Application of a dynamic vertical change rate downscaling method in gridded temperature forecast

To improve spatially fine characteristics and accuracy of objective gridded temperature forecast, a dynamic vertical change rate downscaling method (DRD) considering elevation is proposed. Real-time vertical changes of surface air temperature with elevation (VCE) are calculated using the relationship between surface air temperature forecast and elevation at different grid points in the numerical model, and the results are applied to downscaling forecasts at target grids and stations to generate a more accurate initial background forecast field. Based on ECMWF model forecasts, 5 km resolution gridded elevation information, observations collected at 10154 stations and their geographic information in China, forecast experiments for spring, summer, autumn and winter are carried out. Spatial and temporal distribution characteristics of VCE, the accuracy of DRD temperature forecast and its ability to depict spatially fine characteristics are analyzed. The results show obvious diurnal, seasonal and spatial variations of VCE over China corresponding to diurnal and seasonal variations of surface long wave radiation, thermal properties and topographic dynamic effects. The value of VCE usually reaches the largest in the morning and the smallest in the evening. This means that surface air temperature decreases the slowest or increases the fastest with increasing height in the morning, and decreases the fastest or increases the slowest in the evening. Spatial variability of VCE is the largest in winter and the smallest in summer. The VCE value is closely related to the distributions of topography, land-sea, and inland lakes. Large VCE values usually appear over large topography edges, daytime land-sea margins, and inland lake edges in the daytime in spring and the whole day in summer. VCE often varies more greatly in complex terrain areas. The DRD surface air temperature forecast is significantly better than the bilinear-interpolated value of model prediction (DMO), especially over complex terrain areas. For example, the mean absolute error (MAE) of DRD forecast is about 14.3%—52.5% smaller than that of DMO over the southern Qinghai-Tibet Plateau in spring. At the same time, DRD significantly improves the ability of describing the spatially fine characteristics of surface air temperature. Overall, it is concluded that the DRD method can effectively improve the performance of objective gridded temperature forecast.