土壤侵蚀降低土地生产力, 导致土壤环境恶化, 其中水力侵蚀是土壤侵蚀中最主要的一种形式。 土壤可侵蚀性K值是评价土壤被降雨侵蚀难易程度的一项重要指标。 使用河南、 福建和浙江三省研磨干样可见-近红外(Vis-NIR)漫反射光谱数据, 将其转换为吸收率后进行Savitzky-Golay(SG)平滑去噪； 对土壤有机质(SOM)和机械组成进行精准预测后, 分别采用EPIC和RUSLE2模型估算K值, 并对预测精度进行比较分析, 所得结论如下: (1)建立土壤有机质和机械组成高光谱最佳预测模型, 土壤质地(砂粒、 粉粒和黏粒)预测采用支持向量机(SVM)模型, SOM预测采用局部加权回归(LWR)模型, 模型四分位相对预测误差(RPIQ)为2.27, 3.17, 2.18和3.44； (2)通过土壤质地估算的土壤渗透性等级分类效果较好, Kappa系数为0.62, 同时估测的土壤质地类型与实测土壤质地类型分布特征相近, 质地主要类型均是粉黏土、 砂黏壤土、 壤土、 壤砂土和砂壤土； (3)EPIC和RUSLE2两种模型均具有较为精确的估测能力, EPIC模型预测精度更高, 均方根误差(RMSEP)为0.006 6 (t·ha·h)/(ha·MJ·mm), RPIQ达1.58, 而RUSLE2模型精度相对较低(其中RPIQ为1.43), 因此推荐使用EPIC模型结合Vis-NIR光谱技术估测土壤可侵蚀性K值。 本研究为今后快速准确预测K值提供思路, 并为大面积监测土壤侵蚀提供辅助手段。

Soil erosion reduces the productivity of the soil, leading to the deterioration of soil environment. Water erosion is one of the most important forms of soil erosion. Soil erodibility K value is an important indicator to evaluate soil susceptibility to erosion, the aim of this paper is to evaluate whether Vis-NIR can be used in predicting K value as a rapid method. Soil samples were sampled from Henan, Fujian and Zhejiang provinces, after air-drying and grinding, visible-near infrared (Vis-NIR) diffuse reflectance spectra were measured. Then, soil reflectance spectra were transformed to absorbance spectra and Savitzky-Golay (SG) algorithm was used to eliminate noise. Data mining methods were used to predict soil organic matter (SOM) and soil texture with Vis-NIR spectra, then K values were estimated with EPIC and RUSLE2 models based on predicted SOM and soil texture. The results were as follows: (1) The prediction models with the highest performance were obtained about the SOM and soil texture (sand, silt and clay), the best model for soil texture prediction gained from support vector machine (SVM) model and the best SOM result was performed using locally weighted regression (LWR) model, of which the ratio of performance to inter-quartile distance (RPIQ) was 2.27, 3.17, 2.18 and 3.44 for sand. Silt, clay and SOM. (2) Based on predicted soil texture, the classification accuracy for grade of soil permeability was good (Kappa coefficient was 0.62), and the spatial distribution between predicted values and measured values was similar in soil texture triangle, of which the main types were silty clay, sandy clay loam, loam, loamy sand and sandy loam. (3) The EPIC and RUSLE2 models both had the accurate prediction ability. EPIC model performed better than RUSLE2 model, of which root mean square error of prediction (RMSEP) was 0.006 6 (t·ha·h)/(ha·MJ·mm) and RPIQ reached 1.58, while the accuracy of RUSLE2 model was lower (RPIQ is 1.43). Therefore EPIC model was recommended to estimate K values in combination with Vis-NIR spectroscopic technique. This study presents the potential for estimating soil erodibility K values using Vis-NIR spectroscopy, which provides supplementary method for monitoring soil erosion in large area.