GBGWR模型在降水量空间分布模拟中的应用

Application of GBGWR Model in Spatial Distribution Simulation of Precipitation

近年来,我国大部分地区屡遭洪涝与干旱两种自然灾害侵袭,对重洪涝干旱区域进行空间插值具有重要的意义。针对传统地理加权回归(GWR)模型建模过程中模型识别和参数估计易受观测值异常点影响的问题,本文提出了一种基于吉布斯采样的贝叶斯地理加权回归(GBGWR)方法。运用基于吉布斯采样的马尔可夫链蒙特卡罗贝叶斯方法,估计地理加权回归模型参数,通过平滑函数降低观测值中异常点位数据,最后对湖南省1985—2015年35个观测站点的降水观测数据进行了空间分布模拟。试验结果表明,本文提出的方法相较于GWR模型性能提高了19.8%,相较于BGWR模型性能提高了8.2%,该方法可以有效降低异常值和"弱数据"对回归结果的影响,能够更加真实地模拟湖南省降水量的空间分布。

In recent years,most parts of China were repeatedly attacked by natural disasters, such as floods and droughts.Therefore,it is of great significance to carry out spatial modeling of heavy flood and arid areas. Aiming at the problem that the model recognition and parameter estimation of the geographic weighted regression （GWR） model are easy to be affected by the observed anomalies, a Bayesian geographic weighted regression （ GBGWR ） method based on Gibbs sampling is developed to solve the problem. The Markov Chain Monte Carlo Bayesian method based on Gibbs sampling is used to estimate the parameters of the GWR model, and the anomalous data of the observed values are reduced by the smoothing function.Finally, the precipitation data of 35 observation sites in Hunan province from 1985 to 2015 were simulated.The experimental results show that the performance of this method is 19.8% higher than the GWR model, and is 8.2% higher than the BGWR model. It is proved that the method can effectively reduce the influence of the abnormal value and the ＂weak data＂ on the regression resuh.More realistic simulation of the spatial distribution of precipitation in Hunan Province.