HM-Bayes网络降水空间插值模型的改进及其应用

Improvement of the HM-Bayes networks based precipitation interpolation model and application

基于隐马氏性分析的贝叶斯(HM-Bayes)网络降水空间插值模型是一种空间异相关模型,具备统计学背景强和插值精度高的特点,但该模型未充分考虑典型自然地理要素对降水过程的影响.本文根据降水过程与地表特征的相互作用,建立降水与高程、坡度、坡向的空间分布关联,对HM-:Bayes网络降水插值模型进行了改进,并以无定河流域264:个月32个水文站点的降水记录为例,检测模型精度.结果表明,改进后模型的均方根误差和相对误差的均值分别为24.25毫米/月和0.24,比改进前模型分别降低了23.62%和33.43%.因此,改进后模型插值精度高,提升效果明显,尤其在地形差异大、站点稀疏的地区,精度增幅大.同时,该模型参数物理意义明确,适用性广,运算效率高,灵活性强,具备在地形差异大和无资料地区推广应用的潜力.

The Hidden Markov-Bayes networks （HM-Bayes） based precipitation interpolation method has the strong statistic basis and high precision, but it fails to take the full consideration of the impacts of physical geographical elements on precipitation interpolation. Based on the interactions between precipitation processes and land surface features, this paper improved the HM-Bayes based interpolation model by building up the spatial correlations between precipitation and elevation, slope and aspect, and took the monthly precipitation of 32 hydrologic stations of the Wuding River basin in 264 months as the example to test its performance. Results indicate that the root mean square error and relative error of the improved model are 24.25mm/month and 0.24, which decrease by 23.62／% and 33.43／% relative to the original model. Therefore, the improved model is highly accurate and its accuracy increases significantly. In particular, in the areas with big terrain differences and sparse stations, the accuracy improvement is quite obvious. Meanwhile, this model has the advantages of explicit physics meaning, extensive feasibility, high calculation efficiency, and great flexibility, therefore it has great potentials to be extended to the largely terrain different and ungauged basins.