Forecast Errors and Uncertainties in Atmospheric Rivers

July 15, 2020

ECMWF scientist and former CW3E postdoc, Dr. David Lavers published an article in May 2020 in Weather and Forecasting. Co-authors included representatives from leading global numerical weather prediction centers, several of whom serve on the AR Recon Modeling and Data Assimilation Steering Committee (as given by asterisks): Bruce Ingleby, David Richardson, Mark Rodwell, and Florian Pappenberger* of ECMWF; Vijay Tallapragada* of NCEP; Jim Doyle* and Carolyn Reynolds* of the Naval Research Laboratory; and multiple key academic partners including Aneesh Subramanian* of CU Boulder, Ryan Torn of SUNY Albany, and CW3E Director F. Martin Ralph*. The collaboration on this article between global operational numerical weather prediction centers and academic institutions is an example of how the Atmospheric River Reconnaissance (AR Recon) Program brings together scientific leaders to leverage airborne observations offshore in and around ARs to support improved forecasts of ARs.

Specifically, this study uses the dropsonde observations collected during the AR Recon campaign and the European Centre for Medium-Range Weather Forecasts (ECMWF) Integrated Forecasting System (IFS) to evaluate forecasts of ARs based upon their temperature, wind, and moisture characteristics. Results show that ECMWF IFS forecasts

• were colder than observations throughout the troposphere (Fig. 1);
• were drier and had weaker winds than observations in the lower troposphere, resulting in weaker horizontal water vapor fluxes at low levels (Fig. 1);
• exhibit an underdispersiveness in terms of the water vapor flux (Fig. 2)

The underdispersiveness in water vapor flux observed in this study largely arises from model representativeness errors associated with dropsondes. To supplement the information provided by the dropsondes, four U.S. West Coast radiosonde sites are assessed and confirm the IFS cold bias throughout the winter season.

The issues identified here are likely to affect the model’s hydrological cycle and hence precipitation forecasts. The diagnosis of model behavior using unique dropsonde observations from AR Recon performed by this study helps better understand ECMWF IFS model performance and errors.

This research supports the goals of two Priority Areas within CW3E’s 2019-2024 Strategic Plan–Atmospheric Rivers Research and Applications, and Emerging Technologies–to produce and improve forecasting and decision support tools that meet the needs of western U.S. forecasters, resource managers, and emergency managers.

Figure 1. Figure 3 in Lavers et al. (2020): Shown are O – B departures (in the EDA control member) averaged in 50-hPa layers at the dropsonde locations for (a) specific humidity, (b) temperature, and (c) wind speed. (d) Observed (gray) and background (red) pressure-level water vapor flux magnitude averaged in 50-hPa layers when interpolating specific humidity on to the wind levels and (e) the O – B departures for water vapor flux. The error bars show the 90% confidence interval of the mean. The number of values in each layer is given on the right-hand side of the panel, and the number used across all layers is shown at the top right of each panel.

Figure 2. Figure 7 in Lavers et al. (2020): The modified spread-error relationship of pressure-level water vapor flux magnitude. The square roots of DepVar (solid), EnsVar (dashed), and EnsVar + ObsUnc2 (dotted) are shown at 0-120-h forecast lead times for (a) 925, (b) 850, and (c) 700 hPa.

Lavers, D.A., N.B. Ingleby, A.C. Subramanian, D.S. Richardson, F.M. Ralph, J.D. Doyle, C.A. Reynolds, R.D. Torn, M.J. Rodwell, V. Tallapragada, and F. Pappenberger, 2020: Forecast Errors and Uncertainties in Atmospheric Rivers. Wea. Forecasting, 35, https://doi.org/10.1175/WAF-D-20-0049.1.