PROGNOSTIC DISCUSSION FOR 6 TO 10 AND 8 TO 14 DAY OUTLOOKS NWS CLIMATE PREDICTION CENTER CAMP SPRINGS, MD 300 PM EDT FRI AUGUST 21 2009 THE OPERATIONAL 00Z AND 06Z GFS MODEL SOLUTIONS BEGIN TO BREAK DOWN THE RIDGE OVER THE PACIFIC NORTHWEST WHILE THE HIGH RESOLUTION 00Z ECMWF MAINTAINS A STRONG RIDGE THERE. TELECONNECTIONS FROM THE UPSTREAM TROUGH OVER THE GULF OF ALASKA AND WESTERN ALEUTIAN RIDGE BOTH SUPPORT RIDGING OVER WESTERN NORTH AMERICA WHICH AGREES MORE WITH THE 00Z ECMWF SOLUTION. THIS, IN COMBINATION WITH VERY HIGH 500-HPA ANOMALY CORRELATIONS EXHIBITED DURING THE PAST 60 DAYS, RESULTED IN THE ECMWF-BASED SOLUTIONS BEING FAVORED IN TODAYS OFFICIAL 500-HPA HEIGHT BLEND CHART. TODAY'S OFFICIAL 500-HPA BLEND CONSISTS OF 10% OF TODAY'S OPERATIONAL 6Z GFS CENTERED ON DAY 8...20% OF TODAY'S GFS SUPERENSEMBLE MEAN CENTERED ON DAY 8...20% OF TODAY'S OPERATIONAL 0Z ECMWF CENTERED ON DAY 8...40% OF TODAY'S 0Z ECMWF ENSEMBLE MEAN CENTERED ON DAY 8...AND 10% OF TODAY'S 0Z CMC ENSEMBLE MEAN CENTERED ON DAY 8.
43
Embed
PROGNOSTIC DISCUSSION FOR 6 TO 10 AND 8 TO 14 DAY OUTLOOKS
PROGNOSTIC DISCUSSION FOR 6 TO 10 AND 8 TO 14 DAY OUTLOOKS NWS CLIMATE PREDICTION CENTER CAMP SPRINGS, MD 300 PM EDT FRI AUGUST 21 2009 - PowerPoint PPT Presentation
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
PROGNOSTIC DISCUSSION FOR 6 TO 10 AND 8 TO 14 DAY OUTLOOKS NWS CLIMATE PREDICTION CENTER CAMP SPRINGS, MD 300 PM EDT FRI AUGUST 21 2009
THE OPERATIONAL 00Z AND 06Z GFS MODEL SOLUTIONS BEGIN TO BREAK DOWN THE RIDGE OVER THE PACIFIC NORTHWEST WHILE THE HIGH RESOLUTION 00Z ECMWF MAINTAINS A STRONG RIDGE THERE. TELECONNECTIONS FROM THE UPSTREAM TROUGH OVER THE GULF OF ALASKA AND WESTERN ALEUTIAN RIDGE BOTH SUPPORT RIDGING OVER WESTERN NORTH AMERICA WHICH AGREES MORE WITH THE 00Z ECMWF SOLUTION. THIS, IN COMBINATION WITH VERY HIGH 500-HPA ANOMALY CORRELATIONS EXHIBITED DURING THE PAST 60 DAYS, RESULTED IN THE ECMWF-BASED SOLUTIONS BEING FAVORED IN TODAYS OFFICIAL 500-HPA
HEIGHT BLEND CHART. TODAY'S OFFICIAL 500-HPA BLEND CONSISTS OF 10% OF TODAY'S OPERATIONAL 6Z GFS CENTERED ON DAY 8...20% OF TODAY'S GFS SUPERENSEMBLE MEAN CENTERED ON DAY 8...20% OF TODAY'S OPERATIONAL 0Z ECMWF CENTERED ON DAY 8...40% OF TODAY'S 0Z ECMWF ENSEMBLE MEAN CENTERED ON DAY 8...AND 10% OF
TODAY'S 0Z CMC ENSEMBLE MEAN CENTERED ON DAY 8.
Generic Levels of EM Uncertainty – Lessons from geophysical
modeling and current studies of climate impacts
Nicholas Bond1
Kerim Aydin2, Anne Hollowed2, James Overland3 and Muyin Wang1
1 University of Washington/JISAO
2 NOAA/AFSC3NOAA/PMEL
Techniques for Incorporating Model Ensembles
• Simple means• Means w/ individual bias corrections• Means w/ collective bias corrections• Regularization via EOFs• Bayesian techniques
22 UKMO -HadGem1 UK 1.25°x1.875°L38 (0.33 -1.0°) x 1.0° L40 1+2* 2 1*
Sum 55 40
Models Contributed to IPCC AR4Models Contributed to IPCC AR4
Walsh et al. (2007)
Bayesian Model Averaging (BMA)
• Considers an ensemble of plausible models• Key Idea - The models vary in their skill, and
calibration of this skill produces better forecasts • Forecast PDF estimated through weighting the
PDFs of the individual models, with weights determined by posterior model probabilities
• BMA possesses a range of properties optimal from a theoretical point of view; works well in short-term weather prediction
p(y) is the forecast PDF; fk is the kth forecast model; wk is the posterior probability of forecastk being the best; gk(y | fk) is the PDF conditional on fk being the best forecast.
Weighted ensemble mean of parameter y
Estimating Weights by Maximum Likelihood
• Yields parameter values (weights) that make observed data most likely
• Likelihood function maximized over time and space through determination of model weights for a particular parameter
• Method uses expectation-maximization (EM) algorithm, which resembles a “hotter-colder” game
• Final weights related to how often a particular model constitutes the best model
• Training data set consists of last 40 days of short-term weather forecasts
Ensemble Model Projections for North Pacific Marine Ecosystems
• Initial Selection - Pick models that replicate the observed character of the PDO in their 20th century hindcasts (12 of 22 pass test)
• Regional Perspective - Examine specific parameter(s) in region of interest; consider means, variances, seasonality, etc.
• Model projections - Use quasi-Bayesian method based on “distance” between hindcasts and observations; form weighted ensemble means
• Uncertainty/Confidence - Estimate based on a combination of inter-model and intra-model variances in projections
Present Application• Limited statistics for evaluation (there has been a single
outcome for the past climate) • Compare 20th century hindcast simulations by the climate
models to observations on a regional basis• Observations based on NCEP Reanalysis; good match of
spatial scales• Consider mean, variance, and other measures (trend,
seasonality, etc.) if appropriate• Estimate weights for projections following a scheme
developed for objective analysis of weather observations, Wk = exp(-Dk/Dm)
• Apply intra-model variance from models with 5+ runs to models with fewer runs
Parameters Evaluated• Bering Sea - Flow through Unimak Pass (Nutrient
Final Remarks• Global climate model simulations are being used for a
host of regional applications• There does not seem to be any single “best” method,
but the protocol should include evaluation of model hindcast simulations of key parameters in the region of interest.
• Multi-model ensembles represent a key tool for seasonal climate forecasts, and are being used increasingly for short-term weather prediction.
• On long time horizons, model structural uncertainty dominates initial condition sensitivity.
• From present to mid-21st century, climate change liable to be dominated by thermodynamic effects as opposed to dynamic effects (e.g., winds). The latter will be prone to interannual to decadal natural variability.
• The output from global climate models (perhaps subject to statistical downscaling) can complement that from vertically-integrated numerical models with full dynamics.