Recent heavy rain in Dubai has prompted one of those absurd arguments that the internet is great for. This was the heaviest rain on record for Dubai and led to severe flooding. The UAE has been undertaking regular cloud seeding since 2010 and therefore some believe that the recent flooding is caused by or exacerbated by cloud seeding.
I don’t know a great deal about cloud seeding. I’m vaguely aware that people put various kinds of small particles into clouds in the hope that they will rain more, disperse, or hail less depending on their requirements. Beyond that, I have a vague impression that it’s not terribly effective or that its efficacy is not easy to demonstrate. For a very basic primer see the Climate Brink.
It would have passed me by, but then Judith Curry posted a link to Grazia Magazine with the headline “What caused the UAE storms and torrential rain? Yes, it was cloud seeding” with the comment “Torrential rains in Dubai – unintended consequence of cloud seeding?” That would have passed me by too, though I may have managed a wry smile at that little question mark1. The headline subsequently changed and now the whole page has disappeared. Nevertheless, the response prompted Judith Curry to follow up with:
I would have ignored that too, and another bizarre tweet likening this to COVID lab leaks2, but it was getting a lot of traction – Matt Ridley retweeted the paper – and the whole thing was becoming a thing3. At the same time, here was an actual study – not a Grazia article, an actual study – to get my teeth into and one that was recommended by a bona fide expert on clouds no less.
The study in question is “Study of Impact of Cloud-Seeding on Intensity-Duration-Frequency (IDF) Curves of Sharjah City, the United Arab Emirates” published in the MDPI journal “Water”. It’s always interesting with MDPI articles to see how quickly they were published after submission. In this case, it took a little over three weeks. MDPI is famed for its rapid turnaround on articles and their peer review is second to none4.
Anyway. I had a quick read.
Regular cloud seeding started in the area in 2010, so they compare data before and after 2010, ascribing all changes to cloud seeding without any kind of control.
They claim to use 3 stations, comparing data from 1992-2009 with data from 2010-2020 for each. However, the three station records start in 2003, 2010, and 2014. They fill the very large gaps in all 3 using data back to 1992 using a single series from Dubai airport (which they don’t actually analyse). They do this data-ekeing using “self organizing maps” but they might as well use magic. I don’t believe for a second that they can reconstruct meaningful realistic variability for three different stations from a single station across a period of, in one case, 22 years. They don’t caveat this data in any way, they don’t validate it and they don’t show that it behaves in a realistic way.
They look at the annual maximum daily rainfall for each year and find that its higher on average after 2010 than before. They also find that the return periods (up to 100 years based on 10 years of data for the post 2010 period) for rainfall totals over different accumulation periods are shorter after 2010 than before. i.e. heavier rain is more common after 2010. They claim that this proves “that cloud-seeding operations had led to an obvious increase in IDF values, i.e., the intensities of rainfall storms.“
No attempt is made to put any kind of uncertainty range on the estimated data nor on the return periods estimated from the estimated data. No attempt is made to assess whether there is a meaningful difference between the two periods, or explore alternative causes.
The situation is this. After 2010, they have data from three different stations. Before 2010 they mostly have data from a fourth station, but manipulated in three different ways (one presumes). There is some difference in the calculated summary statistics between the two periods, but no attempt to understand what caused it or whether it might be an artefact of sampling or the data-ekeing process. The analysis doesn’t support – doesn’t even try to support – the contention that cloud seeding caused an increase in rainfall averaged over a ten-year period, let alone that of a single storm. Nor does it try and rule out any alternatives. Given the limitations here regarding the data, I can’t see any way it could be fixed: if I were reviewing the paper, I’d be inclined to recommend a redo at best. One could make it all about Dubai airport, add some tests to show that differences before and after are “significant” but it still doesn’t demonstrate that cloud seeding is the cause, for that you’d need a control or some kind of randomisation. They have neither.
I looked elsewhere for supporting evidence, but it wasn’t much help. There are statements in the introduction which seem to suggest such an attribution had already been made. For example “The first exploratory programs started in the 1990s and altered the quantity of precipitation released from clouds (Malik et al.)“
Malik et al. is a short paper on the general principles of cloud seeding and has nothing to say about the UAE, Sharjah City, or Dubai. It’s mostly about India if it’s about anywhere.
There’s this paragraph too. I quote:
“As noted above, in 2010, the UAE began the cloud-seeding project officially, which registered success in developing rainstorms in the deserts of Dubai and Abu Dhabi (Kumar and Suzuki). By 2016, the UAE reinforced its missions on cloud-seeding with approximately 100 flights per year carrying ionic compounds (such as sodium chloride) to seed the clouds. In 2018, the National Centre for Meteorology in the UAE revealed that the country executed 184 cloud-seeding operations as a consequence of which there was a 55 mm increase in annual rainfall the following year (Ćurić et al.).”
The reference to Ćurić et al., claims it showed a 55mm increase in annual rainfall. It’s a modelling study looking at the theoretical impact of two cloud seeding compounds. It doesn’t show anything about actual rainfall in the real world.
The Kumar Suzuki paper at the start of the paragraph looks at the properties of clouds to assess which might be suitable for cloud seeding and does not show the success or otherwise of the cloud seeding itself.
Both these studies might be considered important precursors to a cloud seeding field campaign – indeed this kind of pre-analysis is recommended by a WMO report on weather modification – but neither one does what is claimed here. The WMO report is worth a read. It goes into a bit more detail about what is required to demonstrate the cloud seeding is effective at a catchment scale (amongst other things).
A more detailed study is cited – Al Hosari et al. – but it uses Dubai Airport and Sharjah Airport as controls (which implies they should be unseeded throughout; they’re not) with the target stations further inland towards the Hajar Mountains, and finds that rainfall declined from 1987 to 2019 at the Sharjah Airport site with no obvious increase after 2010 (contrary to what was argued in the current paper). Cloud seeding for the target locations started at a different date though – 2003 to 2019 – so how that interacts with general cloud seeding starting 2010, which would affect the control stations too, isn’t clear. In this paper they also have an observational break that coincides (more or less) with the start of cloud seeding with obs coming from one source before and another after. A trend changepoint analysis suggests a break in 2011 at the target sites (in the July-October period this time rather than annual means), which corresponds to the start of general cloud seeding(ish) but a break in the control stations (of which Sharjah Airport is one) is only found in 2017, long after the start of operational cloud seeding. This could just be because the timeseries is mostly zeroes up until that point5.
In short, there’s nothing in the paper, not even indirectly via citation, that supports the contention that cloud seeding has any effect on actual rainfall in Dubai, and certainly not on this particular storm. Nor does it demonstrate the opposite. It is therefore almost perfectly useless. Evidence may exist, but it’s not to be found here.
- Such question marks work defensively in the same way that someone who has said something otherwise indefensible might complain woundedly if called out “What? I’m just joking“. ↩︎
- Like the question mark that’s there for a specific audience to tell them what to think. ↩︎
- I find myself drawn to things. This reflects very badly on me and has wasted a lot of my time. On the other hand, I’ve picked up a wide range of almost useful knowledge in the process. ↩︎
- In a very literal sense, if not the most common one. ↩︎
- There’s also an analysis of radar data for two years with storms split into seeded and non-seeded with seemingly big differences in some metrics between the two. However, there’s no indication of how storms were chosen for seeding so one can’t rule out the possibility that they only seeded the storms that looked like they might go large. It’s not random. In fact, it’s all a bit of a mess. This is all very frustrating. ↩︎
Diagram Monkey 
Leave a comment