3B.6 MPAS (Model for Prediction Across Scales): An Overview of Forecast Evaluation and Improvements

Monday, 4 June 2018: 2:45 PM
Colorado B (Grand Hyatt Denver)
Brett A. Wilt, The Weather Company, Andover, MA; and J. P. Cipriani

The Weather Company (TWC) has selected MPAS for its next-generation Deep Thunder global model to replace “RPM”, a Weather Research and Forecasting (WRF) based system. Research and development for the Deep Thunder (DT) MPAS implementation was initiated in January 2017, with daily experimental cycles at 15-km uniform resolution beginning in June 2017. The 144-hour forecasts are cold start initializations using the 0.25-degree NCEP GFS analyses at 00Z, 06Z, 12Z, and 18Z, with the addition of NASA SPoRT 2-km SST and 4-km VIIRS green vegetation fraction data.

Although DT MPAS will ultimately replace all WRF based solutions at TWC, the initial goal is to replace a global 13-km WRF based forecast system. The core requirement in achieving this goal is consistent metrics that yield forecast skill better than or equal to global WRF, as well as positive case study analyses. In order to maximize the research to operations efficiency, a robust DT MPAS lab environment is utilized, which consists of 100 case dates from 2014 and 2016-17 (50 high-impact and 50 evenly distributed).

An overview of the DT MPAS progress will be discussed, along with a variety of physics tuning, improvements, and post-processing diagnostics. As an example of DT MPAS lab, a baseline of 2-meter temperature RMSE and bias statistics are computed for approximately 800 METAR sites across the continental United States (CONUS) using the NCAR MPAS 5.0 code release. With minor tuning to the surface layer, PBL scheme, NOAH land surface model, the addition of NASA SPoRT 2-km SST and 4-km VIIRS green vegetation fraction data, CONUS averaged 72-hour forecast RMSE is reduced by 0.79K during winter months and 0.20K during summer months. Short-term (01-12 hour) forecasts biases are also addressed with a DT MPAS mixing ratio initialization change and the addition of the NCEP GFS cloud analysis. Valuable insight for future improvements will continue to be extracted from the DT MPAS lab, including 10-meter wind speed and precipitation metrics.

The second goal of DT MPAS is an hourly updating, convective allowing forecast system. This will highlight the “Model for Prediction Across Scales” aspect, with a variable mesh that introduces the increasing need for scale-aware physics. A quick overview of forecast impacts related to convective parameterization schemes, both scale-aware and non scale-aware, will be presented for hurricanes Harvey and Irma.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner