Spatial regression with conditional autoregressive (CAR) errors for annual mean relative humidity in Peninsular Malaysia

Modelling observed meteorological elements can be useful. For instance, modelling rainfall has been an interest for many researchers. In a previous research, trend surface analysis was used and it was indicated that the residuals might spatially be correlated. When dealing with spatial data, any modelling technique should take spatial correlation into consideration. Hence, in this project, fitting of spatial regression models, with spatially correlated errors to the annual mean relative humidity observed in Peninsular Malaysia, is illustrated. The data used in this study comprised of the annual mean relative humidity for the year 2000-2004, observed at twenty principal meteorological stations distributed throughout Peninsular Malaysia. The modelling process was done using the S-plus Spatial Statistics Module. A total of twelve models were considered in this study and the selection of the model was based on the p-value. It was found that a possible appropriate model for the annual mean relative humidity should include an intercept and a term of the longitude as covariate, together with a conditional autoregressive error structure. The significance of the coefficient of the covariate and spatial parameter was established using the Likelihood Ratio Test. The usefulness of the proposed model is that it could be used to estimate the annual mean relative humidity at places where observations were not recorded and also for prediction. Some other potential models incorporating the latitude covariate have also been proposed as viable alternatives.

Actions (login required)

View Item