High resolution Bayesain Spatio-Temporal precipitation modelling in Pakistan for the appraisal of trends

Appraisal of spatio-temporal variability of precipitation is important for studying climate change and managing water resources. In this article, data on monsoon precipitation of Pakistan are analyzed for spatio-temporal interpolation using two hierarchical Bayesian spatio-temporal methods which are DLM and AR models. The precipitation records along with three meteorological covariates (temperature, wind speed and humidity) are collected from 53 stations for the years 2000-2008. Initially, we simulated four data sets using AR model with four levels of autoregressive parameter = (0.2, 0.5, 0.7, 0.9) and both models are fitted. It is observed that for the larger values of, DLM gives better results as compared to AR model while AR model performs better for the smaller values of. The estimated value of for precipitation data in the whole study domain is 0.02 which supports AR model for final spatio-temporal mapping. For real data application, data of 45 locations are used for modeling while data of 8 locations have been set aside for cross validation and we use hyper-parameters and of inverse gamma distribution as variance parameters for AR and DLM models. The prediction performances of both models have been examined using Bayesian and non-Bayesian model choice criteria. The predictive inferences are drawn using MCMC algorithm. Prediction plots show that the areas that lie between 32o-36o east latitude and 70o -74o north longitude are high precipitation areas in Pakistan and are helpful in the identification of homogeneous climate zones.