During a conversation last week, a fellow CSAG-er casually mentioned that the probability of precipitation depended on the day of the week. I found this to be puzzling – after all, what does the weather know of the days of the week?
A search of the literature uncovered that this theory is plausible, and that the variability of rainfall occurrence with the (human-introduced concept of) day of the week is human-induced. Numerous studies have linked this phenomenon to pollution, which is in turn influenced by the concept of a “working week”. Roughly speaking, it is suggested that pollution-derived aerosols in the atmosphere change the microphysics of the cloud, not unlike the concept of cloud seeding, thereby increasing the chance of rainfall.
While the literature seemed conclusive regarding the fact the pollution-derived aerosols can affect the weekly cycle of rainfall, it was less so on the direction of the change. A study of near-coastal North West Atlantic ocean areas of the US found that they “receive significantly more precipitation at weekends than on weekdays” [Cerveny and Balling, 1998], due to the build-up of pollutants during the regular working week. However, a later study by Rosenfeld  found “plumes of reduced cloud particle size and suppressed precipitation originating from major urban areas and from industrial facilities such as power plants.[…] both cloud droplet coalescence and ice precipitation formation are inhibited in polluted clouds”. So which is it, and does it happen in South Africa?
Armed with precipitation data from 4000 weather station in Southern Africa, I set out to test this theory. For each station, I tallied the day of the week on which rainfall (defined as precipitation > 0.2 mm) occurred over the time period 1970-2012. At first glance, the distribution of rainfall occurrence by day of week appeared to be uniform (Figure 1).
Myth busted… or so it seemed. Undeterred, I decided to look at the variance in day of week occurrence for each station. This yielded an unexpected result: those stations with high variance all followed the same trend, with a reduced occurrence of rainfall on Friday and Saturday, followed by an increase on Sunday (Figure 2).
While it should not be unexpected that some stations would have a higher variance than others, the fact they all trend in the same direction is surprising. If the errors in measurement were normally distributed, we would expect these to be distributed amongst different days of the week, as well as an equal number of higher and lower rainfall occurrences amongst stations with high variance. A further piece of the puzzle was the fact that the distribution of variances was monotonic (rather than bimodal), suggesting a smooth transition from Figure 2 to Figure 1, and not simply a batch of bad station data.
So what do the stations with higher variance have in common? Given the original hypothesis of pollution-influenced cloud dynamics, I looked at the mean total CO2 emissions from 1970-2008 [EC-JRC/PBL. EDGAR v4.2] (Figure 3). Every station with a high variance in day of week occurrence was located near major population centres and other hotspots of CO2 emissions, with the only anomaly being the CO2 patch over Mozambique, due to a lack of station coverage. In contrast, those stations in the lower emission regions, such as the Northern Cape and Namibia, were closer to uniform distributions of rainfall occurrence by day of week. As these stations make up a significant part of the observation record, it should not be surprising that the changes in rainfall occurrence were not immediately obvious.
Do you have plans this Sunday? If so, bring an umbrella, just in case.