The objective of this project is to evaluate the accuracy of weather prediction in the U.S. from 2014 to 2017. Our dataset contains weather forecast and historical records, such as the predicted and real minimum daily temperature, locations, and the dates that the forecast was made on. We are interested in the spatio-temporal effects on the performance of weather forecasts, such as the effects of time and geographical locations. Performing the Functional Principal Component Analysis (FPCA) on the 50 states in the U.S., we visually detect the spatial correlation of the prediction error among various states and capture the main pattern of the variance in the prediction error over time. Then, we will conduct the spatial analysis on the obtained PC scores of each state to further investigate the geographical correlation. Furthermore, we will regress multivariate covariates on the functional responses on the basis that all the responses and covariates will be considered as objects with a function of time.