Hybrid Wind Nowcasting for a Seamless Prediction of Wind Profiles over an Airport up to Four Hours

For save air traffic operations timely and accurate weather information is of crucial importance as expressed in the NextGen Weather or SESAR 2020 programs. Costly delays or runway capacity limitations due to adverse weather situations may be mitigated, due to improved thunderstorm predictions or wake vortex separation management as proposed in SESAR2020 projects PJ.04 and PJ.02, respectively. However, many studies point out the gap in prediction accuracy for the very-short range forecast horizon between 30 minutes and four hours. There, the skill of nowcasting techniques drops significantly while weather prediction models are still affected by numerical spin-up effects.

To fill this gap, three approaches appear obvious: improve the nowcasting, improve or optimize numerical weather prediction systems, or use adequate blending strategies of both. We present a joint effort of DLR and DWD to increase wind prediction accuracy tailored to the international airport of Munich (MUC) and compare it to the currently available forecasts of the COSMO-DE (Consortium for Small-Scale Modeling) model of DWD. The experimental system, called COSMO-MUC, pursues a two-fold approach. First a regionalized prototype of the COSMO-DE model, COSMO-MUC, has been adapted to local conditions at MUC. Adaptions comprise an increased horizontal resolution, enhanced land use information and orography data sets, and an hourly assimilation-forecast cycle. Additionally, newly available measurement data are included in the assimilation process. The second step consists of an adaptive blending of nowcasts of locally available wind profiles and the numerical weather prediction output by a Kalman Filter technique. Local data stems from wind profiles from a previously installed  SELEX Meteor 50DX radar and a Lockheed Martin WTX Wind Tracer lidar at MUC.

The experimental COSMO-MUC nowcast system is evaluated for two test periods in 2014. While prediction accuracy seems to be only slightly enhanced by the adaptions to the numerical weather prediction model, the hybrid nowcasting promises an efficient measure to significantly improve wind prediction accuracy, given adequate measurement data.  A reduction in RMSEV of 10% - 15% for horizontal wind is observed in the complete forecast horizon of four hours compared to persistence nowcast in the first hour and numerical predictions later on.  

Future work will concentrate on the confidence bounds of the wind prediction by adaptive strategies to confine the model error in the filter equations.