Terminal Ceiling & Visibility PDT AWRP Program Management Review 19 November 2002 Dave Clark PMR_Nov_02-1
Terminal C&V Project Areas SFO Marine Stratus Forecast System - 2002 Activities - Continued development NYC Winter Weather C&V - Scope - Initial Efforts PMR_Nov_02-2
Stratus Impact on SFO Approach Mt. Diablo Golden Gate Bridge Hunters Pt. Almeda NAS 6000 Oakland Intl. Hayward 3500 Cedes Intl. 11000 7 AM San Francisco Intl. San Mateo Bridge 7000 8000 Briny Intl. Halfmoon Bay San Carlos Porte Intl. 8000 6000 Palo Alto 4000 Moffett NAS San Jose Intl. Reid Hillview 9 AM Offshore to So. California 13000 Pesca Intl. San Francisco Bay Area Bay Approach Major Jet Arrival and Departure Routes PMR_Nov_02-3 Arrivals Departures 11 AM
System Sensing Network SODAR Sensor Suite Fort Funston S. F. Bay Oakland SFO Pacific Ocean Half Moon Bay San Carlos ARTCC Surface Weather Observation SODAR (Acoustic Sounder) Pyranometer (SW Radiation) Sonic Anemometer Radiosonde (Weather Balloon) San Jose PMR_Nov_02-4
PMR_Nov_02-5
2002 Highlights Major effort over winter to automate model development process - Allows testing of many more predictors and paths - Facilitates development at other airports (e.g. NYC) Upgraded SODARs prior to start of demonstration 2002 Demonstration June through October - Excellent SODAR performance - Two versions of models were implemented PMR_Nov_02-6
2002 SODAR Upgrade Issue: Control of audio signal based on aging technology; longterm maintainability problem. Poor signal quality at SFO Solution: Replace with PC sound card technology SFO SODAR 2001 SFO SODAR 2002 PMR_Nov_02-7
Forecast Model Development COBEL Physical Column Model Key Inputs: High resolution temp, humidity, wind, solar radiation Consensus Forecast Local Statistical Forecast Model Key Inputs: Local measurements of cloud base and inversion height, wind, pressure Regional Statistical Forecast Model Key Inputs: Regional hourly surface observations, 12Z Oakland sounding Satellite Statistical Forecast Model Key Input: GOES visible satellite PMR_Nov_02-8
Statistical Model Development Process Data Re-scale Predictors Select potential predictors (up to several hundred). Re-scale nonlinearly based on correlation with predictand (side-by time). Predictor Nulling Examine correlations and redundancies to reduce establish pool of best candidate predictors. (20 predictors) All-Paths Testing Fit all combinations of 20 best predictors (~ 1 million forecast equations). For each models size (# of predictors, 1-20), order by best fit. Select best 200 of each size. (4000 equations) Crossvalidation Cross-validate (10 samples). Sort, eliminate all fits worse than 1-minute error from best fit. Rank remaining equations by error and stability. (10-100 equations) Final Selection Select from candidates based on preferred predictors (e.g. based on predictor availability, etc.) PMR_Nov_02-9
2002 Forecast Models 2001 2001A 2001B 2002 2002A 2002B Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Operational Demo Automation Process Developed Operational Demo Version 2002A models implemented in June -Prototype version from automated processing -Some inadequecies were revealed New models (2002B) implemented mid-august - Improvements to correct model over fitting - Case selection during development process - Reduction in overestimate of forecast confidence (impacts model weighting) -1 st overnight forecasts (RSFM at 09z &11z) PMR_Nov_02-10
Model Development Historical Performance Consensus Forecast, 2002 Data Median Absolute Error (hours) 13z 14z 15z 16z 17z 18z 2001B 0.91 0.83 0.59 0.61 0.60 0.74 2002A 1.02 0.71 0.63 0.64 0.53 0.62 2002B 0.81 0.76 0.62 0.43 0.67 0.53 PMR_Nov_02-11
Intermodel Comparison 2002 Data, Model Version 2002B Median Absolute Error (hours) 09z 11z 13z 14z 15z 16z 17z 18z Consensus 0.81 0.76 0.62 0.43 0.67 0.53 COBEL 1.00 0.89 0.79 0.75 0.84 1.00 Local SFM 0.96 0.91 0.58 0.62 0.64 0.63 Regional SFM 1.00 1.00 0.96 1.01 0.91 0.66 0.67 1.09 Satellite SFM 0.68 0.70 0.58 0.63 0.53 PMR_Nov_02-12
Objectives for Forecast Upgrades Extension of some products to overnight hours -Zi algorithm - COBEL and Local SFM Investigate improvements to Satellite & Regional SFMs Objective Identification of pro-active decision opportunities CONSENSUS FORECAST 16z Approach Clear at 17:25 GMT Run Confidence MOD Probability of Clearing By: 17Z 18Z 19Z 20Z 65% 90% 93% 95% COMPONENT FORECASTS Run Model Fcst Conf 16z COBEL 18:00 MOD 16z Local 17:50 MOD 16z Regional 17:45 MOD 16z Satellite 16:55 MOD CWSU - REPORT RATE CHANGE Display Modification Concept PMR_Nov_02-13
Northeast Winter C&V Prediction Joint effort with National C&V and Winter Wx PDTs NYC area to act as testbed Initial user meeting held on Long Island in March - Air Traffic Managers - NWS Forecasters (CWSU, FO, and Region) - Airlines: forecasting and dispatch - Researchers PMR_Nov_02-14
Northeast C&V Initial efforts Winter 2002-03 to survey needs to steer development - Web-based Discussion Board: Air traffic managers, forecasters, product developers - Investigate application of NCEP model data Sensor testbed deployments at Brookhaven & Rutgers Initial product development - Precipitation visibility nowcast based on WSSDM technology - Regional SFM based on SFO technology PMR_Nov_02-15
Summary Completion of 2002 Demonstration at SFO SODAR upgrades Statistical model development automation Two versions of models tested Continuing work at SFO Final model changes Extension of products to overnight hours Development of proactive decision aid Initial efforts beginning on Winter Weather C&V (NYC) PMR_Nov_02-16