Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn...

31
Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada

Transcript of Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn...

Page 1: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Fuzzy Traffic Light Methods

byW. Silvert, IPIMAR, Portugal

andP. Fanning, R. Halliday,

and R. Mohn DFO, Canada

Page 2: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Why are we doing this?

The first question to be asked about any approach is why it is needed.

SGPA Term of Reference D is to revise the description of PA concepts to make them more intelligible for non-fishery users.

The Traffic Light Approach is one of the types of descriptions currently under investigation to simplify the process of management decision-making.

Page 3: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

The Traffic Light Method

The Precautionary Approach (and Risk Management in general, not just for fisheries) requires that masses of complex data be presented clearly to managers, fishermen and other stakeholders.

The Traffic Light Method is an easily understood way of presenting information about stock conditions.

Page 4: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Indicators & Characteristics

We speak of Indicators, which are basic properties of the system, and Characteristics, which are integrated variables representing several Indicators.

Abundance is a typical Characteristic, since it represents the result of combining several Indicators, such as:Research Trawl dataVPA analysisCatch per Unit Effort

Page 5: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Standard Traffic Lights

Each Indicator or Characteristic is represented by a single traffic light, red, yellow or green in the standard traffic light representation.

There is no smooth transition, just two sharp lines separating red-yellow and yellow-green.

The meaning of the lights can be very sensitive to the location of these cuts.

Page 6: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Example: 4VsW Cod

A “crisp” traffic light indicatorsharp transitions between colours make

positioning the boundaries very criticalA lot of information is lost

Page 7: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Criteria for Improvement

The objective is to develop a more general approach with the following characteristics:

ResolutionUncertaintyWeighting

Page 8: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Resolution

The most serious problem with the standard traffic light method is the way that the lights change discontinuously when the Indicators change smoothly.

There is general agreement that there must be a more gradual representation of the significance of changing indicators.

Page 9: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Uncertainty

A less obvious point, but one which is clearly relevant to fisheries management, is the need to represent the degree of uncertainty in the interpretation of Indicators, and to provide a mechanism for expressing conflicting evidence or interpretation.

Page 10: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Weighting

It is also clear that not all Indicators are equally significant. They can be:

Of varying accuracyOf different relevanceOf dubious valueNew and untested

Page 11: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Alternative Approaches

Most alternatives to the standard traffic light method use some sort of averaging to show that an Indicator is on the border between red and yellow or between yellow and green.

One example is using intermediate colours, such as orange between red and yellow.

Page 12: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Fuzzy Traffic Lights

Fuzzy Sets offer one way to improve the standard traffic light method.

With fuzzy traffic lights an Indicator can correspond to more than one light.

For example, instead of using orange to show that an Indicator is on the red-yellow boundary, we can simply show both red and yellow lights.

Page 13: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Advantages of Fuzzy

Fuzzy traffic lights are continuous, we can switch between colours gradually to achieve higher resolution.

Fuzzy traffic lights show uncertainty if we illuminate several lights at once.

Fuzzy traffic lights can be weighted to show relative importance of indicators.

Page 14: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Memberships

The key idea behind Fuzzy Set Theory is that something can belong to more than one set at a time.

When we say that a light is red, that means that it belongs to the set “red”.

With fuzzy sets we can have a light that is 50% in set red and 50% in yellow.

Page 15: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Membership Example

Let the amount of each light displayed vary with the level of the indicatorTraffic Light Fuzzy Set

0

0.25

0.5

0.75

1

1.25

0 5000 10000 15000 20000 25000 30000 35000

Indicator

Deg

ree

Mean0.6*Mean

Strict Default Rules

Page 16: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Fuzzy Indicators

Use a combination of coloursgradual transitions show uncertainty and

contain more information than solid colour bars

Note that some bars have multiple colours

Page 17: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Application to Haddock

Note how much data is included on this figure, and how easy it is to see a pattern

Char Weight

Cod SSB Mana 1.0 Fraction over 42 Mana 1.0

Exploitation (%)(ages 5-10) Fish 1.0 Misaine Temperature Prod 0.5

Spring RV condition Prod 0.5 Spring RV 50% mat Prod 1.0

Summer RV growth age7(len) Prod 1.0 Summer RV Condition Prod 0.5

VPA Rec Prod 1.0 Density(1-29cm) Prod 0.5

Area occupied(1-29cm) Prod 1.0 VPA SSB Abun 1.0

Density(30cm+) Abun 0.5 Area occupied(30cm+) Abun 1.0

Sentinel (kg/set) Abun 1.0 Summer RV #/tow(42cm+) Abun 1.0

Summer RV #/tow(26-41cm) Abun 1.0

4TVW Haddock Summer RV #/tow(26-41cm)

0

15

30

45

60

75

1970 1975 1980 1985 1990 1995 2000

ManagementFishingM Production

Abundance

C:/1paulf/PA/hadd6c_fuzzy_traffic.txt

Page 18: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Application to White Hake

We have no VPA results, but we still can present an assessment

Page 19: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Current Developments

Char Weight

Temperature (Area >6C) Envi 1.0 Relative F Fish 1.0

Condition Factor Prod 0.5 July Area Occupied (<45cm) Prod 0.5

July Survey numbers (<45) Prod 1.0 July Survey Z Prod 1.0

July Survey mean weight Abun 0.5 July Area Occupied (>45cm) Abun 0.5 Halibut Survey numbers/set Abun 1.0

ITQ Survey numbers/set Abun 1.0 July Survey numbers (>45) Abun 1.0

4X/5 white hake July Survey numbers (>45)

0

6000

12000

18000

24000

1970 1975 1980 1985 1990 1995 2000

3 year trend

Environment Fishing Mortality

Production Abundance

C:/1paulf/PA/TLwhhakesimple4x_2.txt

Page 20: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Uncertain Reference LevelsWide yellow zone reflects uncertainty

Char Weight

Temperature (Area >6C) Envi 1.0 Relative F Fish 1.0

Condition Factor Prod 0.5 July Area Occupied (<45cm) Prod 0.5

July Survey numbers (<45) Prod 1.0 July Survey Z Prod 1.0

July Survey mean weight Abun 0.5 July Area Occupied (>45cm) Abun 0.5 Halibut Survey numbers/set Abun 1.0

ITQ Survey numbers/set Abun 1.0

4X/5 white hake ITQ Survey numbers/set

1.9

3.8

5.7

7.6

9.5

July Survey numbers (>45) Abun 1.0

1970 1975 1980 1985 1990 1995 2000

3 year trend

Environment Fishing Mortality

Production Abundance

C:/1paulf/PA/TLwhhakesimple4x_2.txt

Page 21: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Fuzzy Rules

The use of Fuzzy Traffic Lights to represent stock status means that we also use fuzzy rules to make management decisions.

Some typical (and familiar) fuzzy rules:IF it feels cold THEN light a fireIF you are hungry THEN eat something

Fuzzy rules are like crisp rules:IF the temperature falls below 14.7º C THEN

switch on the heater

Page 22: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Fuzzy Control of Fisheries

Fuzzy rules are of the form: IF (condition) THEN (act)

IF management= green AND production= green AND abundance= green

THEN tac_increment is large positive

IF production= green AND abundance= green THEN tac_increment is small positive

IF production= green AND abundance=yellow THEN tac_increment is no change

IF production=green AND abundance=red THEN tac_increment is small negative

IF production=yellow AND abundance=green THEN tac_increment is no change

IF production=yellow AND abundance=yellow THEN tac_increment is small negative

IF production=yellow AND abundance=red THEN tac_increment is large negative

IF production=red AND abundance=green THEN tac_increment is small negative

IF production=red AND abundance<>green THEN tac_increment is large negative

Page 23: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Displaying Fuzzy Lights

There are several ways to show a fuzzy traffic light:

Bubble charts, which look a lot like real traffic lights

Pie charts, which display information more quantitatively

Stacked bar graphs, which are less familiar but very effective

Page 24: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Bubble Charts

A Bubble Chart looks like a regular traffic light, but the sizes of the ”lights” are proportional to the membership in each of the three sets, red yellow & green.

Page 25: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Pie Charts

A pie chart looks less like a traffic light, but it gives a more quantitative picture of how much of each light is lit,

The area of each slice represents the fuzzy membership.

Page 26: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Stacked Bar Graphs

A stacked bar graph is somewhat like a traffic light with rectangular bulbs.

The area of each part of the bar represents the membership in the corresponding set.

Page 27: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Choosing the Display

The bubble chart resembles traffic lights most, but it does not give a good sense of the quantitative information about memberships.

The pie chart and the stacked bar graph both represent the relative memberships clearly.

Page 28: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Displaying Weighting

The bubble graph does not give a good idea of the relative weights of the different Indicators.

By varying the diameter of the pie charts or the width of the bar graphs we can show the relative importance of different indicators.

At present weighting has not been well implemented in trial applications and it is difficult to achieve agreement on it.

Page 29: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Comparison of Pie Charts

Page 30: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Comparison of Bar Graphs

Page 31: Fuzzy Traffic Light Methods by W. Silvert, IPIMAR, Portugal and P. Fanning, R. Halliday, and R. Mohn DFO, Canada.

Conclusions

Traffic Lights offer a clear way to present complex fisheries data.

Fuzzy Traffic Lights provide more information with little loss of clarity.