“probability forecast use” - initial findings - jan verkade edwin welles february 2012

39
“Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

Upload: oscar-dean

Post on 12-Jan-2016

214 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

“Probability forecast use”- initial findings -

Jan Verkade

Edwin Welles

February 2012

Page 2: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Outline

• Probability forecast study – overview• Initial findings• Summary and conclusions

Page 3: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

But first: thanks to…

Page 4: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Thank-you to

• Diane Cooper, Service Hydrologist @ Chanhassen WFO• Reggina Cabrera, Chief, Hydrologic Services Division @ NWS

Eastern Region• John Blood, Senior Planner @ State of Minnesota Homeland

Security Emergency Management Services• Allen Glass, Bruce Elder, Tom Miller and Heather Winkleglass @

City of St Paul Emergency Management / Dept of Public Works• Scott Jutila and Patrick Foley @ USACE, St Paul District• Chris Franks, Meteorologist @ Chanhassen WFO

Page 5: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Thank-you to

• Greg Kruse, Chief of Hydrologic Monitoring Unit @ MN Dept of Natural Resources

• Marc Deutschmann, VP @ Houston Engineering• Greg Spoden, Climatologist @ State of Minnesota Climatology

Office, University of Minnesota• Walter Potts @ ESRI• Bill McAuliffe, Jim Kern, Ray Grumney and David Braunger @ Star

Tribune • Jim Bodensteiner and Jeff Berrington @ Xcel Energy

Page 6: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Last, but certainly not least, many thanks to…

Steve BuanService Coordination Hydrologist @ NWS NCRFC

Page 7: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Probability forecast study – purpose

• General move towards probabilistic forecasting, varying reasons• To capture perceived benefits, simply having the probability

forecast is not enough• Additional effort may be

required:

• visualisation

• communication

• decision-making

• verification

• “downstream” DSSs

• adaptation of businessprocesses / procedures

Page 8: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Probability forecast study – method

• Desk research

• Forecasting exercise with Dutch Waterboard

• Interviews with

• US NWS forecasters and forecast users

• Dutch Meuse River forecasters and forecast users

• Scottish Environment Protection Agency forecasters and forecast users

• Possible wrap up workshop with all participants

Page 9: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

Findings

Page 10: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Page 11: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Page 12: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Page 13: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Preliminary findings: themes

• Visualisation and communication• Reflection of physical processes• Decision making• Training and education• Use of and rationale for probability forecasting

Page 14: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Why people use forecasts

• planning• planning• planning• awareness raising• decision making

Page 15: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Preliminary findings: visualisation

• Users may have issues with visualisation of probability forecasts• Main issue: probability forecasting adds a dimension to the

problem• In addition to:

• space (2 dimensions already)

• time

• a variable such as either flow or stage• A 5th dimension is added: probability

visualisation not trivial

Page 16: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Preliminary findings: visualisation• Suggested approach for visualisation:• As we can visualise only 2 (graphs) or 3 (maps) dimensions, we

need to “fix” the others• Choices for fixing vary

by user and use

this “dimensionality issue” requires somewhat sophisticated tools to visualise probability forecasts(e.g. GIS, programming languages)

Page 17: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Preliminary findings: communication

• Ties in with visualisation• Additional issue: “flow” and “stage” may not be meaningful to many

end users• Variables such as “inundated area”, “number of properties

affected”, “reference floods” may be effective to communicate risks

may require post-processing

Page 18: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Preliminary findings: physical processes

• Often heard: users compare forecasts with their knowledge of the physical system

• This is quite difficult when “abstract probabilities” come into play these cannot be confirmed with knowledge of hydrology

• some indication may be needed that the physics are okay

not an easy task; maybe solution is found inprovision of metadata

providing information on past performanceof probability forecasts will be helpful(for this: forecast verification is required)

Page 19: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Preliminary findings: decision-making

• Decisions are made based on a consequence, not on a hazard

• Must transform probabilities to make a binary decision

> Often times: intuition only

• Decision criteria can now include explicit expression of risk

• Risk = Probability x Consequence

• NWS concentrates on forecasting a possible hazard; decision support is needed to translate this into meaningful consequences

Page 20: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Preliminary findings: training and education

• Information overload is a real issue

• Trying to understand the forecast is not conducive to good decision making

• This should then be done prior to a “crisis situation”• Links between “hydrological variables” and real issues not always clear

for non-experts• Everyone found the NWS webinars extremely useful• Need to explain how to access the information in the forecasts by fixing

a dimension.

Page 21: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Preliminary findings: why probabilities

• Rationale for moving towards probability forecasting:

• “More realistic forecasts” widely accepted> Deterministic forecasts are “over confident”

• “Risk based” decision making easier said than done> Support systems not generally in place

• Benefits may not reside with both forecaster and end user> Forecasters provide more complete information> Users must do more interpretation and post-processing

Page 22: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

Some ideas…

Page 23: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Decision support: from outlooks to planning (1)

Exc.probs Max stage[-] [ft]

98% 1395% 1490% 1580% 1750% 2520% 3210% 345% 352% 381% 40

Page 24: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Decision support: from outlooks to planning (2)

Determine potential consequences from inundation

maps

Do this for all relevant stages

Page 25: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Decision support: from outlooks to planning (2)

Page 26: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Decision support: from outlooks to planning (2)

Max stage #houses affected[ft] [-]13 014 015 017 025 032 534 2535 5738 10040 250

Page 27: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Max stage #houses affected[ft] [-]13 014 015 017 025 032 534 2535 5738 10040 250

Exc.probs #houses affected[-] [-]

98% 095% 090% 080% 050% 020% 510% 255% 572% 1001% 250

Decision support: from outlooks to planning (3)Exc.probs Max stage

[-] [ft]98% 1395% 1490% 1580% 1750% 2520% 3210% 345% 352% 381% 40

Page 28: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Decision support: from outlooks to planning (4)#houses affected [-]

0

50

100

150

200

250

300

0%10%20%30%40%50%60%70%80%90%100%

Exceedence probability [-]

Hou

ses

affec

ted

[ft]

Page 29: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

Another idea…

Page 30: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

36 Serious flood damage occurs at the University of Iowa campus.

34 Flood protection becomes necessary at the University of Iowa. Water affects several industrial businesses and warehouses along Commercial Drive.

32 Water floods county road bridge approaches along the river and affects streets and parking lots along Commercial Drive.

30 Flooding occurs in Coralville. Water inundates the Cedar Rapids and Iowa City Rail line near Coralville.

25.5 Water affects the north Wastewater Treatment Plant. Considerable flooding occurs at the University of Iowa.

25 Water floods Hancher Auditorium at the University of Iowa. Flooding problems occur elsewhere on the University of Iowa campus.

22 Urban flood damage occurs in Iowa City. Water enters homes along Edgewater Drive.

21 Flooding occurs in homes near Taft Speedway in Iowa City. Homes on east side of Quarry Road in Coralville require protection. Edgewater Drive becomes impassable.

20.5 Water affects the north Wastewater Treatment Plant.

19 Lowland flooding occurs in the Iowa City Park area.

Gather the impacts.

Page 31: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Assign probabilities to the stages.

21 ft 70%

25 ft 50%

30 ft 30%

34 ft 10%

Page 32: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

36 Serious flood damage occurs at the University of Iowa campus.

34 Flood protection becomes necessary at the University of Iowa. Water affects several industrial businesses and warehouses along Commercial Drive.

32 Water floods county road bridge approaches along the river and affects streets and parking lots along Commercial Drive.

30 Flooding occurs in Coralville. Water inundates the Cedar Rapids and Iowa City Rail line near Coralville.

25.5 Water affects the north Wastewater Treatment Plant. Considerable flooding occurs at the University of Iowa.

25 Water floods Hancher Auditorium at the University of Iowa. Flooding problems occur elsewhere on the University of Iowa campus.

22 Urban flood damage occurs in Iowa City. Water enters homes along Edgewater Drive.

21 Flooding occurs in homes near Taft Speedway in Iowa City. Homes on east side of Quarry Road in Coralville require protection. Edgewater Drive becomes impassable.

20.5 Water affects the north Wastewater Treatment Plant.

19 Lowland flooding occurs in the Iowa City Park area.

70%

50%

30%

10%

Assign probabilities to the stages.

Page 33: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

And some ideas about visualisation…

Page 34: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Recommendations: map-type visualisation

stage / flow

high/medium/lowexceedence prob

“Play” button fortime dimension

Fix any continuous variable;or define the event otherwise

Page 35: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Page 36: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Recommendations: graph-type visualisation

Fixed:•location•probability

Va

ria

ble

A

time

prob

abili

ty

timeFixed:•location•variable / event

variable = f(time) probability = f(time)

Page 37: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Conclusions• Multi-dimensional nature of probability forecasts makes them hard

to understand, (as opposed to the probabilities being too complicated).

• Everyone needs to fix at least one dimension to make sense of the probability forecasts

• Impacts/ consequences

• Planning Scenarios

• Analagous Events

• Background information on the

forecast origins helps. So does verification.

• Probabiilty forecasts are more informative! And useful!

Page 38: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

February 10, 2012

Contact details

Edwin Welles

[email protected] , 301-642-2505

Jan Verkade

[email protected] , +31.88.335.8348

On the web:

http://www.deltares.nl/en

This presentation can be downloaded from:

http://publicwiki.deltares.nl/display/~verkade

Page 39: “Probability forecast use” - initial findings - Jan Verkade Edwin Welles February 2012

Thank you for your attention