Improving Near-Surface Temperature Forecasts in the NCEP Global Forecast System

Like precipitation, the accurate prediction of near-surface temperature is one of the essential components for numerical weather prediction, and is also considered a challenging task because of the multiplicity of physical processes and their complex interactions. It has long been known that the NCEP Global Forecast System (GFS) has large errors in the forecast of near-surface temperature in some seasons. In particular, large biases in late afternoon and nighttime 2-m temperatures usually happen in spring, autumn and winter seasons. This study focuses on improving near-surface air temperature forecasts under stable conditions in the GFS model. We identify the systematic deficiencies and cause of errors in near-surface temperature forecasts by investigating the physics of the Noah land surface model and land-atmosphere interactions, and find a practical solution to reduce these kind of forecasting errors. Sensitivity tests for case studies and a one-month experiment have been performed. The results demonstrate substantial reduction of errors in near-surface 2-m air temperature forecasts using the proposed modifications, and includes a notable reduction of bias and root-mean-square of temperature in the lower atmosphere.