| 引用本文: | 邢贞相,张涵,付强,等.考虑气象因子不确定性的概率降水预报研究[J].灌溉排水学报,2018,37(10):100-107. |
| XING Zhenxiang,ZHANG Han,FU Qiang,et al.考虑气象因子不确定性的概率降水预报研究[J].灌溉排水学报,2018,37(10):100-107. |
|
| 摘要: |
| 降水量与气象因子的变化密切相关，因而气象因子的不确定性必然影响着降水量预报的精度。【目的】获取气象因子不确定性特征，提高降水预报结果的精度。【方法】以三江平原典型农场—友谊农场为研究区域，在贝叶斯概率预报系统（BFS）的理论框架下，利用径向基人工神经网络（RBF-ANN）描述月降水主要影响因子与月降水量的映射关系，并将其作为BFS的似然函数，以实测降水量为后验信息，对降水主要影响因子先验信息进行贝叶斯修正，利用基于自适应算法的马尔可夫链蒙特卡罗随机模拟技术（AM-MCMC）获取了各月的蒸发量、平均气温、相对温度的后验密度。结合RBF-ANN构建了考虑主要影响因子不确定性的概率降水预报模型（IFU-PBF），研究了各月降水的均值预报过程及其指定概率的置信区间。【结果】与传统RBF-ANN的确定性预报结果相比，预报期IFU-PBF计算结果的各精度评价参数均有所提高，纳什效率系数提高了3%；均方根误差降低了51%；相关系数提高了2%。尤其对极值降水的适用性更好，极大值的预报相对误差平均提高了55%，极小值的预报相对误差提高了24%。【结论】考虑气象因子不确定性，开展降水的概率预报更符合降水及其影响因素的随机过程的本质，提高预报精度的同时，能够考虑预报结果的不确定度。 |
| 关键词: 叶斯预报系统； RBF-ANN； 气象因子； 不确定性； 概率降水预报 |
| DOI：10.13522/j.cnki.ggps.20170058 |
| 分类号: |
| 基金项目: |
|
| Predicting the Probability of Precipitation with Meteorological Uncertainties in Consideration |
|
XING Zhenxiang, ZHANG Han, FU Qiang, GONG Xinglong, LI Heng
|
|
1. School of Water Conservancy and Civil Engineering, Northeast Agricultural University, Harbin 150030, China;2. The Key Laboratory of Efficient Utilization of Agricultural Water Resources of Ministry of Agriculture, Harbin 150030, China
|
| Abstract: |
| 【Objective】The amount of precipitation is closely related to the change in meteorological factors, while the uncertainty of the meteorological factors affects the accuracy of precipitation forecast. This paper analyzed how to improve the accuracy of precipitation prediction considering uncertainty of the meteorological factors. 【Method】 Based on the Bayesian probabilistic forecasting system (BFS), a radial artificial neural network (RBF-ANN) was proposed to describe the relationship between the monthly precipitation and the main meteorological factors including evaporation, mean temperature and relative temperature for Youyi Farm in Sanjiang Plain. The above relationship was taken as a likelihood function of BFS, and the posterior information based on observed rainfall was used in the Bayesian modification of the pre-information of the main influencing factors. The posterior density of evaporation, mean temperature and relative temperature for each month was then obtained by the adaptive sampling algorithm in the Markov Chain Monte Carlo simulation method (AM-MCMC), from which the probability precipitation prediction model (IFU-PBF) was constructed. Using the IFU-PBF, we forecasted the mean precipitation and its confidence interval with specified probability. The uncertainty of the predicted precipitation was calculated.【Result】 Compared the deterministic prediction model RBF-ANN, the IFU-PBF improved the forecasted accuracy with the Nash efficiency coefficient increased by 3%, the mean square error reduced by 51% and the correlation coefficient increased by 2%. For predicting extreme precipitation, the relative error of the predicted the maximum value and minimum value using the proposed model increased by 55% and 24% respectively. 【Conclusion】 The IFU-PBF substantially improves prediction accuracy of the precipitation with the uncertainty of meteorological factors in consideration. The results provide a basis for water resource management and utilization. The probabilistic forecast is more appropriate by considering the stochastic nature of the precipitation and its associated factors. |
| Key words: Bayesian Forecasting System; RBF-ANN; meteorological factors; uncertainty; probabilistic precipitation forecasts |
