Weather Regimes and Forecast Errors in the Pacific Northwest Lynn McMurdie (co-author Joe Casola) University of Washington

Why the Pacific Northwest and the North Pacific Ocean? The Pacific is one of the larger areas of sparse insitu observations in the world Uncertainty over the Pacific has a large impact on predictability over the North American continent and beyond (Hakim 2003 & 2005, Chang 2005, Reynolds & Gelaro 2001, Colle – CW) Short-term forecast errors of SLP often largest over west coast of North America and over North Pacific than other regions of North America (McMurdie and Mass 2004, Wedam et al. 2008, Colle and Charles 2008)

Why the North Pacific Ocean, cont.? E. Pacific W. Atlantic Colle and Charles (2008) used a feature tracking algorithm to calculate GFS position and central pressure errors of surface cyclones. Wedam et al. 2008, SLP errors along east and west coasts by NAM

00 hr GFS 72 hr GFS 12 UTC 12 Nov 2007 – Veteran’s Day Windstorm Subsequent Errors 36h: 10 – 15 hPa 24h: 5 – 15 hPa 12h: 5 – 12 hPa Large Position Errors Errors by Other models CMCG 15 – 22 hPa ECMWF 12 – 20 hPa NAM 18 – 26 hPa Large Position Errors

Why do large short-term forecast errors occur? Important to investigate the circumstances under which large forecast errors occur Do large forecast error events occur during particular ‘flow (or weather) regimes’?Do large forecast error events occur during particular ‘flow (or weather) regimes’? Define Weather Regime: Recurrent spatial structures within the atmosphere (e.g. Cheng and Wallace, 1993)

Pacific NW Weather Regimes MethodMethod - Cluster a single 500-hPa geopotential height contour (540 dm) to identify regime patterns DataData – NCAR Reanalysis data, DJFM , 150°E – 60°W (Casola and Wallace, 2007) MotivationsMotivations – Using a single contour – Appeals to synoptic intuition – Yields a simple framework ResultsResults – Four regimes identified: Offshore Trough, Alaskan Ridge, Coastal Ridge and Rockies Ridge OTARCRRR

4 Regimes of Interest Off-Shore TroughAlaskan RidgeCoastal RidgeRockies Ridge Full Field – 510dam:552dam, 6 dam interval, 540 dam in blue Anomalies – 2 dam interval; (+) in Black; (-) in Blue

Regimes for This Study For the period November 2002 – March 2008 (winter months only), calculated 3-day running mean of 500-hPa 540 dm and matched each day to one of the four regimes that matched the closest. transitionIf the pattern correlation was less than 0.5, then there was no match, and that day was called ‘transition’ In this study, forecast errors of sea level pressure and 2-m temperature are compiled for each of the 4 weather regimes

Large Forecast Errors - SLP h Forecasts of sea level pressure (slp) by the GFS and NAM models for 6 winter seasons (November 2002 – March 2008) were compared to slp observations at coastal and offshore stations Average error, standard deviation, and Mean Absolute Error (MAE) calculated at each station Daily ErrorDaily Error = average MAE for all stations for each day, each forecast lead time If error > 2*SD at an individual station, it was a large error day for that station (also used 3*SD) Large Error DayLarge Error Day = more than 2 stations experienced errors > 2*SD.

Big Small 22% 13% 15% 17% 33% NAM +Bias GFS -Bias F48 NAM errors > GFS errors F48 (Others same) Daily errors of MAE and ME

Rockies Ridge experiences highest frequency of large error days Coastal Ridge experiences the lowest Offshore Trough has 2 nd highest frequency of large error days -- NAM GFSNAM Frequency of Large Error Days

What’s so special about Rockies Ridge? Has straight jet across Pacific Has largest variance of SLP in region closest to the PacNW Windstorms and Floods often in this regime RR

What about Jetstream Strength? Anecdotal and theoretical evidence that when there is a strong jet across the Pacific – larger uncertainty in forecasts and larger forecast errors, esp. SLP Used 300 hPa windspeed from NCAR reanalysis data ( ) to define the Jet in a box across Pacific – 180°W to 125°W; 40°N to 55°N Separate dataset into weak/neutral/strong 300 hPa windspeed.

F48 Other forecast hours NOT the same: F72, no significant differences Other ‘wind boxes’, results NOT the same The MAE during strong jets is NOT always larger than the MAE during weaker jets

Frequency of Large Error Days Large Error events happen more frequently during strong Pacific Jets than weaker jets. GFS NAM

Forecast Errors of Temperature Verification Region: Pacific NW including WA, OR, ID, southern BC and northern CA Land stations and nearshore buoys, CMAN stations (networks included: SA, CM, BF, SS, RW, HN, AW, AM, EC) Usually had 500+ observations/day --> 1 error/day Calculated MAE, ME at 00 UTC and 12 UTC for each day and averaged for each Regime

Example to show distribution of observations Large Error Event: GFS 48hr for 29 Nov 2002, positive errors negative errors

BigSmall 12 UTC F48 (Other forecast hours same. 00 UTC similar but not as significant)

Frequency of Large Error Days 12 UTC GFS 12 UTC NAM Coastal Ridge has highest frequency of large errors for both models and all forecast hours All other regimes have below or average number of large error days Coastal Ridge is typically ‘benign’ with light winds – model errors, such as the model parameterization of nocturnal boundary layer, may be significant factor.

Take Home Message(s) large SLP errors events strong jets Strong jets are not necessarily associated with large errors of SLP, but large SLP errors events are often associated with strong jets. errors of SLP RockiesRidge Large forecast errors of SLP in my ‘box’ are a function of the ‘shape’ of the upper level flow (with the Rockies Ridge regime experiencing the largest average MAE and the Coastal Ridge the smallest.RR errors of minimum Temperature Coastal Ridge Large forecast errors of minimum Temperature over the PacNW are also related to weather regime, with the Coastal Ridge experiencing the largest errors.CR

Implications shapeThe statistical evidence here demonstrates that upper level flow shape modulates the observed forecast errors (i.e. RR regime) initial condition errors may still dominateProcesses of downstream development, amplification of jet still contribute to error growth and propagation, but large errors on west coast are not limited to these processes alone – initial condition errors may still dominate There are other ways to quantify forecast errors: sensitivity fields, ensemble spread. Can these results be reproduced using these metrics?

Composite Jet for large error events Climatology from NDJFM (Data source NCEP – NCAR reanalysis) Large Error Event Wind anomalies

How Important are these forecast errors? Do large forecast errors occur for storms that impact society/property …Do large forecast errors occur for storms that impact society/property … Snow Winds … Floods…

High (significant) Impact Event? High Impact EventHigh Impact Event = Any event mentioned in the Storm Data Publications during Oct – Mar of 2002 – 2007, west of Cascades WA & OR, affecting significant area (high wind events, significant rain/flooding, lowland snow) Large Error DayLarge Error Day = Any day where the forecast of sea level pressure at coastal and offshore stations is large (more than 2*sd) at 2 or more locations Are large error days associated with high impact events?Are large error days associated with high impact events?

911 Days, 5 winter Seasons High Impact Events ~17% Large Error Days ~ 25% 55% of High Impact Events have large errors. If the large error date lags 1 day or is same day as a high impact event, then 70% of High Impact Events have large forecast errors.

Why Cluster using a “Limited Contour?” Chosen contour (540dm) runs through variance maximum in the North Pacific Advantages over linear methods (e.g., PCA). There are no constraints on the patterns (they don’t have to be orthogonal nor anti-symmetric to one another). Easy to relate individual records to clusters.

Periods of Large Errors Periods of Small Errors Error > 2*sd

Distinguishing Ridges The height of the 500 hPa surface at 45°N latitude for each cluster; Climatology (-), Cluster (- -); Ridges in red, Troughs in blue. Same as above, climatology removed.

Limited Contours 1008 individual spaghetti strands to be clustered Each “spaghetti strand” is a 5-day average of the 540-dam contour of the 500-hPa geopotential height field, taken from NCEP-NCAR Reanalysis data DJFM , 150°E – 60°W Casola and Wallace (2007, JAMC)

20%12%15%21%32% Big Small +Bias -Bias

Some years have regimes that lastSome years have regimes that last Some don’tSome don’t Largest errors found more frequently during Rockies Ridge (& Offshore Trough)Largest errors found more frequently during Rockies Ridge (& Offshore Trough)

SLP Variance for the different Regimes