One of the models time series which also involves spatial aspects (spatio-temporal) is Generalized Space Time Autoregressive (GSTAR). Until now, GSTAR modelling don't involve metric-type, which is called GSTARX. Parameter estimation for spatio temporal modeling is still limited by using Ordinary Least Square (OLS) which is less efficient because the residuals are correlated. Generalized Least Square (GLS) is one of the alternative methods for parameter estimation residuals are correlated. In this study would like to looking at the efficiency of GLS estimation method is compared with OLS to correlated data in GSTARX model. Simulation results show that the estimation GLS method is more efficient than using OLS if residual correlated.