Why you can get ‘500 year floods’ two years in a row by Anne Jefferson:

Flood probabilities are based on historical records of stream discharge. Let’s use the Iowa River at Marengo, Iowa as an example. It reached a record discharge of 46,600 cubic feet per second* (1320 m3/s) on 12 June. That flow was estimated to have a 500 year recurrence interval, based on 51 years of peak flow records

…

When you are extrapolating beyond your data by an order of magnitude, the highest points in the dataset start to have a lot of leverage. Let’s imagine that there’s another big flood on the Iowa River next year and we do the same analysis. Now our dataset has 52 points, with the highest being the flood of 2008. When that point is included in the analysis, a discharge of 46,600 cubic feet per second* (1320 m3/s) has a recurrence interval of <150 years (>0.6%). It’s still a darn big flow, but it doesn’t sound quite so biblical anymore.

…

Urbanization and the adding of impervious surface is one cause of increasing flood peaks, but in Iowa, a more likely culprit is agricultural.

This post is a good explanation that the 500 year flood idea is just way of saying .2% probability (that some people confuse as meaning it can only happen every 500 years). But I actually am more interested in the other factor which is how much estimation is in “500 year prediction.” We don’t have 500 years of data. And the conditions today (I believe) are much more likely to create extreme conditions. So taking comfort in 500 year (.2%), or even 100 year (1% probability) flood “predictions” is dangerous.

It would seem to me, in fact, actually having a 500 year flood actually increases the odds for it happening again (because the data now includes that case which had not been included before). It doesn’t actually increase the likelihood of it happening but the predictions we make are based on the data we have (so given that it happens our previous 500 year prediction is questionable). With a coin toss we know the odds are 50%, getting 3 heads in a row doesn’t convince us that our prediction was bad. And therefore the previous record of heads or tails in the coin toss have no predictive value.

I can’t see why we would think that for floods. With the new data showing a flood, (it seems to me) most any model is likely to show an increased risk (and pretty substantial I would think) of it happening again in the next 100 years (especially in any area with substantial human construction – where conditions could well be very different than it was for our data that is 20, 40… years old). And if we are entering a period of more extreme weather then that will likely be a factor too…

The comments on the original blog post make some interesting points too – don’t miss those.

Related: Two 500-Year Floods Within 15 Years: What are the Odds? USGS – All Models Are Wrong But Some Are Useful by George Box – Cancer Deaths – Declining Trend? – Megaflood Created the English Channel – Seeing Patterns Where None Exists – Dangers of Forgetting the Proxy Nature of Data – Understanding Data