16
Fuzzy Logic in Natural Resource Modelling William Silvert William Silvert Emeritus Research Scientist Bedford Institute of Oceanography Dartmouth, Nova Scotia

Fuzzy Logic in Natural Resource Modelling William Silvert Emeritus Research Scientist Bedford Institute of Oceanography Dartmouth, Nova Scotia

Embed Size (px)

Citation preview

Page 1: Fuzzy Logic in Natural Resource Modelling William Silvert Emeritus Research Scientist Bedford Institute of Oceanography Dartmouth, Nova Scotia

Fuzzy Logic in Natural Resource Modelling

William SilvertWilliam Silvert

Emeritus Research ScientistBedford Institute of

OceanographyDartmouth, Nova Scotia

Page 2: Fuzzy Logic in Natural Resource Modelling William Silvert Emeritus Research Scientist Bedford Institute of Oceanography Dartmouth, Nova Scotia

There is aProblem with Numbers

We usually express model results as numerical values, possibly with confidence limits.

The more rigorously we express these numbers, the more complicated we make things.

A really good scientific description of model output is unlikely to be adequately assimilated by managers and other client groups.

We have to get good answers and present them in a way that can be understood and used.

Page 3: Fuzzy Logic in Natural Resource Modelling William Silvert Emeritus Research Scientist Bedford Institute of Oceanography Dartmouth, Nova Scotia

TAC, Version 1

The TAC (Total Allowable Catch) for a fish stock is often presented in this way:

TAC = 42,000 tonnes

Page 4: Fuzzy Logic in Natural Resource Modelling William Silvert Emeritus Research Scientist Bedford Institute of Oceanography Dartmouth, Nova Scotia

TAC, Version 2

Of course we know that the TAC should be presented like this:

TAC = (42,000 ± 6,000) tonnes

Page 5: Fuzzy Logic in Natural Resource Modelling William Silvert Emeritus Research Scientist Bedford Institute of Oceanography Dartmouth, Nova Scotia

TAC, Version 3

Even better, the TAC should be presented as:

30 35 40 45 50 55

TAC

Pro

ba

bili

ty

Page 6: Fuzzy Logic in Natural Resource Modelling William Silvert Emeritus Research Scientist Bedford Institute of Oceanography Dartmouth, Nova Scotia

Model Output is More and More Complex!

The expression of model results gets more and more complex as it becomes more sophisticated.

A simple TAC is easy to grasp, but not very informative. It is just one number.

A TAC with a confidence interval is more informative, but there is an additional number.

The full Probability Distribution Function is most informative, but probably too complex for most fisheries managers to deal with.

Page 7: Fuzzy Logic in Natural Resource Modelling William Silvert Emeritus Research Scientist Bedford Institute of Oceanography Dartmouth, Nova Scotia

Discrete Model Output

Our conceptual understanding of model output tends to use discrete categories.

Although model output is usually numerical, when we look at the numbers we think “That is a very low TAC” or “It will be a good year for the fishing industry”. In other words, we map the numbers into discrete categories.

We should therefore consider using discrete categories for model output.

Page 8: Fuzzy Logic in Natural Resource Modelling William Silvert Emeritus Research Scientist Bedford Institute of Oceanography Dartmouth, Nova Scotia

Counter-Arguments

The main problem with using discrete categories is that it gives rise to artificial discontinuities.

1

TAC

Bad Medium GoodBad Medium Good

Page 9: Fuzzy Logic in Natural Resource Modelling William Silvert Emeritus Research Scientist Bedford Institute of Oceanography Dartmouth, Nova Scotia

The Fuzzy Solution

We can avoid these artificial discontinuities by using fuzzy categories, so that we can have a mixture of bad,

0%

50%

100%

1

TAC

Mem

ber

shipmedium, and

good to characterise model output.

Page 10: Fuzzy Logic in Natural Resource Modelling William Silvert Emeritus Research Scientist Bedford Institute of Oceanography Dartmouth, Nova Scotia

Flexibility

Fuzzy classification can be more flexible than numerical descriptions. For example, the bimodal Probability Distribution Function shown earlier,

is described as 40% bad, 25% medium, and 35% good. An error bar tells us less.

30 35 40 45 50 55

TAC

Pro

ba

bili

ty

Page 11: Fuzzy Logic in Natural Resource Modelling William Silvert Emeritus Research Scientist Bedford Institute of Oceanography Dartmouth, Nova Scotia

Other Considerations

Fuzzy Logic is ideally suited for including subjective information, such as expert but non-quantitative judgements about the quality of a natural resource (e.g., the taste of fish).

Different types of information, possibly contradictory, can be reconciled within the framework of Fuzzy Logic.

Missing data does not pose as great a problem as it does with more quantitative methods.

Page 12: Fuzzy Logic in Natural Resource Modelling William Silvert Emeritus Research Scientist Bedford Institute of Oceanography Dartmouth, Nova Scotia

Defuzzification The use of Fuzzy Logic is not a final

commitment — it is possible to convert fuzzy memberships into a simple number through a process known as “defuzzification”.

Many managers equate Fuzzy Logic with Fuzzy Thinking and resist it vigorously.

This may be why countries like Japan are so far ahead of North America in applications of Fuzzy Logic, such as steady-shot camcorders!

Page 13: Fuzzy Logic in Natural Resource Modelling William Silvert Emeritus Research Scientist Bedford Institute of Oceanography Dartmouth, Nova Scotia

Defuzzified TAC’s

For example, we can convert the fuzzy TAC (40% Bad, 25% Medium, 35% Good) into a single numerical value if we associate Medium with the range of values 40-45 kt, Bad with lower values under 40 kt, and Good with TACs over 45 kt. We get a defuzzified value from the weighted sum which is a reasonable approximation to the weighted mean TAC obtained from the Probability Distribution Function.

Page 14: Fuzzy Logic in Natural Resource Modelling William Silvert Emeritus Research Scientist Bedford Institute of Oceanography Dartmouth, Nova Scotia

Defuzzification Issues Defuzzification is always possible and lets us

hide the use of Fuzzy Logic if it doesn’t “sell”. Unfortunately, defuzzification also hides

useful information, such as the bimodal distribution in this example.

So defuzzify if you have to, but not if you can avoid it.

30 35 40 45 50 55

TAC

Pro

ba

bili

ty

Page 15: Fuzzy Logic in Natural Resource Modelling William Silvert Emeritus Research Scientist Bedford Institute of Oceanography Dartmouth, Nova Scotia

Model Validation Discrete predictions are easy to test,

since they are either right or wrong. Continuous prediction is inherently fuzzy. How do we test whether a result was predicted?

If we overfish and the TAC should have been 38 kt, can we say that the model was correct if the predicted value was (42 ± 6) kt?

What about (42 ± 2) kt?How do we evaluate models that make vague predictions against ones with more precision?

Page 16: Fuzzy Logic in Natural Resource Modelling William Silvert Emeritus Research Scientist Bedford Institute of Oceanography Dartmouth, Nova Scotia

Conclusions

Fuzzy Logic offers an alternative form of model output, and a mechanism for model testing.

It facilitates discrete classification and allows for a degree of subjective evaluation.

Numerical output may appear more scientific and objective, but this can be both misleading and impractical.

Additional advantages include means to merge contradictory data and even to deal with missing information.