两种模型对森林土壤中量元素空间分布预测

The Prediction of Forestry Soil Elements Spatial Distribution by Two Models

采用BP人工神经网络模型与空间插值模型中的泛克里格插值法、三次样条插值法、反距离插值法相结合,预测广东省云浮市云城区与云安区森林土壤中量元素含量的空间分布。结果表明:研究区中量元素交换钙的含量相对较高,为17.333~1 169.033 mg·kg^(-1),其次为全硫,含量为60.787~354.600mg·kg^(-1),交换镁含量最低,为8.320~51.580 mg·kg^(-1);从养分的变异系数来看,3种中量元素在整个研究区的变异系数在33.43%~106.34%之间,其中除云安区的交换钙变异系数为106.34%,属于强变异以外,其他两个元素均属于中等变异。3种插值方法中,泛克里格对交换镁的预测性较好,而对交换钙、全硫预测性欠佳。通过模型的对比发现,BP人工神经网络模型为最适模型。

The BP artificial neural network and the spatial interpolations model（Universal Kriging, Spline method and Inverse Distance Weighted） were adopted to predict the forest soil elements（exchangeable calcium, sulfer and exchangeable magnesium） spatial distribution in Yuncheng district and Yun＇an district in Yunfu city, Guangdong province. It was showed that the highest content element in the forestry soil was exchangeable Ca, which concentration was 17.333~1 169.033 mg · kg^（-1）, followed by total sulfur and exchangeable Mg, which concentration were 60.787~354.600 and 8.320~51.580 mg · kg^（-1） respectively. The coefficient of variation for the three elements were between 33.43%~106.34%. The variation coefficient of exchangeable Ca in Yun＇an district reached 106.34%, indicating that the concentration of exchangeable Ca had lavge variation. In the three kinds of interpolation methods, the Universal Kriging deviation was small than other methods. But the Universal Kriging was unable to predict other elements well in the three elements, except exchangeable Mg. By comparison, we can reach the conclusion that the BP artificial neural network was the best model to predict the element spatial distribution in this study.