Download presentation
Presentation is loading. Please wait.
Published byDustin Carter Modified over 9 years ago
1
Using Ensemble Models to Develop a Long-Range Forecast and Decision Making Tool Brandon Hertell, CCM Con Edison of New York Brian A. Colle, Mike Erickson, Nathan Korfe Stony Brook University
2
Motivation Actionable weather forecasts required at longer lead times Conveying certainty/uncertainty in weather forecasts at any time length challenging Day 0 Accuracy/Confidence Lead Time Low High 2
3
Challenging Forecasts “Normal” Weather Extreme Weather High impact, low probability events are the most difficult 3
4
Timeline > 5 Days Monitoring Day 5 Monitoring Notifications Day 3 Preparation Resource Decisions Mobilization Day 0 – Storm Impact Ride out storm 4
5
Decisions Being Made Sooner Do Nothing Better be 100% correct $$ cost of being wrong is high Unhappy customers Bad publicity Regulation Do Something Pre-mobilization Scheduling Planning Resources Mutual Assistance Decision 5
6
Current Methodology Meteorologist experience and knowledge of the current weather, combined with the model forecasts and other data dictates confidence level in the weather forecast Hypothesis- An ensemble weather model may provide an engineered solution to quantifying forecast probabilities at any time scale …how would decision making change if this were the case? 6
7
Ensemble Decision Tool 2014 Phase 1 – Develop visualizations that show the probability of incoming weather – based on company weather triggers Storm track – testing phase Coming soon – – High winds, heavy precipitation, freezing line – Attempt to classify the probability of weather solutions by a “most probable”, “best case”, “worst-case” scenario 7
8
Ensemble Datasets Operational cyclone tracking website uses 4 ensembles – 21 member GEFS: Global Ensemble Forecast System – 21 member CMC: Canadian Meteorological Center Ensemble – 21 member SREF: Short Range Ensemble Forecast System – 10 member FNMOC: Fleet Numerical Meteorology and Oceanography Center (NOGAPS Ensemble) Forecasts update when data is available GEFS and SREF 4 times daily CMC and FNMOC 2 times daily 8
9
http://wavy.somas.stonybrook.edu/cyclonetracks/ 9
10
Significant Cyclone Track Archive Track historical cyclone cases using Hodges (1995) surface cyclone tracking scheme – Cyclone conditions: 24 h lifetime and 1000 km distance – TIGGE: THORPEX Interactive Grand Global Ensemble ECMWF, CMC, and NCEP ensembles utilized 00Z and 12Z MSLP data with 1˚x1˚ resolution Download and Convert Data Preprocess Data: Bandpass Filter Hodges Cyclone Tracking Calculate Cyclone Intensity Box Method Tracks Probability Shading Instantaneous Probability 10
11
Superstorm Sandy – 5 Day Forecast 11
12
Additional Parameters Are Being Tested Using GEFS Wind Speed Temperature Precipitation 12
13
Dec 2010 Blizzard – 120 hours 13
14
Dec 2010 Blizzard – 72 hours 14
15
Superstorm Sandy – 120 hours 15
16
Superstorm Sandy – 72 hours 16
17
Tropical Storm Irene – 120 hours 17
18
Tropical Storm Irene – 72 hours 18
19
February 2013 Blizzard – 120 hours 19
20
February 2013 Blizzard – 72 hours 20
21
Ensemble Decision Tool 2015 Phase 2 – Incorporate Historical Data – By using a historical data set of storms, the ensemble model can be “trained” toward pattern recognition – Probability estimates can be improved by comparing the forecast to past events 21
22
Questions? Brandon Hertell, CCM Meteorologist Con Edison Emergency Management hertellb@coned.com 212-460-3129 22
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.