spatiotemporal dynamics is essential for conserving Mollisols. However, few studies have mapped MET over long-term periods, and the observed samples exhibit a zero-inflated distribution, which can markedly reduce predictive accuracy in traditional models because smoothing of zero values introduces substantial bias. To address the distributional characteristics of MET, this study developed a two-stage model combining random forest classification and regression. This model was built using 358 soil samples collected between 1985 and 2015. Initially, a classification model was used to determine the presence or absence of mollic epipedon. Then, for areas where mollic epipedon presence, a regression model was employed to estimate its thickness. The predicted results were used to generate MET distribution maps at five-year intervals from 1985 to 2020, and its inter-annual variability was analyzed. Our study showed that: The two-stage model achieved long-term prediction of MET in Northeast China (R2 = 0.61, LCCC = 0.77, RMSE = 18.40 cm, and MAE = 11.75 cm). Compared with the traditional model, the two-stage model increased R2 and LCCC by 17.31% and 26.23% respectively, while reducing RMSE and MAE by 15.28% and 32.28%. These results indicate that the two-stage model predicted MET more accurately than the traditional model. From 1985 to 2020, the mean MET ranged from 31.05 to 32.89 cm. Its inter-annual variability generally exhibited an initial declining trend, followed by an increase during the period of 2000-2005. Over the past 35 years, the average MET decreased by 0.7 cm. However, the thicker mollic epipedon in newly reclaimed cultivated land partly offset this overall declining trend. The area with declining MET accounted for 35.79%, exceeding the area with increasing MET, which accounted for 28.20%. Among the different regions, Sanjiang plain region exhibited the largest decline in MET, indicating the most severe degradation. This study provides methods and data for long-term mapping of MET, and offers a scientific basis for the development of soil conservation strategies in China's Mollisols region.
Spatiotemporal distribution of cultivated land mollic epipedon thickness in Northeast China from 1985 to 2020 based on a two-stage model.pdf