应用基于PLSR的土壤-环境模型预测土壤属性

PREDICTION OF SOIL PROPERTIES USING PLSR-BASED SOIL-ENVIRONMENT MODELS

土壤-环境模型对于正确理解土壤属性与环境因子间的关系,以及进行土壤属性预测与制图均具有重要的意义。研究区位于陕西省长武县内多年退耕还林还草沟壑区域,采集72个土壤表层样本,选择3/4的样本作为建模集,其余1/4的样本作为验证集;环境因子选择容易获取的地形因子和由遥感影像提取的植被因子和湿度因子,建立基于偏最小二乘回归(PLSR)的土壤-环境模型。结果表明:全氮、速效钾、全钾、有机质与环境因子间均有显著相关性;建立的PLSR模型可解释土壤属性的空间变异从23%(全氮)到27%(全钾);与逐步回归方法构建的模型相比,利用PLSR构建的土壤-环境模型可以更好地表征土壤属性与环境变量间的关系,拟合精度和预测精度也相对较高,说明PLSR建立的模型可以更好地应用于相似区域的土壤属性预测。

Soil-environment models can be of great importance to proper understanding of relationships between soil properties and environmental factors, and to predicting and mapping of soil properties, as well. A gully area where the ＂Grain for Green＂ policy had been implemented for years was selected in Changwu County, Shaanxi Province, a Loess Plateau region in China. A total of 72 surface soil samples were collected, and 3 fourths of the samples were used as a cal- ibration set of samples and the rest as a validation set. Several easily aquired environmental factors, such as topographic factor, vegetation index and wetness index were used in a PLSR （partial least squares regression）-based soil-environment model established for the study. Quantitative analysis of relationships between environmental factors and soil properties of the samples was done. Results show that soil properties, including available potassium, total potassium, organic matter and total nitrogen, were significantly correlated with environmental factors. The PLSR-based model could well explain 23% to 27% the spatial variability of soil properties. Compared with the stepwise regression model used, the PLSR model was much better at characterizing soil-environment relationships with better fitting and prediction accuracy, suggesting that the PLSR-based model is applicable to prediction of soil properties of similar regions.