Diagnosing systematic numerical weather prediction model bias from thermodynamic tendencies in short-term forecasts

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Thursday, 2 July 2015: 8:30 AM
Salon A-2 (Hilton Chicago)
Steven M. Cavallo, University of Oklahoma, Norman, OK; and J. Berner and C. Snyder

Accurate predictions in numerical weather models depend on the ability to accurately represent physical processes across a wide range of scales. We evaluate the utility of using model thermodynamic forecast tendencies to diagnose systematic forecast biases in the Advanced Research Weather Research and Forecasting (ARW) numerical model. This method is implemented using ARW with an Ensemble Adjustment Kalman Filter (EAKF) within the framework of the Data Assimilation Research Testbed (DART). We apply this method in two experiments selected by their contrasting geographic regions of the globe with regionally-specific configurations of the ARW used for real-time forecasting: (1) the Advanced Hurricane Weather Research and Forecasting (AHW) modeling system and (2) the Antarctic Mesoscale Prediction System (AMPS). These experiments are designed to efficiently identify the precise model parameterizations that lead to systematic model bias in order to better focus on devising methods to improve the physical representation of particular atmospheric processes.

The experiment in the tropical Atlantic was performed during the period August-November 2010 where a systematic warm temperature bias with a magnitude ~0.5 K occurred in a deep tropospheric layer centered ~700 hPa. This method revealed that this bias originated primarily from the Kain-Fritsch convective heating parameterization scheme, and furthermore that this bias developed within the first 30-minutes of each forecast. The experiment over the Antarctic was performed during the period September-October 2010 to coin-cide with the Concordiasi intensive observation period, when special droponsonde observations were available as an independent source of model validation. Here, where convection is relatively small, lower-tropospheric biases arose primarily from parameterized boundary layer processes, with results sensitive to the underlying surface type and orography of the Antarctic continent. These experiments furthermore quantify the evolution of model spin-up, where it is concluded that spin-up has distinct phases that can only be minimized when initializing model forecasts from an ensemble member that is generated using a data assimilation system using an identical model configuration.