海涂围垦区土壤盐分空间变异模拟的比较研究

以苏北海涂围垦区为例,利用人工神经网络(ANN)、普通克里格(OK)插值和序贯高斯模拟(SGS)对典型地块土壤盐分空间分布进行了模拟、插值与预测,获取了各方法的优化结构与参数,并就各方法对土壤盐分分布特征与空间结构的预测能力进行了比较分析。结果表明:ANN、OK和SGS法均较好地模拟和预测了土壤盐分的空间分布,达到了较高模拟、插值与预测精度;ANN获得的土壤盐分空间分布最为连续,SGS法整体分布相对离散;ANN能较好地预测盐分较低的样点,但ANN对高盐分样点的预测结果不如SGS和OK;SGS预测结果最符合实测值的波动特点,ANN预测结果波动范围最窄,SGS较ANN和OK更能反应数据随机变量的结构性和波动性,在整体上要优于ANN和OK法。该结果为滨海地区盐渍土壤的精准评估与高效改良提供了参考依据。

Comparative study of simulations of spatial variability of soil salinity in coastal polders

Artificial neural network (ANN), ordinary kriging (OK) and sequential Gaussian simulation (SGS) was introduced separately to simulation, interpolation and prediction of spatial distribution of soil salinity in a field typical of the coastal polder in North Jiangsu, to work out optimal structures and parameters of various methods for comparison of the methods in efficiency of predicting distribution characteristics and spatial structure of soil salinity. Results show that all the three methods, ANN, OK and SGS, were quite good at simulating and predicting spatial distribution of soil salinity and displayed quite high accuracy in the simulation, interpolation and prediction. The spatial distribution obtained by the ANN method was the most continuous and smooth, while that obtained by the SGS method was relatively discrete and fluctuant. The ANN method exhibited a relatively high prediction accuracy at sites of low soil salinity, but a much lower accuracy than the SGS and OK did at sites of high soil salinity. Furthermore, the prediction of SGS tallied the most with the fluctuation trait of measured value, and the narrowest fluctuation range was observed in prediction using the ANN method. The SGS method was better than the ANN and OK methods at reflecting spatial structure and fluctuation of the random variables of data, indicating that SGS is superior to ANN and OK as a whole. The findings may be cited as reference for precision assessment and high-efficiency amelioration of saline soil in coastal polders.