Forecasting techniques for solar and wind energy are essential for controlling their variability and are being heavily researched. However, regional differences in predictability in these forecast models have rarely been studied. Regional differences refer to differences in forecast performance based on location. In this study, regional differences in time-series forecast models often used for short-term forecasts were quantitatively analyzed. A three-step methodology was devised to extract significant features for forecasting performance and develop a model to estimate forecast performance. (1) Multiple forecasting models, including machine learning, were applied to 100 random sites to calculate performance. (2) The time-series features of each site were statistically extracted and compared with forecast performance. (3) A formula-based modefl was developed to estimate forecast performance based on highly correlated metrics. The standard deviations of the rolling mean and those of the rolling standard deviations were significant metrics among the rolling statistics useful for interpreting time-series data and showed a correlation of >0.8 with regional forecast performance for both solar and wind. These findings can reduce the uncertainty of forecast model applications and minimize the risk of power system operations.