1. Development of new forecast products for weeks 3-4 Nat Johnson 1 Stephen Baxter 2,3, Steven Feldstein 4, Jiaxin Feng 5,6, Dan Harnos 2, Michelle L’Heureux 2, and Shang-Ping Xie 5 1 Cooperative Institute for Climate Science, Princeton University 2 NOAA/NCEP Climate Prediction Center 3 University of Maryland 4 Penn State University 5 Scripps Institution of Oceanography, University of California, San Diego 6 NOAA Geophysical Fluid Dynamics Laboratory NOAA Climate Test Bed Meeting

2. NOAA CPC is bridging the forecast gap in weeks 3-4 Key Questions:  What are the sources of skill for lead times of 3-4 weeks?  How do we translate that knowledge into operational forecast guidance? http://www.cpc.ncep.noaa.gov/products/ predictions/WK34/

3. Many studies indicate that the tropics are a key source of extended-range midlatitude predictability. Specifically, relationships with the El Niño/Southern Oscillation (ENSO) and Madden-Julian Oscillation (MJO) hold promise for weeks 3-4. Riddle, Stoner, Johnson, L’Heureux, Collins, and Feldstein (2013, Climate Dynamics) 500-hPa height anomalies (m) Temperature anomalies (ᵒC) Days that MJO precedes pattern Anomalous frequency of cluster pattern (top left) occurrence (%) MJO influence on cluster pattern The MJO gives information on pattern occurrence 10-25 days in advance MJO phase A weekly cluster pattern

4. How do we transition the gains from our research into an operational forecast tool? We generate winter North American temperature forecasts for weeks 3-4 based on empirical relationships with MJO, ENSO, and the linear trend (Johnson et al. 2014, Weather and Forecasting) Climatology MJO+ENSO+trend forecast Simple but transparent Translates research into practical guidance for the forecaster Produces skillful forecasts

5. For some initial states of the MJO and ENSO, the skill scores of the weeks 3-4 T2m forecasts from the empirical model are substantially higher than the typical skill scores of dynamical models.

6. Operational Adaptation Extend periods from DJFM to 12 running 3- month periods. Shift from ERA-Interim to daily observations: – CPC Internal T2m Data (Janowiak et al. 1999) – CPC Unified Gauge-Based Analysis (Xie et al. 2010) Fourth root taken to increase distribution normality. Shift from three-class to two-class forecast. Combined product for Weeks 3 and 4.

7. Weeks 3+4 Heidke Skill Score from combined effects of ENSO+MJO+Trend

8. Embedded periods of lower/higher skill dependent upon background climate state. Note even with a weak MJO, skill is present. Active background state not necessary for a skillful forecast (due to trend and nonlinearity of ENSO).

9. Temperature Top left: Forecast (from 6/19) Mid/Bot left: Forecast tool output Bot right: Observations (7/4-17) Coherence between statistical tool and forecast, despite tool being unavailable to forecasters. Forecast and tool verified well.

10. Precipitation Top left: Forecast (from 6/19) Mid/Bot left: Forecast tool output Bot right: Observations (7/4-17) Less agreement between tool and forecast made without it. Forecast outperformed tool.

12. O2R: How can we do better?  Accounting for MJO and ENSO amplitude: multiple linear regression or logistic regression?  Accounting for nonlinear ENSO effect? (Johnson and Kosaka, submitted) How do we expand beyond ENSO- and MJO-based forecast guidance?  Refining dynamical forecast guidance  Incorporating other skillful predictors in statistical guidance

13. Predicting large-scale teleconnection pattern indices: What are the skillful predictors and how much skill?  Partial least squares regression model of DJF PNA index  Predictors: tropical OLR and Northern Hemisphere z300  Comparison with CFSv2 skill Day 16 predictor patterns OLR z300 -0.2-0.100.1 0.2 CFSv2 more skillful through week 3, statistical model more skillful in week 4 Jiaxin Feng, Steve Baxter, Nat Johnson

14. Question: What is the relationship between interference, tropical convection, and surface air temperature, sea ice, the stratospheric polar vortex, and the Arctic Oscillation? Stationary wave index (SWI): Defined as the projection of the daily 300-hPa streamfunction onto the 300-hPa climatological stationary eddies. Steven Feldstein and Michael Goss

15. Surface Air Temperature: Constructive interference with and without Warm Pool convection Western Pacific OLR < -0.5 Western Pacific -0.5 < OLR < 0.5 Western Pacific OLR > 0.5 Goss, Feldstein, and Lee (submitted)

16. Refining dynamical forecast guidance: Are particular initial conditions associated with high skill in CFSv2? High (Week3 HSS > 1.0 σ) minus low (Week3 HSS < -1.0 σ) skill composite initial 300-hPa streamfunction and tropical precipitation anomalies Ψ300 (m 2 s -1 ) Precip (mm d -1 ) Goss and Feldstein (2015, Mon. Wea. Rev): Initial atmospheric flow over northeastern Asia, Arctic coast of Siberia, and northwestern North America may have large impact on the extratropical response to the MJO.

17. Conclusions An empirical MJO/ENSO phase model has been undergone successful R2O prior to CPC’s experimental Week 3-4 forecast product going live, and has become a key component of the forecasting process. Research team member participation in the Week 3-4 forecasting process has aided statistical tool interpretation by forecasters, and led to feedback for subsequent product development in an O2R sense. Focus in FY16: Improving understanding of skillful predictors for weeks 3-4, translating that knowledge into forecast guidance, CFSv2 calibrations and possible statistical-dynamical hybrid

18. Dynamical forecast guidance: How well does the CFSv2 perform in weeks 3-4? 4 x daily retrospective forecasts out to 45 days, 1999-2009 Mean of four daily forecasts, bias correction applied DJFM T2m, North America Week 3 Week 4 HSS Week 2