The Application of CDF-T in Bias Correction of Short-term and Medium-term Temperature Forecast Products of Numerical Weather Prediction Model

To improve the accuracy of short-term and medium-term temperature forecasts at stations and enhance the capability of predicting extreme temperatures, this study uses ground observation data from 11 national stations in Shanghai (2017–2024) and the dynamical downscaling system forecast products of the NCEP-CFSv2 global climate model. Based on the CDF-T bias correction method, the 1–14 day forecast temperatures from the NCEP-CFSv2 dynamical downscaling system are corrected, and the correction effects are evaluated with error analysis. Independent verification and evaluation for 2021–2024 show that, regarding extreme temperatures, the dynamical downscaling system exhibits the characteristic of underestimating maximum temperatures and overestimating minimum temperatures relative to observations. After applying the CDF-T method, the extremes of daily maximum and minimum temperatures align more closely with observations. The correction effects for maximum temperatures in spring and summer are more significant and remain stable over the 1–14 day forecast lead times, with the most significant effect in summer. For minimum temperatures, significant correction effects are observed in spring, autumn, and winter, remaining stable within the 1–14 day forecast lead times, with the most notable effect in winter. In terms of spatial distribution, the correction effects for summer daily maximum temperatures are better in eastern coastal regions than western regions, while those for winter daily minimum temperatures are