Presentation on theme: "A Brief Guide to MDL's SREF Winter Guidance (SWinG) Version 1.0 January 2013."— Presentation transcript:
A Brief Guide to MDL's SREF Winter Guidance (SWinG) Version 1.0 January 2013
What's this all about? An innovative way to view and understand SREF output Calibrated probabilistic forecast guidance, based on NCEP's Short Range Ensemble Forecast (SREF) system--SREF Winter Guidance (SWinG) Prototype includes weather elements that focus on rain/snow/freezing rain forecast decisions Available on the web for all SREF forecast cycles and time projections at a limited number of stations
Why use calibrated probabilities? Ensembles are often overconfident (underdispersed). Too frequently the verification falls outside the spread of the ensemble members. SWinG forecasts are calibrated. True measure of forecast confidence. Statistically reliable spread.
How do I use it? If precipitation type is a question, and You expect SREF to be skillful Assess the meteograms for your stations SREF/xml/meteoform_sref.php
How to Assess Meteograms Time series forecasts of weather elements related to precipitation type Black line is 50th percentile
How to Assess Meteograms Grey areas show spread of the distribution (10th, 30th, 70th, and 90th percentiles) 10% 90% 70% 30%
How to Assess Meteograms Red lines show the station specific climatological boundary for rain/snow, if available
How to Assess Meteograms Tri-color lines show rule of thumb values for rain/freezing/frozen Rain/freezing frozen threshold Freezing/snow line All-snow threshold
Up Close Close up of 850 mb Temperature
Up Close Compare spread at 0300 and 1500 UTC. Forecast is more confident at 0300.
Up Close At 0300, guidance indicates ~70% chance of 850 mb temperature below key value (-1.5° C)
Up Close At 1500, SWinG indicates ~20% chance of 850 mb temperature below key value (-1.5° C)
Which weather elements? Current 2-m Temperature 850 mb Temperature mb Thickness mb Thickness mb Thickness mb Thickness Future Dendritic Growth Zone Depth Omega Freezing Level Positive/Negative Energy Snow Liquid Equivalent Why these weather elements? There are better parameters for winter weather!
Why these weather elements? We have a very short sample. SREF began running in this configuration 21 Aug We are using a modified form of Bayesian Model Averaging (BMA). This technique can only forecast weather parameters that are observed daily. Currently, SREF vertical velocities for NMM and NMMB have problems. It's a new technique. We started with the easiest weather elements.
Which Stations? On NCEP's Central Computing System, we generate SWinG for more than 3000 stations Adapted from the BUFR station list used at NCEP BUFR station list is source for BUFKIT application On our web page, we generate images for ~400 stations All upper air stations in CONUS and Alaska Additional stations to support WFO LWX Winter Weather Pilot Project. We can, and will, add stations to the web page Contact us if you want us to add stations
How do we make SWinG? Using most recent verification... Correct bias of each member Weight the bias-corrected members (ARW, NMM, NMMB members) Correct forecast spread Compute probabilities We have named this technique Decaying Average Bayesian Model Averaging (DABMA).
Previous Forecasts Today's Observation Update Bias Corrections Bias correction for each model core... Latest SREF Forecast Correct Bias of Each Member We track and remove the bias of each member. We update this bias correction daily with the most recent verification. Previous Bias Corrections New Estimate = 0.95 x Previous Estimate x Today's Estimate Latest SREF Forecast Correct the Bias of Each Member i.e., 1 bias correction value each for ARW, NMM, NMMB, which is applied to each of their respective members more
Previous Bias- Corrected Forecasts Today's Observation Update Relative Weights Relative weights for each model core Using the most recent verification, we compute relative weights for bias-corrected ARW, NMM, NMMB members Previous Weights
Using most recent verification, correct forecast spread Previous Forecasts Today's Observation Compute optimal spread Optimal spread Raw Spread Corrected
We compute probabilities using a Normal Mixture Model to combine member forecasts.
Illustration: Three members (blue) contrib- ute to final probability distribution (black) For SREF, we use all 21 members.
We compute probabilities using a Normal Mixture Model to combine member forecasts. Relative model weights set height of each blue curve
We compute probabilities using a Normal Mixture Model to combine member forecasts. Bias-corrected SREF forecasts set position of each blue curve on X-axis
We compute probabilities using a Normal Mixture Model to combine member forecasts. The optimal spread deter- mines the spread of each blue curve.
Join the conversation! We are using the NWS Innovation Web Portal (IWP) to gather feedback from forecasters. https://nws.weather.gov/innovate/group/guest/communities You will find Additional documentation and case studies Forum where you can submit questions and comments For access Follow the URL and login with NOAA credentials Select “Available Communities” tab Find “SREF Winter Guidance” and “Join”