Abstract:Using the 3 h observation data from national-level stations in Beijing Region in spring (February to April) during 2019-2021, we evaluate the temperature forecasts made by numerical models including European Centre for Medium-Range Weather Forecasting (ECMWF) and Global and Regional Assimilation and Prediction System (CMA-GFS), and by national gridded guidance forecast product (SCMOC) as well as the revised feedback gridded forecast product at provincial level (SMERGE). The results show that the spring temperature forecasts in Beijing by the models of ECMWF and CMA-GFS often show negative biases. The biases have no significant difference in mountainous and plain areas, but more prominent during nighttime periods. The gridded forecast products (SCMOC and SMERGE) have a good ability to correct the temperature forecasted by the models (ECMWF and CMA-GFS). The temperature forecast biases of the gridded products are concentrated in the range of -1-1℃, meanwhile the forecast accuracy is higher and the mean absolute error is lower than that of model forecast. Some problems are found in the forecasts of 24 h temperature change and diurnal temperature range from the four products. The amplitude of intense 24 h temperature change forecasted by all products is relatively smaller than that of observation, and the gridded forecast products fail to demonstrate significant correction ability. In addition, the diurnal temperature range forecasted by all products has a positive deviation of 1-3℃ compared to the observation. SCMOC has a better correction ability for the diurnal temperature range by the models, while SMERGE overestimates the difference more prominently than that by models. The positive deviation of diurnal temperature range forecast is closely related to the underestimation of low temperature at 05:00 BT in the models, while the overestimation of high temperature at 14:00 BT can not be ignored in the grid forecasts. The refined analysis suggests that gridded forecast products should not only focus on improving overall accuracy (reducing biases), but also on the development and evolution of synoptic processes.