Most of the current pollutant concentration prediction methods do not take into account the temporal and spatial characteristics of meteorological data, which results in a great loss of forecasting accuracy. To this end, this paper presents a spatio-temporal hybrid prediction method (GPR-EN), which can extract both temporal and spatial characteristics of process data and improve the prediction accuracy of pollutant concentration significantly. First, elastic net algorithm is used to comprehensively analyze the spatial correlation between the sample points and the target points, and reconstruct the spatiotemporal data, so as to provide the optimal variable input for the prediction model. Second, with the strong generalization ability of the Gaussian Process Regression model, the complex nonlinear characteristics of spatiotemporal data can be effectively handled, so as to obtain better prediction results. Finally, simulation experiments are carried out on AQI data set and air SO2 concentration data set, and the experimental results show that the prediction accuracy of the proposed method is more than 22% higher than that of the contrast methods.