Perhaps your
attention has drifted over to football, but the aftermath of hurricane Isaac
has given way to another national sport.
Isaac was the category-1 storm that just pummeled New Orleans and
neighboring areas. Isaac landed close to the spot where Category-5 hurricane
Katrina landed, back in 2005. We now
know that Isaac’s destruction and havoc – and especially the rainfall - was
higher than what was expected. Many were
apparently unprepared for the storm’s intensity– as was evident after the fact,
even though they had taken the seeming necessary recommended precautions. Put simply, it seems many people were
unpleasantly surprised by Isaac –despite having received and acted on the
advance warnings provided by the authorities.
Predictably,
the chorus of complaints impugning the hurricane scale rankings has surfaced. And with the outcry one hears calls for
revising the hurricane warning scales – presumably, in a direction that will
convey better information. Even the New
York Times joined the fray – hosting a “debate” on the matter titled “How Could the Storm Ratings be
Improved?” I argue
that to replace the current system with a more complex one is
inappropriate. And I argue in fact, that
to advocate replacing the scale is to call attention to the wrong issue.
The by now
familiar hurricane scale is the Saffir-Simpson scale – a 5-point category
ordering originally created by Howard Safir.
The scale was later enhanced and popularized by Robert H. Simpson of the
National Hurricane Center. It was an
effort at prediction in a manner intended to inform decisions both by
individuals and by cities, states and regions.
Mr. Saffir’s ordering was a storm-destruction ranking system
keyed primarily on wind speed. To be
considered a Category 1 hurricane, the system’s wind speed must clock within
the 74-95 mph range. At the other
extreme are Category 5 monster cyclones where wind speed exceeds 155 miles per
hour. Three other thresholds between 74
and 155 mph winds separate Category 2-4 hurricanes.
Although it might seem somewhat obvious, the beauty of Mr.
Saffir’s innovation was not only in establishing the threshold (thereby
creating the categories) but also in associating each category with the damage
the winds may cause. To formally do this
he examined historical data. He documented
the positive relationship between wind-speed and property and other physical
damage. Thus, trees and unanchored
mobile homes receive the primary damage in a Category 1 storm whereas – at the
other end of the scale – a Category 5, involves the complete failure of roofs
and some structures. The other three descriptions
of destruction were then matched with the sustained wind speeds that would
produce the corresponding damage. What
follows then is to recognize that past is prologue.
What Mr. Safir and Mr. Simpson created was a model, a simplification
of a real event intended to convey decision-making information (material
destruction) based on a particular, distinctive attribute – like wind-velocity. By design,
because it is an abstraction of reality,
it will (practically) always be
mistaken. To expect a model to perfectly
convey as much information as the actual event it aims to describe is a
mistake. Jorge Luis Borges most
elegantly captured this fallacy in his Exactitude in Science. As an aside, it’s worth noting that this
beautiful story stands to be one of the most influential (in my opinion), if
not the most influential, in the history of the written word given the number
of words expended (it has less than 150 words).
Borges tells of a map-making competition where different generations of
cartographers aimed to surpass the previous generation by building more and
more accurate maps in the pursuit of the perfect map. They ended the map competition by building a
map that replicated the world. A map
that was useless.
In threshold setting – which is at the core of the
Safir-Simpson method and other ordered scale warning systems – there can be two
types of errors. Suppose that the
authorities announce a Category 1 storm (and to keep it simple, assume there
are only two levels). When the cyclone finally
arrives we find the property damage ends up being what would normally be
associated with a Category 2 hurricane (for sake of argument) – an unpleasant
mistake. On the other hand, suppose the
authorities announce a Category 2 hurricane – and when the winds are quieted we
find material damage of Category 1 magnitude.
The error, this time, is a more welcome one. In principle, we could move the thresholds
to try to maximize the “welcome” error.
Lower the wind-speed threshold to 80 miles per hour. Thus, any storm with winds between 75 and 79
miles per hour will be a Category 1 and now anything above 80 and 110 is a Category
2. But moving the thresholds we
encounter several unintended consequences. We would reduce the chances of
incurring the unpleasant mistake and enhance the chances of the welcome
mistake. But over time since everything is going to be a Category 2 hurricane –
any information contained in the warning is lost – it becomes meaningless –
like Borges’ map. Think about it – why not avoid unpleasant mistakes at all by
eliminating the various categories and only have 1 category. In this case you can only be pleasantly
surprised.
So what is being left out in the map-building – the
Saffir-Simpson scale – and thereby a potential source of error? There is more to a hurricane’s damage
potential that high winds. The storm
surge is wall of water at the leading edge of a storm – is especially
destructive in low lying areas. The
rainfall associated with storms presents flooding threats – especially in
regions with saturated soils or already swollen rivers. Two other things (at least) surge and rainfall
in addition to a cyclone’s wind speed are associated with the ultimate concern
material damage. The association between
these three measurable features of a cyclone and material damage – aside from
being positive – is different and distinct.
Thus, relying on one aspect of a hurricane wind-speed to categorize them
in a manner that tells us how much damage to expect will imperfectly capture
the association between the other aspects as well – practically guaranteeing a
source of error.
Will a more complex model – perhaps one that combines wind-speed,
surge and rainfall - do better? Not
likely – error and therefore judgment cannot be avoided.
Thus, a model builder has to tradeoff error and
accuracy. How much error? Alan Alda’s character in Nothing but the Truth arguing a point of law set forth the
parameters: “is this mistake like wearing white after labor day or is it like
invading Russia in winter?” Scale
constructors often opt to minimize errors – and associated costs of errors. And
to educate the population – so that we understand the nature of the underlying
decision-making model and thereby take its efficacy into account into our own
decision-making and importantly, ex post performance appraisal.
Arod
No comments:
Post a Comment