How severe is Cape Town's “Day Zero” drought?
Abstract
With reservoirs running dry, Capetonians are bracing themselves for the day when their water supply may be switched off. Piotr Wolski delves into the rainfall record to put this drought into historical context
During the last months of 2017, Cape Town in South Africa was expected to run out of water. Reservoirs were dwindling, and “Day Zero” – the day when the city's 3.7 million metro-area residents would have their taps switched off – was a very real prospect. Strict rations would be in force from that point on, and long queues would be expected to form at water collection points throughout the city.
The only certain way to avoid this was to cut water usage by more than half, to 450 million litres per day or less, as then supplies would last until the rainy months of July to October 2018. So far, thanks to the efforts of Capetonians, Day Zero has been delayed. Household water consumption has been reduced, and agricultural use has also declined. But at the time of writing, daily consumption was still hovering around 520 million litres, and taps may still have to be switched off at the beginning of July should the rains fail.
Cape Town's water crisis is, understandably, a hotly contested issue. No one disputes that the city is experiencing a drought, which began in 2015. But some argue that the drought is a mild one, and that the root cause of the water crisis is poor planning and mismanagement on the part of the Western Cape Water Supply System (WCWSS). Others argue that the drought is severe, and while human factors may have aggravated the situation, Cape Town has simply been unlucky in the climate lottery.
In these disputes, facts are few and opinions plenty. But as a hydrologist, I can apply some statistical analysis to answer the question, “How severe is the current drought?”
The 36-year view
There are a number of weather stations that measure rainfall in the vicinity of WCWSS reservoirs (known as “dams” in South African English), and data for them are available from the Department of Water and Sanitation (DWS) website (dwaf.gov.za/hydrology). Unfortunately, not all stations in that data set have good rainfall records – there are numerous gaps, and the availability of the record varies over the period 1981–2017. In fact, there are only four stations located near WCWSS dams that have no significant gaps and no systematic errors.
Figure 1 shows total annual rainfall recorded at these stations since 1981. Note that the years run from November to October. This is because October is more or less the end of the rainy season and is the last month when an increase in dam levels may be recorded.
Total annual rainfall recorded at four Department of Water and Sanitation stations – Vogel Vallij, Zacharashoek, Theewaterskloof and Kogelbaai – for the period 1981–2017. Years end in October.
From Figure 1 it is clear that 2017 rainfall was low, perhaps lower than it has been since 2000. However, the current perilously low water levels in WCWSS dams are not only a result of low rainfall in 2017; previous years were also relatively dry. This can be seen in Figure 2. To produce this figure, I calculated average rainfall over 2–4 years for each station and then produced a simple average of the data from the individual stations. That average may not express the true rainfall pattern that affects the dams. For that, one would need a much denser rainfall network. But at least it does take care of possible inaccuracies in the data of individual stations.
Total annual rainfall averaged over four Department of Water and Sanitation stations, with a running mean over 2, 3 and 4 years.
The plots of average rainfall tell a consistent story: the amounts for 2017, as well as the preceding 2-, 3- and 4-year averages, seem to be the lowest since 1981. But 36 years of data provide a fairly limited view of the historical rainfall record. We know that there were droughts before 1981 – notably in the 1920s and 1970s. To compare the severity of the current drought to those older ones, we need data stretching back further in time.
The 84-year view
To put the 2017 drought in a more extensive context, we turn to a different data source: the South African Weather Service (SAWS). Rainfall data for some SAWS stations in the Western Cape go as far back as the late 1800s, but unfortunately there is no overlap between the DWS and SAWS rain stations, so we cannot simply extend the record of the four stations used earlier. We need to repeat the analyses on the SAWS data set.
The most recent drought, manifested through low rainfall in 2015–2017, is very rare and severe indeed
I started with a total of 96 stations in the Western Cape region, but of that set only 13 had data going back to the 1930s – and of those 13 stations, only three were suitable for this task. The rest were located too far away from the WCWSS dams, did not have consistent data covering the last 3 years, or had major gaps in their records.
So, for these three stations I again calculated average rainfall over 2–4 years and then produced a simple average of the data from the individual stations. The results are in Figure 3. This shows that 2017 was the lowest rainfall year since 1933. Average rainfall for the 3-year period 2015–2017 was also the lowest in the record, but this was not the case for the 2- and 4-year means leading to 2017.
Total annual rainfall averaged over three South African Weather Service stations, with a running mean over 2, 3 and 4 years.
Rare and severe?
When discussing events such as droughts and floods, water and climate scientists often quote a “return period”, or “recurrence interval”, as an indication of the rarity (or not) of a particular event. The SAWS data from Figure 3 would suggest that 2017 and the period 2015–2017 were the driest since 1933. This translates simply into a drought return period of once in 84 years. But the 2017 drought might be seen to be rarer still were we able to look back further in time.
It is possible to assess the return period based on a statistical distribution of rainfall values rather than on the actual data. The rationale for this is that observational data (i.e. rainfall) are in reality only a relatively short-term sample of all possible rainfall events that can occur at a location under a given climate. The fitted distribution function, rather than the sample itself, describes that climate. Having that distribution function allows for extrapolation beyond the length of the observational data series. It also allows for determination of the uncertainty of the estimate of the return period.
This analysis is illustrated in Figure 4. There is no clear theoretical basis for what type of distribution the data on 3-year rainfall should follow. Obvious candidates are truncated Gaussian and gamma distributions, and members of the generalised extreme value distributions family. I have tested these types, and the distribution function that fits the observations best is the gamma distribution, shown in Figure 4(a). Although it does not describe well the higher frequencies in the observed range, the estimates of the return interval for rare events based on that distribution fit those based on data very well. This is illustrated by the agreement of the solid line and the data-based dots in Figure 4(b). In that plot, observed data (markers) are plotted in a way that takes into account the uncertainty of observations and the derived return interval. So the return interval for 2017 does not correspond to what one could simply derive from the data (i.e. 84 years).
(a) Mean annual rainfall (3-year moving averages), 1933–2017, with fitted gamma distribution, and (b) 1/probability of exceedance (= return period). Exceedance refers to less than the observed mean annual rainfall amount.
The 90% confidence interval in Figure 4(b) is obtained by bootstrapping. I have used the so-called block bootstrapping procedure so that the confidence interval takes into account the autocorrelation in the data (for there is autocorrelation in annual rainfall totals, and in the 3-year running mean which is used to determine the return interval of the drought).
So what comes out of this analysis? My best estimate of the return interval of the meteorological drought in the region of WCWSS dams is 311 years, with 90% confidence that it actually falls between 105 and 1280 years. In other words, the most recent drought, manifested through low rainfall in 2015–2017, is very rare and severe indeed.
Capetonians will no doubt continue to debate whether mismanagement played a part in this crisis. But my analysis suggests the authorities were faced with a situation that would have been hard to cope with regardless.
Note
This article is based on Piotr Wolski's CSAG blog post, bit.ly/2EuL7aS.
