Making sense of probabilistic flood forecasts –

decision support methods

Murray Dale Hydrometeorologist (murray.dale@ch2m.com)

• Probabilistic flood forecasting and why do it

• Decision-support: how those taking FIM decisions can

use probabilistic forecasts – outcomes of the

Environment Agency research project SC090032

Plan of presentation

What is probabilistic flood forecasting?

• In plain English: “What is the

chance of a flood happening?”

• Instead of saying it might flood / it

might not, a probabilistic forecast

means you can say: “There’s a

60% chance it will flood”

• And also: “…but there’s a 10%

chance it might be a really

damaging / dangerous flood”

• BUT… how do we know what to

do consistently with this

information?

Why bother?

• "Apparently a lady rang the BBC and said she

heard that there was a hurricane on the way. Well,

don't worry, if you're watching, there isn't."

Decision-support methods – project SC090032

6

Project SC090032 (2009-13)

Project outcomes

• easy-to-apply framework for practitioners

• case studies

• techniques for:

– a. Setting risk based thresholds for probabilistic flood forecasts

– b. Assessing the ‘costs’ and ‘benefits’ of relevant FIM actions

– c. Considering ‘soft’ factors in decision making

– d. Determining performance of probabilistic flood forecasts

1. Coastal surge: Colne Barrier

(Essex)

2. Coastal surge (&fluvial): Thames

Barrier, London

3. Fluvial using European Flood

Alert System model: River

Severn @ Bewdley

4. Fluvial using new Grid-to-grid

model: Cornwall (using floods of

17 November 2010 as case

event)

5. Surface Water using Extreme

Rainfall Alert system: multiple

locations

12

3

4

5

Case studies

Decisions on what FIM actions?

10 9 8 7 6 5 4 3 2 1 0.5 0

60 54 48 42 36 30 24 18 12 6 2 0

Monitoring & forecasting

Event preparation

On-site activities

Warning dissemination

Lead time (Large Catchment) in days

Lead time (Small Catchment) in hours

Deployment of staff to respond operationally to floods and/or monitor flooding in

communities

Routine & enhanced forecasting

Initiate enhanced monitoring

Flood Advisory Teleconferences

Staff Preparedness

Issue Severe Flood Warnings to public and

partners

Issue Flood Warnings to

professional partners

Issue Flood Warnings to

public

Flood awareness raising with public

Structural checks and watercourse

clearances

Deploy temporary and

demountable defences

Operate active control

structures

3 decision-support methods

• Basic method – use a probability threshold for a specific

FIM action based on judgement and local knowledge (e.g.

20%, 40%, 60%)

• Simplified method – use a probability threshold based on

the ratio of FIM action cost and the monetised benefit of that

action (a cost-benefit approach in the widest sense)

• Detailed method – establish a water level-impact

relationship for use in real-time and, in real-time, determine

whether average flood impact of the forecast water levels (if

no FIM action is taken) is greater than the FIM action cost

Detailed method – deriving average benefit

Detailed method – decision support

FIM Decision Support Tool

Date 01/05/2014 Time 12:22:42

Team XXX User XXX

Site/Community

Potential FIM action

Decision Support Method

Standard cost/benefit method Simple threshold method

Action cost £4,000 Probability threshold 0%

Forecast benefit £6,142 Forecast probability 100%

Initial recommendation Take action Initial recommendation Take action

Soft factors influencing the decision include:

1. Do you want to use this event as a practice or training event or as a PR exercise? [could change a ‘No’ into ‘Yes’]

2. Is the community at risk in danger of being desensitised (i.e. too many false alarms?) [could change a ‘Yes’ into ‘No’]

3. Is this a highly sensitive location with recent flooding? [could change a ‘No’ into ‘Yes’]

4. Have there been any missed flooding events (not forecast) at this site ? [could change a ‘No’ into ‘Yes’]

Final action decision

Justification

1 Forecast benefit comprises monetised impact of reduction in risk to life/serious injury, social impact, residential properties damage,

business/agriculture damage and infrastructure disruption.

Colne Barrier

D2. Operate active structures as necessary (e.g. close barriers)

Standard Method

Save As PDF

Load Probabilistic Forecast Result Data

Which method to pick?

• PFF forecast reliability (a good relationship between forecast

probability and observed frequency) – this includes reliability

of forecast models (hydrology & hydraulic)

• The cultural shift in forecasting & warning required

• Cost-benefit for decision making vs. the audit trail / process

Challenges

Colne Barrier events > 3.3mAOD: Forecast Surge Peak + Forecast Astronomical Peak(Forecast 1 = One tidal cycle [T + 5 to 12 hours]; Forecast 2 = Two or more tidal cycles [T + 12 to 36 hours])

2.5

2.7

2.9

3.1

3.3

3.5

3.7

0 1 2 3 4 5

Event Number

mA

OD

Actual Peak Deterministic Forecast

25/11/2007 12:15 (Ev1) 25/11/2007 12:15 (Ev1)

23/02/2008 01:30 (Ev2) 23/02/2008 01:30 (Ev2)

10/02/2009 12:30 (Ev3) 10/02/2009 12:30 (Ev3)

03/02/2010 02:45 (Ev4) 03/02/2010 02:45 (Ev4)

01/03/2010 12:30 (Ev5) 01/03/2010 12:30 (Ev5)

Upper Threshold Low er Threshold

Reliability - example

Summary

• We can’t escape uncertainty

• Objectivity in decision-making is

needed

• PFF decision-support benefits:

– An audit trail

– No such thing as a ‘false

alarm’!

– Enable better & earlier

forecasts, costs to be

saved

– Operate on all sources

(coastal, pluvial, fluvial)