Why do 100 year events happen so often?
By Paul Homewood
According to Mickey Mann, it’s because of climate change.
But back in 1990, scientists knew the truth, as this article from the BOM explains:
Analyses of data from rainfall gauges and the use of statistical theory enables one to estimate the probability that a particular rainfall depth will be equalled or exceeded at a particular place, within a particular time interval (duration), and over any given period of time. For example, the probability that 48.2 mm or more will fall in any 1-hour duration in a period of one year at the site of the Bureau’s official Melbourne raingauge. Curves representing these values are known as rainfall intensity-frequency-duration (IFD) curves. Rainfall IFD analyses are available for all locations in Australia.
The probability of a particular rainfall amount for a specified duration being equalled or exceeded in any 1 year period can be expressed as a percentage (the annual exceedence probability or AEP) or as “on the average once in every x years” (an average recurrence interval, or ARI, of x years). As an example, for Melbourne, a rainfall amount of 48.2 mm in 1 hour can be expected to be equalled or exceeded on average once every 100 years. In this case, the ARI is 100 years and the AEP is 1%. It is important to note that an ARI of, say, 100 years does not mean that the event will only occur once every 100 years. In fact, for each and every year, there is a 1% chance (a 1 in 100 chance) that the event (in this example, 48.2 mm in 1 hour) will be equalled or exceeded (once or more than once). The use of annual exceedance probability (AEP) to describe the chance of a particular rainfall is preferred as it conveys the probability or chance that exists for each year. The alternative, ARI, is a term which is easily misunderstood.
Statistical theory (see for example, Laurenson, E.M, “back to basics on flood frequency analysis”, Civ Eng Trans, 1987) can be used to calculate the chance this (1% AEP) value has of occurring or being exceeded over a longer period, say the 50 years from 2001 to 2050. This chance is calculated to be about 40%. That is, a 1% AEP (100 year ARI) event has a high chance (40%, or between a 1 in 2 chance and a 1 in 3 chance) of being equalled or exceeded over a 50 year period. Similarly, if a storm water drain is designed to cope with a 10% AEP (10-year ARI) rainfall event there is a 10% chance of it’s being overtopped next year, a 39% chance in the next 5 years, and 63% chance in the next 10 years.
The effect of area
The above discussion applies to a particular point or location. If a larger area is considered (say, the whole of a city or town), the chance of receiving an intense rainfall event (say a 1% AEP event) somewhere over the larger area is increased. In practice, the most intense part of a thunderstorm, or of a cold frontal band, is a small area, across which the rainfall values are almost the same. Values for one hour duration taken at points about 5 km apart would be almost totally independent of each other in some storms but partly related in others. For example, one could make a grid of about 100 such points in the greater Melbourne area. The 1-hour, 1% AEP (100 year ARI) value has 1% chance of occurring at each of these points in a particular year, which means that there is a good chance (63% assuming full independence of points) of a 1-hour, 1% AEP (100 year ARI) event occurring somewhere in the general Melbourne area in each calendar year. (This is an approximation only as the assumption of independence of – zero correlation amongst – points varies with both the duration of rainfall under consideration and the type of rainfall-producing weather mechanisms which can operate in the area.)
Add events of different duration
The above reasoning can be repeated using different durations for rainfalls of a given probability. If a storm occurred, it would be most severe in terms of rainfall intensity and expected probability of occurrence for some particular duration, e.g. it may be a 0.5% AEP (200-year ARI) event at a duration of 1 hour but a 2% AEP (50 year ARI) event for a 30 minute duration, and a 1% AEP (100 year ARI) event for 2 hours duration. On another occasion a storm at a particular locality may peak at a different duration. At any particular place there is a 1% chance in any one year of experiencing a rainfall intensity equal to or greater than the 100 year ARI value for a number of different durations for which the rainfalls are independent, or nearly so. For example, durations of: 5 minutes, 1 hour, 6 hours, and 3 days may display a considerable degree of independence.
Given this range of durations from 5 minutes to 3 days, if one talks simply about the “100-year event” (regardless of the duration at which it occurs) then its probability of occurrence somewhere over a large area, say the greater Melbourne area, is increased significantly above the 63% estimate above – which refers to one duration only.
Thus the reason that some people may perceive the “100-year event” to occur more frequently than its name implies is that, instead of focussing on a single point (raingauge), and a single duration, they include in their considerations an area of significant size and a wide range of durations.
Based on an article by Malcolm Kennedy (1990)
http://www.bom.gov.au/water/designRainfalls/rainfallEvents/why100years.shtml
Comments are closed.
NOT A LOT OF PEOPLE KNOW THAT 
Excellent: thank you. Now, where to bookmark that?
Any chance of getting Harrabin, Shukman and McGrath to read it too?
How often do you bet on rank outsiders (using crutches before the race starts)?
As followers of the Discworld know, Million to one chances come off nine times out of ten.
http://wiki.lspace.org/mediawiki/Million-to-one_chance
The whole article is bogus (apologies to all).
Studying our past can only ever be a rough guide to our future and certainly does not enable statistical predictions to be made, will we ever learn?
The article is genuine. Your objection seems to be to the statistical methodology. Considering it’s probability, not certainty, I am uncertain why the objection. It’s being used as “a rough guide” and really is a very, very rough estimation. Presumably, you object to putting probability on a throw of the dice or flipping a coin, since this is basically the same thing.
In a sense you are correct if there is no allowance for changes made by man. Going back to the infamous Somerset Levels floods, a statistical analysis of records would not of seen this coming. The flooding was due to deliberate non-action by government agencies and wildlife charities. Had the locals been left in charge it would not have happened.
Changes made by man such as not dredging, etc, can be factored in. The global warming believers are never going to do that—they only recognize CO2 as a cause of all things bad in weather.
Many things, if left to locals, would not happen. Locals are much better at managing their space because they are the ones with the most to lose if they make a mistake. Getting the government out of these things would be very, very good.
Man made changes can be factored in but I suspect they are not. After all, it is unlikely they would want to show that failure to dredge channels would cause a big flood. Similarly where developments create a faster flow through or a restriction, which result in either upstream or downstream flooding, they wouldn’t want to declare that.
Along with the Somerset Levels is the Norfolk Broads as a major wetland area in the UK. Nothing happened on the Norfolk Broads. The watercourses of the Broads are managed by the local drainage board and keep the likes of the Environment Agency and RSPB out. I am not certain if the Levels got a local drainage board put in by the changes made by Owen Paterson to sort it out but I am sure that is what they wanted.
I think there is more to this than simple statistics. There is a town in Maryland called Point of Rocks that was flooded one winter in the early 1990’s when two cold nor’easters were followed by a warm one which released three storms of snow and rains at one into the watershed above the town causing two rivers to create a 500 year flood. In the fall, a hurricane’s remnants dumped massive amounts of rain into the same watershed creating a second 500 year flood in 6 months. This area of the country is subject to feast or famine weather patterns with respect to large storms spinning in or along the coast. Our 20 in snowfall average is three, 8″ years followed by a 50+” year. If this bi-modal precipitation distribution is used to make a Gaussian statistical prediction based on a single average, you would get the wrong expected frequency of exceptional years.
Yes, of course you get a wrong frequency of exceptional years in some cases. These are ESTIMATES. They are not necessarily predictive. Back to the coin toss. It tells you how many time to expect heads based on probability. Yet people can and do get heads 5 times in a row. It’s not about the ACTUAL outcome. It’s about the probability of an outcome. There will always be exceptions. I suppose the only way to avoid this is to simply say it might rain tomorrow, it might not. It could be hot, it might not. ALL weather forecasts are made using statistics. Shall we cancel weather forecasts altogether? Forget the whole thing because it’s just an estimate?
Note that the probabilities need to fit into known climate patterns such as the 11yr, 30yr and 600yr solar cycles. hence in the south of uk we get snow in the winter for 5yrs then no snow in winter for 5yrs – like clockwork. So flood event probabilities will be cyclical as well with solar rhythmns.
I’m getting the feeling from comments here that only 100% accuracy is accepted, which means we toss out every statistical process out there. I agree that the current language of a hundred year flood is not correct. It’s a 1% chance in any year. However, it seems one should correct the misconception of the statistic, not just throw out all statistics and probability. If we don’t use probability or stastics, we’re left with “a big hurricane in heading in some direction and might hit a continent”. That seems like a drastic move to avoid being wrong. However, maybe people would be more comfortable with that, rather than an educated guess at the direction of the progress of the storm. That was, the weather service can never, ever make a mistake.
I gain the impression that readers question the certainty to which these claims are made, ie, without making it clear that they are uncertain and not predictive, and people are questioning whether we really have sufficient data with which to compile the claim.
The coin toss whilst illustrative as to principle and whilst it well demonstrates that it is both rare to see a heads then tails repeated pattern, and it is not unusual to see a medium to long series of the same side coming up, only takes matters so far since there are essentially very limited outcomes.
When ( for practical purposes) there are only 2 variables, it is easy to make predictions. But where there are many variables (such as in weather) and where chaos plays a more significant role, it is far more difficult to construct a realistic statistical model.
This is made more difficult due to lack of historic data. Consider Australia, how do you make predictions of a 1 in 100 event when you only possess data going back 80 to 150 years? One would need upwards of 500 years worth of data before something meaningful could be produced.
As I see it, a lot of these predictions are guestimates. They are little.more than informative as to whether something is unusual or commonplace.
I don’t think that’s the point being made. Probability is fine when you have sufficient understanding of what can happen, but misleading or worse when you do not. And probability only works if the future resembles the past – which it may not. See Black Swans and Fooled by Randomness.
Statistics are constantly misused and abused in any number of fields. The number of areas where statistics are actually useful is pretty small. This stuff -100 year storms – seems to be an area where statistics creates more obfuscation and confusion than clarity. And the statistics themselves may well fail the two basic tests – do we understand sufficient possible occurrences and will the future resemble the past?
For climate, it is worse than that.
You will often here the meme that climate vs weather is like boiling water vs bubbles – one cannot predict individual bubbles, but one can predict general boiling behaviour from incresed temperature. This is not the whole truth, however – boiling behaviour is largely constrained by bother temperature AND pressure. If these thwo variables do not correlate as a non-zero integer, then temperature is to some extent not predictive of boiling behaviour – there will be some correlation, but it is by no means certain that an increase in temperature will result in an increase in boiling behaviour.
For climate then, we need to consider temperature, humidity and advection (wind) at the very least. These variables are not 100% correlated anywhere on the surface of this planet, and the correlation varies from place to place and over time.Basing predictions of any one of these variables based on a single other variable (partial CO2 pressure) then is highly dubious to sat the least.
Yet here we are…
I have no problem with statistics, I do have a problem with people who manipulate statistics as they say, ‘figures can’t lie but liars can figure’.
Statistics can be a useful tool but er like the model’s won’t tell yer and no matter how many algorerhymes are at best just that, a ‘best guess’, statistical analysis of past events can produce pointers but headlnes insofar as ‘wet’ are concerned, like “hundred year event” are just scaremongering bollocks – they know it, we know it.
And then what [statistical analysis] other than the alarmist claque – what, who does it serve?
Shit happens, Houston was unfortunately, appallingly flooded very badly. Yet, the Texans are probably [with Alaskans] some of the toughest, independent, self assured, confident of their own ability – set of people in the US.
It seems to me, that, they [Texans] patiently determined are just getting on with it, with none of the sort of outcry we witnessed a few years prior, further east in Louisiana and may we all respect, marvel at them – for that.
Try explaining, contextualizing historical statistical precipitation analysis to a Texan just now would likely get you some interesting answer, maybe an answer of the hot lead variety.
I agreed that calling this a 100 year event is deceptive. It should be a 1% chance. Most people realize that 1% events do occur but not often. Some understand that these events can happen two years in a row. The problem is the misuse of statistics. As far as I can see, the proper action is to educate people on how these mathematics work. Certainly not during a hurricane—only a clod (or a really nasty liberal) would do that. I can’t see any way to get rid of the stats, nor any reason to. Just use them properly.
agreed, totally.
I think this has been covered before.
I seem to recall the expert on BBC More or Less saying something like
“Everyone in the industry knows that one in one hundred year floods often happen more than once in every hundred years, the term is really a misnomer”
Maybe it was here http://www.bbc.co.uk/programmes/b06wbghr
Even the Wikipedia page says this
“A common misunderstanding exists that a 100-year flood is likely to occur only once in a 100-year period. In fact, there is approximately a 63.4% chance of one or more 100-year floods occurring in any 100-year period. ”
https://en.wikipedia.org/wiki/100-year_flood
David Spiegelhalter says the return period is “one of the worst possible ways to communicate risk”:
“A better approach is to use the percentage chance of a flood over the coming five or 10 years”
If a weather event is likely to happen on average every 100 years, it can’t be attributed to human effects on that basis. One argument left for alarmists in such cases is that it might have been made worse by human effects, but even that is highly debatable.
Low frequency of severe landfalling storms/hurricanes in recent years is opposite to alarmist predictions, and sea levels have been rising slowly but steadily ever since the end of the Little Ice Age.
Reblogged this on WeatherAction News and commented:
The moreb places that are measured and the longer duration of weather records, the more there are to be broken. Added to this you then get new records such as bank holidays which are ready to break due to the short duration of that record.
But if that doesn’t work, you can always decide to lower the previous record by a degree to claim the new one, eh Met Office?
Indeed, meaning there’s a ready supply of narrative