以云浮市云城区和云安区森林土壤为研究对象，应用传统统计学和地统计学方法结合 GIS 技术分析其土壤养分的空间变异性，预测土壤养分空间分布。地统计学主要从空间插值模型和人工神经网络模型的角度对土壤养分的空间变异性进行诠释，并利用平均绝对误差(MAE)和均方根误差(RMSE)两个指标以及模型预测点与实测点间的相关系数作为判断模型好坏的标准。研究结果认为插值模型中泛克里格插值法在样点密度较小时显示出明显优势，而 BP-ANN 在模型的稳定性和推广性表现尤为突出，最后对泛克里格插值模型和 BP-ANN 下的有机碳、全氮、全磷、全钾4种养分空间预测分布特征进行描述。%This research analyzed the spatial variability of forest soil nutrients and forecasted forest soil nutrient distribution in Yuncheng district and Yun’an district. Geostatistics mainly described the spatial variability of soil nutrients in the perspective of the spatial interpolation model and back-propagation artificial neural network model. Besides, the root mean square error(RMSE) and the mean absolute error(MAE) were also used to assess the model accuracy. The model with higher correlation coefficients between prediction data and reference and lower RMSE and MAE was considered to be a successful model. Study results suggest that the Universal Kriging model displayed a clear advantage while the sample density was small. However, back propagation artificial neural network model’s stability and promotional ability was much better than other interpolation models. The spatial distribution characteristics of carbon, nitrogen, phosphorus and potassium that predicted by Universal Kriging and back propagation artificial neural network were described. Both of the UK model and BP- ANN model prediction distribution of four kinds of nutrients were described.
展开▼
