Spatial Rainfall Prediction Based on PGD-MRF Hybrid Model

Abstract:

It is of great significance for decision making in water resources planning and management to predict climate variation, especially rainfall, accurately. The main goal of this study is to put forward a PGD-MRF hybrid model for monthly rainfall forecast, which is based on Poisson Gamma Distribution (PGD) and Markov Random Field(MRF) models, to make up for the deficiency of the atmospheric general circulation model (GCM) in low spatial resolution and difficultly simulating regional climate change. The Wei River Basin was taken as an case study to investigate the accuracy of PGD-MRF hybrid model. Based on the monthly rainfall data from 1960 to 2010 at eight meteorological stations, the PGD model was firstly set up to fit the statistical relationship between monthly precipitation and GCM output factors. Then the monthly rainfall data in historical period of 1960-2010 were simulated through the spatial correlation of monthly rainfall analyzed by MRF model. The statistical downscaling model (SDSM) was also employed to evaluate the performance of the PGD-MRF model. The comparison of results revealed that the PGD-MRF model had provided a superior alternative to SDSM for forecasting monthly rainfall at all these eight meteorological stations. To further illustrate the stability and representativeness of the PGD-MRF model, the monthly rainfall data from 2001 to 2010 at Huashan station were used to verify the model. The results showed that the PGD-MRF model had a good stability and great representativeness as well as a high prediction precision.