model jumpiness and the need for ensembles richard grumm national weather service office and lance...

26
Model Jumpiness and the Model Jumpiness and the Need for Ensembles Need for Ensembles Richard Grumm Richard Grumm National Weather Service Office National Weather Service Office and and Lance Bosart Lance Bosart State Univesity of New York at Albany State Univesity of New York at Albany

Upload: bartholomew-snow

Post on 18-Jan-2016

217 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

Model Jumpiness and the Need for Model Jumpiness and the Need for EnsemblesEnsembles

Model Jumpiness and the Need for Model Jumpiness and the Need for EnsemblesEnsembles

Richard GrummRichard GrummNational Weather Service OfficeNational Weather Service Office

and and Lance BosartLance Bosart

State Univesity of New York at AlbanyState Univesity of New York at Albany

Richard GrummRichard GrummNational Weather Service OfficeNational Weather Service Office

and and Lance BosartLance Bosart

State Univesity of New York at AlbanyState Univesity of New York at Albany

Page 2: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

OBJECTIVESOBJECTIVESOBJECTIVESOBJECTIVES

• Get a fix on some aspects of uncertainty and be able to Get a fix on some aspects of uncertainty and be able to recognize uncertainty in forecasts.recognize uncertainty in forecasts.

• Provide examples of uncertainty in the NCEP GFS and its Provide examples of uncertainty in the NCEP GFS and its impact on the MREF and SREF ensemble prediction impact on the MREF and SREF ensemble prediction systems via examples of model jumpiness.systems via examples of model jumpiness.

• Define model jumpinessDefine model jumpiness as the changes or differences of as the changes or differences of forecasts of features and parameters from one run-to-run forecasts of features and parameters from one run-to-run from a single numerical model. These inconsistencies may from a single numerical model. These inconsistencies may span intensity, gradient, and location of a feature or span intensity, gradient, and location of a feature or parameter.parameter.

Page 3: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

Model Jumpiness through the eyes of a Model Jumpiness through the eyes of a model or prediction systemmodel or prediction system

Model Jumpiness through the eyes of a Model Jumpiness through the eyes of a model or prediction systemmodel or prediction system

• We all see We all see uncertaintyuncertainty in deterministic models on in deterministic models on a daily basis. Some common include:a daily basis. Some common include:

– Significant run-to-run differencesSignificant run-to-run differences

– The NAM or GFS may change the track and intensity of a cyclone or The NAM or GFS may change the track and intensity of a cyclone or frontal system.frontal system.

– The precipitation shield shifts to the east (north) or west (south). The precipitation shield shifts to the east (north) or west (south).

– The problem is typically The problem is typically – worse at longer forecast range though not always.worse at longer forecast range though not always.– A function of scale and mesoscale details can be quite changeableA function of scale and mesoscale details can be quite changeable

– Like the rain snow line or area for heavy snow/rainLike the rain snow line or area for heavy snow/rain

• Not to mention differences between different Not to mention differences between different models!models!

Page 4: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

4 GFS Runs 4 GFS Runs Big cyclone disappears? Big cyclone disappears?All images valid the same time!All images valid the same time!

4 GFS Runs 4 GFS Runs Big cyclone disappears? Big cyclone disappears?All images valid the same time!All images valid the same time!

Big cyclone most of PA might be rain.

Weak storm…PA is now cold…snow in SE?

Page 5: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

Return of the cyclone?Return of the cyclone?all images valid the same time!all images valid the same time!

Return of the cyclone?Return of the cyclone?all images valid the same time!all images valid the same time!

Page 6: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

Things cannot get so bad so fast!Things cannot get so bad so fast!Or can they as Robin might say “Holy short-wave Batman”Or can they as Robin might say “Holy short-wave Batman”Things cannot get so bad so fast!Things cannot get so bad so fast!Or can they as Robin might say “Holy short-wave Batman”Or can they as Robin might say “Holy short-wave Batman”

Page 7: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

Getting closer to event timeGetting closer to event time…still lots of…still lots of uncertainty uncertainty

Getting closer to event timeGetting closer to event time…still lots of…still lots of uncertainty uncertainty

Page 8: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

Well…it passed to our West!Well…it passed to our West!Warm windy winter rainWarm windy winter rain

Well…it passed to our West!Well…it passed to our West!Warm windy winter rainWarm windy winter rain

Page 9: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

““Jump” right into some pointsJump” right into some points““Jump” right into some pointsJump” right into some points

• The GFS showed run-to-run inconsistenciesThe GFS showed run-to-run inconsistencies

– These inconsistence These inconsistence uncertainty.uncertainty.– CausesCauses same model each time suggests uncertainty in the initial same model each time suggests uncertainty in the initial

conditions. The need for multiple sets of IC’sconditions. The need for multiple sets of IC’s

• Significant impacts on sensible weather Significant impacts on sensible weather elementselements..

– Areas and amounts of rain or early on, snowAreas and amounts of rain or early on, snow– POPS and temperatures POPS and temperatures – Winds to include direction changes of over 180 degrees!Winds to include direction changes of over 180 degrees!

• We need to acknowledge, visualize, and be deal We need to acknowledge, visualize, and be deal with uncertainty and quantify it.with uncertainty and quantify it.

• Do you think this case is unique?Do you think this case is unique? It happened It happened within 7 day of this eventwithin 7 day of this event and it and it does all the timedoes all the time!!

Page 10: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

Weather on 12 Feb 2006?Weather on 12 Feb 2006?In Washington, DC and NYC pick clouds, wind direction and PTYPEIn Washington, DC and NYC pick clouds, wind direction and PTYPE

Weather on 12 Feb 2006?Weather on 12 Feb 2006?In Washington, DC and NYC pick clouds, wind direction and PTYPEIn Washington, DC and NYC pick clouds, wind direction and PTYPE

Light winds Possibly precip

Rain?Rain/Snow wind?

Page 11: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

Pleasant NW winds or a NE gale?Pleasant NW winds or a NE gale?…and we want those winds in 3-hour increments….…and we want those winds in 3-hour increments….

Pleasant NW winds or a NE gale?Pleasant NW winds or a NE gale?…and we want those winds in 3-hour increments….…and we want those winds in 3-hour increments….

Whale storm

Major East Coast Storm

Details of center location and pressure still varyl

Page 12: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

So there will be a stormSo there will be a stormbut look at the variation of the depth and locationbut look at the variation of the depth and location

So there will be a stormSo there will be a stormbut look at the variation of the depth and locationbut look at the variation of the depth and location

Details still uncertain

Over Cape Cod

No

Make that south of

Page 13: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

Large East Coast Storm solutionLarge East Coast Storm solutionBut where will it snow big and not at all…winds for RI pleaseBut where will it snow big and not at all…winds for RI please

Large East Coast Storm solutionLarge East Coast Storm solutionBut where will it snow big and not at all…winds for RI pleaseBut where will it snow big and not at all…winds for RI please

At finer scales the devil is in the details.

Page 14: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

““Jump” right into more pointsJump” right into more points““Jump” right into more pointsJump” right into more points

• Run-to-Run inconsistencies Run-to-Run inconsistencies – even at 6-hr incrementseven at 6-hr increments– Close in we got the Big StormClose in we got the Big Storm– But we had problems with the location and intensityBut we had problems with the location and intensity

• Still hard to get the details nailed down Still hard to get the details nailed down – Winds direction and rain snow line looked elusiveWinds direction and rain snow line looked elusive– Did not show QPF but it too must was hard to nail Did not show QPF but it too must was hard to nail

down.down.

• At the smaller scales, States and Counties the At the smaller scales, States and Counties the details due to jumpiness still remain elusive.details due to jumpiness still remain elusive.

Page 15: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

MREF forecasts-06UTC 8 FebMREF forecasts-06UTC 8 FebMREF forecasts-06UTC 8 FebMREF forecasts-06UTC 8 Feb

Page 16: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

MREF forecasts-12UTC 8 FebMREF forecasts-12UTC 8 FebMREF forecasts-12UTC 8 FebMREF forecasts-12UTC 8 Feb

Page 17: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

MREF forecasts-12UTC 9 FebMREF forecasts-12UTC 9 FebMREF forecasts-12UTC 9 FebMREF forecasts-12UTC 9 Feb

Page 18: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

MREF Comparative QPFMREF Comparative QPFprecipitation shield is moving east!precipitation shield is moving east!MREF Comparative QPFMREF Comparative QPFprecipitation shield is moving east!precipitation shield is moving east!

Page 19: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

SREF 09 and 21 UTC 9 FebSREF 09 and 21 UTC 9 Febprecipitation shield is moving east!precipitation shield is moving east!SREF 09 and 21 UTC 9 FebSREF 09 and 21 UTC 9 Febprecipitation shield is moving east!precipitation shield is moving east!

Page 20: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

Coastal stormCoastal stormor an offshore track even or EPS has issuesor an offshore track even or EPS has issues

Coastal stormCoastal stormor an offshore track even or EPS has issuesor an offshore track even or EPS has issues

Page 21: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

A few more pointsA few more pointsA few more pointsA few more points

• The MREF & the GFS showed run-to-run The MREF & the GFS showed run-to-run inconsistencies.inconsistencies.– But had a cyclone in its solutions that could affect the coast But had a cyclone in its solutions that could affect the coast

before the single GFSbefore the single GFS– It slowly converged on a solution about T-4 days.It slowly converged on a solution about T-4 days.

• The impacts on the forecast were significant The impacts on the forecast were significant even the SREF had trends and moved the even the SREF had trends and moved the threat areathreat areaEASTEAST– Sunny NW winds or rain…or snowSunny NW winds or rain…or snow– It was not too clear where it would snow until about T-2days!It was not too clear where it would snow until about T-2days!

• The cases of 5 and 12 February are NOT The cases of 5 and 12 February are NOT uniqueunique– They are ubiquitousThey are ubiquitous

Page 22: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

30 August GFS forecast heavy rains-30 August GFS forecast heavy rains-Front and ErnestoFront and Ernesto

30 August GFS forecast heavy rains-30 August GFS forecast heavy rains-Front and ErnestoFront and Ernesto

Page 23: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

31 August GFS Heavy rain forecasts31 August GFS Heavy rain forecastsaxis/location heavy rain and end timeaxis/location heavy rain and end time

31 August GFS Heavy rain forecasts31 August GFS Heavy rain forecastsaxis/location heavy rain and end timeaxis/location heavy rain and end time

Page 24: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

ConclusionsConclusionsConclusionsConclusions

• There is considerable uncertainty in There is considerable uncertainty in weather forecastingweather forecasting– Model jumpiness is a signalModel jumpiness is a signal– Model differences are signalsModel differences are signals– Ensembles help us identify these signalsEnsembles help us identify these signals

• Model uncertaintyModel uncertainty– Due to initial conditions and data are not unique.Due to initial conditions and data are not unique.– We deal with them at various forecast lengths, and We deal with them at various forecast lengths, and

meteorological scales.meteorological scales.– We see these problems on a daily basis.We see these problems on a daily basis.

Page 25: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

HubrisHubrisooverbearing pride or presumption; arroganceverbearing pride or presumption; arrogance::

HubrisHubrisooverbearing pride or presumption; arroganceverbearing pride or presumption; arrogance::

• The 13 March 1993 storm was a relative The 13 March 1993 storm was a relative successsuccess

• It gave us confidence in modelsIt gave us confidence in models

• How big a success was it?How big a success was it?– Lucky at the scales presented.Lucky at the scales presented.– We still have storms that are hard to predictWe still have storms that are hard to predict– The mesoscale details are even harder to get rightThe mesoscale details are even harder to get right..

• High confidence and precise forecasts are High confidence and precise forecasts are quite likely quite likely hubris.hubris.

Page 26: Model Jumpiness and the Need for Ensembles Richard Grumm National Weather Service Office and Lance Bosart State Univesity of New York at Albany Richard

NWA- Probabilistic Forecasting

Questions?Questions?Questions?Questions?