黄土高原县域土壤养分空间变异特征及预测——以陕西省横山县为例

SPATIAL VARIABILITY AND PREDICTION OF SOIL NUTRIENTS ON A COUNTY SCALE ON THE LOESS PLATEAU—A CASE STUDY OF HENGSHAN COUNTY,SHAANXI PROVINCE

研究土壤属性空间变异及其分布特征与环境因子的关系，对于了解生态系统的过程具有重要意义。以横山县为例，采集了254个耕层（0～20cm）土样，利用数字地形与遥感影像分析技术，提取相关地形与遥感指数，分析不同土地利用、地形条件下土壤养分空间变异及分布特征，并结合回归分析与地统计学进行空间分布预测。结果表明，不同土地利用类型其养分含量差异显著，水田和川地的有机质和全氮含量明显高于其他土地利用类型，而全磷含量以梯田最高。不同坡度分析表明，“0～3°”坡度等级有机质和全氮含量显著高于其他坡度等级；不同坡向土壤养分含量差异均不显著，但存在一个明显的趋势，即阴坡有机质和全氮含量整体上要较阳坡高。土壤有机质与高程H呈现负相关，与坡向的余弦值cosα、复合地形指数CTI、修正后的土壤调节植被指数MSAVI及湿度指数WI呈正相关。土壤全氮与相关环境因子的关系基本与有机质的一致。而土壤全磷与修正后的土壤调节植被指数MSAVI及湿度指数WI正相关。土壤有机质和全氮用相关环境变量的多元线性逐步回归模型拟合预测较好，而对于全磷，预测结果较差；回归一克里格预测有效地减小了残差，消除了平滑效应，与实测值较为接近，预测精度高于多元线性逐步回归预测。

Being the most important determinants of soil quality, soil properties significantly influence land use and ecological processes. Study on spatial variability of soil properties is vital for sustainable land management. Samples of surface soils （0 - 20 cm） collected from 254 sampling sites all over Hengshan County, on the Loess Plateau were analyzed to study spatial variation of soil nutrients （include soil organic matter （SOM）, total nitrogen （TN） and total phosphorus （TP）） with land use types and topography conditions. Correlation analyses were carried out of soil nutrients with terrain attributes and remote sensing indices. Finally, environment indicators were used to predict soil nutrients spatial distribution by multiple-linear regression analysis and geo-statistics. Significant differences were found between different land use types in soil nutrients, with the highest values in SOM and TN measured in soils from paddy field, and the highest value of TP in soils from terrace farmland. Fields with slope gradients ranging in 0 - 3 were significantly higher than fields with higher slope gradient in SOM and TN. And little difference was found in soil nutrients between fields different in slope aspect, but a tendency was discovered that SOM and TN in fields on northern slopes was higher. Different correlations were found of soil nutrients with terrain attributes and remote sensing indices. The multivariate linear stepwise regression model was relatively precise for SOM and TN, but for TP, it was not so good. Such techniques may be applied as a first step in unmapped areas to guide soil sampling and model development. The regression-kriging method can effectively reduce residuals in prediction by eliminating smoothing effect. So its predicted values are quite close to the measuresd, demonstrating that the regression-kriging method improves accuracy of prediction.