Regime Dependent Bias Correction of Ensemble Output

Open Access
Author:
Boden, Joshua Randal
Graduate Program:
Meteorology
Degree:
Master of Science
Document Type:
Master Thesis
Date of Defense:
November 26, 2012
Committee Members:
  • George Spencer Young, Thesis Advisor
  • Johannes Verlinde, Thesis Advisor
  • Anne Mee Thompson, Thesis Advisor
  • Paul Gerard Knight, Thesis Advisor
Keywords:
  • Ensemble
  • Regime
  • Forecasting
Abstract:
Today’s most commonly used point sensible weather forecasting guidance (MOS) does not fully exploit recent increases in modeling prowess and continues to struggle with atmospheric non-linearities. A more accurate and versatile forecasting technique is proposed which merges forecasts from the non-operational 21 member SREF system with flow dependent (regime) error correction of 2-meter temperature and QPF. Previous flow regime work identified the atmospheric patterns but only by purely statistical means and over weeks to yearlong syntonic scale applications. This work defines short-term (3-hr) dynamic, physical regimes, specific to the variable being forecasted, with the relevant regime for each forecast diagnosed via series of ground truth, satellite and radar observations. The system parses point SREF ensemble forecasts by physics schemes into four parent member groups, performing a regime dependent debiasing of each group and then averaging across all groups to develop a forecast consensus. The technique was tested via cross validation. Sinusoidal fitting of the seasonal bias trend in the remaining error was also investigated as a secondary prognostic tool. Deterministic temperature and QPF forecasts were generated over an annual cycle, with the technique increasing skill as compared to the SREF Ensemble Mean, operational model and its associated MOS guidance across six climatically different locations tested. Annual MAE and KSS test analysis points to consistently increased 2-meter temperature and QPF skill over MOS, revealing the importance of regime definition when debiasing model forecasts.