Abstract:

The karst region in eastern Yunnan, China, which is known for its fragile ecosystem, suffers from severe desertification, soil erosion, and landscape fragmentation. This complex background significantly affects soil properties and carbon cycling, and causes difficulties in assessing soil organic carbon (SOC) stocks. Identifying the spatial distribution of SOC and its affecting factors, thus improving the karst SOC sink, is of high significance for achieving carbon neutrality and reducing global warming. In this study, the characteristics of SOC and its affecting factors were examined using geostatistical methods, the Random Forest Model (RF), and the SHapley Additive exPlanations (SHAP) method based on soil survey data. The findings were as follows: 1) SOC in the surface layer varied significantly across the region, ranging from 1.45 to 56.0 g∙kg⁻¹, and the coefficient variation (CV) was approximately 46.7%. 2) Significant differences in SOC existed among different land use types and soil types, and the surface SOC decreased and then increased as elevation increased. 3) Moderate autocorrelation (nugget coefficient: 48.4%) was observed in the spatial distribution of SOC. Total nitrogen (TN), total phosphorus (TP), precipitation, altitude, and pH were the primary influencing factors. Soil TN and TP, as the key factors, explained approximately 88.5% of the spatial variation in SOC. 4) A threshold or peak effect was observed in the impact of key factors on the SOC. For instance, the effect of TN on SOC changed from negative to positive when TN exceeded 1.75 g∙kg⁻¹. The positive effect of TP reached its peak value at 1.50 g∙kg⁻¹. This implies that excessive phosphorus fertilizer input may lead to resource waste rather than improve soil fertility. Altitude also played a significant role in SOC, with the peak value at 1750 m for positive impact. Similarly, the pH had a peak positive effect of approximately 4.45. These findings on the threshold or peak effects of key factors are especially valuable for understanding the complex mechanisms of SOC accumulation in karst regions.

Key words: karst fault basin, soil organic carbon (SOC), spatial variation, random forest model, SHapley Additive exPlanations (SHAP)

摘要：

滇东喀斯特地区作为典型的生态脆弱区，土地石漠化和水土流失严重、景观空间异质性高，这些因素强烈影响着土壤性质和碳循环过程，为评估该地区土壤有机碳储量动态变化带来巨大困难。通过探究喀斯特地区土壤有机碳空间分布特征及其关键影响因素，进而提升喀斯特土壤碳汇能力，对实现碳中和目标以及减缓全球气候变暖具有重要意义。基于滇东喀斯特地区典型区域土壤调查数据，利用地统计学方法、随机森林模型（RF）和SHAP解释方法（SHAP），分析了土壤有机碳空间分布特征和关键影响因子及其影响效应。结果表明，1）研究区表层土壤有机碳质量分数为1.45－56.0 g∙kg⁻¹，变异系数为46.7%，空间变异较大。2）不同土地利用类型、土壤类型之间土壤有机碳含量差异显著（p<0.010），随着海拔上升，表层土壤有机碳含量呈现先下降后上升的趋势。3）土壤有机碳空间分布具有中等程度的空间自相关性（块金系数：48.4%），空间变异主要由总氮、总磷、降水量、海拔和pH主导，土壤氮、磷等关键因子解释了土壤有机碳88.5%的空间变异。4）关键影响因子对土壤有机碳的影响存在阈值或峰值效应。当总氮超过阈值点1.75 g∙kg⁻¹后，其对土壤有机碳的影响由负转正；总磷对土壤有机碳的正向贡献在1.50 g∙kg⁻¹左右时达到峰值，侧面证明了过度施磷肥并不能起到增加土壤肥力的作用，而可能会造成资源浪费和土地生态污染问题；海拔对土壤有机碳的正向贡献在1750 m左右时达到最大值；pH对土壤有机碳的正向贡献在4.45左右达到峰值。研究表明需要考虑关键因子的阈值或峰值效应，以便了解土壤有机碳积累的潜在过程。

关键词: 喀斯特断陷盆地, 土壤有机碳, 空间变异, 随机森林模型, SHAP