The inherent differences between observational topography and model terrain have seriously affected the verification accuracies of 2 m temperature levels. The traditional two-dimensional interpolation schemes are only able to ensure the forecasting elements and observational consistency in latitude and longitude locations of two-dimensional spaces, while ignoring the vertical direction consistency. This has the effect of the forecasting and observational verification results not originating from the same spatial positions, thereby causing misleading evaluations. The diurnal cycles are important features of the 2 m temperatures. However, due to the limitations of the physical processes(such as radiation), large bias have consistently appeared in the diurnal cycle forecasts. In this research study, three-dimensional forecast variables were combined with the near-surface elements of the forecasting products, and an advanced three-dimensional interpolation scheme was developed in order to ensure a consistency with the observations in the three-dimensional spatial forecasting processes. Then, based on topography correction methods, the monthly forecasting errors were used as reference bias products for the purpose of eliminating systematic errors and obtaining forecasting products with characteristic diurnal cycles. The abnormal datasets were rejected using a significance test in order to ensure the validity of the samples. In this study, using a classification analysis based on 27 typical observational gauges selected in the complex terrain of Shanxi Province, six major gauge stations were selected which were known to have different height biases between the model terrain and observational heights. The 48-hour forecasting products in August of 2016 were used for this study's comparison process. It was found that the three-dimensional interpolation scheme effectively solved the misleading evaluations caused by the height bias between the model terrain and observation topography, regardless of whether the large height bias gauge stations or small height bias gauge stations were examined. However, it was observed that the scheme had not effectively improved the diurnal cycle trends of 2 m temperature forecasting. Therefore, it was determined that the three-dimensional interpolation scheme could only modify the overall bias magnitude, and could not improve the forecasting abilities of the diurnal cycles. However, it was observed that after systematic error corrections were adopted, the diurnal cycle forecasting features had been obviously improved. In particular, it was found that a better consistency with the observations had been attained, as well as higher skill scores, particularly in the first 24 hours. The results of the seasonal statistical evaluation of the summer of 2016 indicated that, after the bias corrections, the 2 m temperatures could be effectively improve the oscillation of the periodic errors. Furthermore, the RMSE had been maintained at approximately 2 K, which indicated the obvious advantages of the improvements. This study focused on the effectiveness of the bias correction method, and was most concerned with the improvement trends. The systematic errors required monthly forecasting data for many years as reference errors, and the number of forecasting samples was found to restrict the bias correction effects to some extent. Therefore, it was concluded in this study that by increasing the forecasting samples, more reference samples could be added to ensure the error correction methods were perfected. In this way, the proposed bias correction effects could potentially be more significant in the future. At the same time, some of the related operational models have been running for long periods of time(such as the NCEPGFS, ECMWF, T639, and so on). A more ideal reference data base could be obtained using these long period forecasting products, which would potentially display superior effectiveness and applicability in 2 m temperature bias corrections in future studies.