Application of artificial neural network to screen out the dominant meteorological parameters for prediction of air temperature

Precise weather forecasting is one of the most significant challenges in the modern world. The significant meteorological parameter, air temperature (TP), is often measured with limited spatial resolution, which necessitates the prediction of them at places far from monitoring stations and in the cases of accidentally missing data from the data loggers of monitoring stations. This study supports a non-location-specific model for air temperature prediction by combining a three-layer backpropagation artificial neural network (ANN) and meteorological data. The predictor identification method (PIM) & cross-validation method incorporated with the ANN model reveals density altitude (DA), heat index (HI), relative humidity (RH), and wet bulb temperature (WB) as potential input variables for the prediction of TP. DA and HI are strongly correlated to TP for the whole year for all types of land covers, whereas the dependency of TP on RH varies seasonally. RH is always a topping variable for air temperature prediction, which undoubtedly enhances the prediction accuracy. In pre-monsoon and monsoon, there are only three dominant input variables for predicting air temperature, i.e., DA, HI, and RH. In the post-monsoon, WB comes into the role of an additional predictor. The temperature prediction model shows a good agreement between ANN-estimated air temperature and measured air temperature, with the coefficient of determination value of 99.46%, 99.23%, 99.73%, and 99.18% for pre-monsoon, monsoon, post-monsoon, and winter season, respectively using the predictor set of potential input variables. Therefore, in this study, ANN modeling emerged as a reliable method for air temperature prediction.