Postprocessing Model-Predicted Rainfall Fields in the Spectral Domain Using Phase Information from Radar Observations

Abstract

Abstract In an attempt to combine the short-term skill of radar nowcasting and the long-term skill of numerical models, successive 15-min rainfall accumulations obtained from the U.S. national radar composites and predicted by the Weather Research and Forecasting (WRF) model are corrected by adjusting the phases of Fourier components in the spectral domain while keeping the amplitudes invariant. The phase information is first obtained by decomposing the radar and model fields into Fourier space. Then, to correct the model outputs, a new set of Fourier components is constructed using the phases from the radar image and amplitudes from the model image. The corrected model image can be obtained by inverse transformation of the new set of Fourier components. Scales of wavelengths starting from 50 to 500 km are playing a crucial role in correcting the positional and intensity errors of continental-scale precipitating systems predicted by the WRF model, while scales greater than 500 km are contributing very little, which is negligible. The model errors are characterized in terms of phase and power. Compared to power/amplitude correction, phase correction plays a major role in eliminating the positional and intensity errors. Scales that are responsible for the errors or not predicted by the model have a decorrelation time of about 2 h in both the zonal and meridional directions. The higher critical success index (CSI) and smaller RMS error values of the radar-extrapolated and phase-corrected fields compared with the WRF model–predicted fields indicate that the extrapolated phase information can be used to correct the model outputs up to a lead time of about 4 h.