The impact of El Nino Southern Oscillation on space time PM10 levels in Peninsular Malaysia: the hierarchical spatio-temporal autoregressive models approach

Hierarchical spatio-temporal autoregressive models are useful to understand the impact of predictors on a spatio-temporal-dependent variable. This study aims to fit the model to monthly PM10 concentration using potential predictors from 33 monitoring stations within Peninsular Malaysia from 2006 to 2015 and predict the space-time data spatially and temporally. Using Monte Carlo Markov Chain (MCMC), spatial predictions are obtained based on the posterior and predictive distributions of the model. The posterior distribution of the model that is without covariates exhibits a strong temporal correlation between successive months and also a strong spatial correlation with an effective range of 300 km. Spatio-temporal models were fitted to the data with a sine term, a cosine term, and a lagged El Nino Southern Oscillation (ENSO) index as predictors. Of the 33 monitoring sites, 8 were selected randomly for validation sets. The predictions and forecasts are validated using the root mean square error (RMSE), the mean absolute error (MAE), and the predictive model choice criteria (PMCC). The model with a sine term and a cosine term as predictors produces a reasonable RMSE, MAE, and PMCC of 7.23, 5.91, and 114.54, respectively. It is lower compared to those of the other models. The coverage percentage of the forecast 5-95 percentile range is 89.2% implying good prediction results. The results also show that none of the ENSO indices has a significant impact on the spatial distribution of the PM10 concentration.

Automated summary

Hierarchical spatio-temporal autoregressive models were used to study the PM10 concentration in Peninsular Malaysia. The results showed strong temporal correlation between successive months and spatial correlation with an effective range of 300 km. The inclusion of sine and cosine terms as predictors improved the prediction results.