Unprecedented flooding caused by climate change? Not really.

Liz Truss, Secretary of State for Environment, Food and Rural Affairs, linked extreme weather with climate change (albeit implicitly):

The hon. Lady is absolutely right about the extreme weather patterns that we are seeing. As we say, that is consistent with climate change trends. Climate change is factored into all the modelling work that the Environmental [sic] Agency does, but in the light of this extreme weather we must look at that modelling and ensure that it is fit for purpose for future decisions. We constantly review investment in flood defences. It is important that we remain fair to people across the country, and that the people of Cumbria understand why decisions have been made and get the proper protection they deserve.

(From Hansard, viewable here.) As an aside, I take exception to her statement that "climate change is factored into all the modelling work that the Environmental [sic] Agency does" (see my blog here), but in this post I mostly want to look into the link between extreme flooding and climate change.

A team from Cambridge, Aberystwyth and Glasgow universities have looked at geological evidence for historic floods (article here) and come to the conclusion that current flooding isn't unprecedented. I thought it'd be interesting to try to do the same with recorded river flows - so I've looked into how often we'd expect to see an extreme flood (i.e. a flow bigger than we've ever seen before at the gauge in question) across the UK gauging station network.

The number of times a record breaking flow is observed is actually an interesting statistic to work with - because it's independent of how the flow is distributed statistically. Fitting distributions to short series is fraught with problems (see here), so not having to worry about this is a definite advantage. The probability of seeing a record breaking value is simply 1/N, where N is the length of the record¹ - if we assume that there are no trends in the data. This probability decreases as we collect more records - as we'd expect, because in a long record we've seen it all before, and the probability of a record being broken is therefore small. If there is an upward trend in the data, we'd expect records to be broken more often than this - so this could be a useful test for increasing flows coming, for example, from climate change.

I've calculated how many record high flows we'd expect to see in UK flow gauging stations assuming no climate change, along with how many we actually have seen over the past - shown in the figure below. Unfortunately it takes a while for new flow data to get incorporated into publicly available data sets, so this data doesn't include the most recent 2015-6 floods.

The expected number of maxima (grey line) peaks in the 1970s. This is purely due to the increase in the number of gauges (blue line) between approximately 1960 and 1980 - there were a lot of new gauges around in the 70s which caused a lot of record flows to be recorded. The observed incidence of record breaking high flows (orange line) show a lot of variation - with big peaks in 1960, 1968, 1974, 1979 and 2000, corresponding to big flood years. Interestingly, the peak corresponding to 2007 isn't that big - perhaps reflecting the significant role of surface water flooding in that year.

There's a lot of variation in the signal, and we expect some years to show more records than others. Is it possible to pick out a signal from climate change? I don't think so - the observed number of records follows the expected number pretty well, but with a lot of scatter. I've simulated what we might expect to see if there were a trend in flows (plot below), increasing by 2% per decade (this is broadly in line with guidance on current rates of climate change effects for river flows). The difference is much smaller than the natural variations in maxima we see from year to year.

So looking at the record breaking flows we see every year, I don't think we can assign these to climate change - it's just natural variability that leads to some record breaking years.


¹ One of the series has got the be the maximum - and the probability that it occurs at the end, i.e. we've just broken a record, is therefore 1/N. You can prove this with fancier maths too.

Climate Change - it's happening now

The winter 2015-16 floods in the UK have sparked the usual debate - we're not spending enough on flood protection, building on floodplains has made things worse, it's all because we're paving over driveways etc. Many commentators (including my old colleague Reza Ahmadian) have cited climate change as a contributory factor - but how true is this?

Lots of work on climate change seeks to look well into the future - some work I recently did for the Committee on Climate Change (report here) focused on changes by the 2050s and 2080s (although we did look at the 2020s too). But the recent news that we're already approaching 1°C above the pre industrial average prompts a question - should we therefore also see a change in flood risk? The link between global temperatures, rainfall and river flows is well established (in the minds of climate scientists at least), so in this post I'll try to understand what this means practically for flood risk.

It's important first to understand a key concept in climate studies - the baseline. While we're almost at 1°C above the pre industrial average for global mean temperature, that's only around 0.7°C above the temperatures in the period 1961-1990, which is typically used as the baseline in climate studies (like the IPCC). The chart below shows global mean temperature relative to this 1961-1990 baseline, showing that the baseline almost corresponds to a relatively stable period for global temperatures between 1940 and 1980, but that there was significant period of warming before that. I've added a thick red line to show a "cartoon" of running average temperature, which is constant before 1920 and from 1940 to 1980.

This global mean temperature history has two potentially significant effects for how we understand flood risk. Firstly, our records of flows from gauging stations might not be stationary - there might be trends in there which we need to correct for before doing any statistical analysis. Secondly, there has been an increase in temperature since the 1961-1990 baseline which could have caused an increase in flood flows.

What does this mean practically? I've done some calculations for 5 gauging stations across the UK, using the latest advice on climate change impacts on river flows, and an assumption that increases in flows are roughly proportional to global temperature increase. I've expressed the results as the 1 in 100 Annual Exceedance Probability (AEP) flow for the baseline period 1961-1990, and the annual exceedance probability of that flow now in 2016.

The results are shown in the table below - for the Eden at Sheepmount in Carlisle, the baseline 1 in 100 flow of 1405m³/s now has an AEP of 1 in 62 - an increase in frequency of ~70%. The Blackwater in Northern Ireland shows a doubling of frequency for a big flood. The effect is less pronounced for the Tay, and the Severn and the Wye show no increases (because the climate change impact guidance indicates no sensitivity to climate in the short term up to 2025). 

So for some catchments, we're seeing significant increases in the likelihood of large floods compared to the baseline period 1961-1990. We need to think about these effects when we undertake flood risk analysis - rather than assuming our records of flow are representative of conditions now, and that climate change is something that only happens in the future.