A supercool opportunity

Hi there,

Hope you’re enjoying Climate Week so far if you’re in New York, and a more restful week than those of us who are if you’re not!

Today, we’re ~targeting~ a trajectory back towards Keep Cool’s roots in more tech-focused coverage, as opposed to me opining about more general things going on. If I hear collective sighs of relief, trust that I won’t take them personally.

DEEP DIVE

I don’t want much American football anymore these days, neither NFL nor college. But as I let my mental gears turn over about how I wanted to structure this piece on mile ~500 of my roadtrip across the country, the fact that there is a 15-yard penalty in American football that can be levied on a team for “targeting,” i.e., for a particularly vicious tackle, or, more specifically, defined by the NCAA as: “…any hit that ‘goes beyond making a legal tackle or a legal block or playing the ball…’” came to mind.

As a fan of precise English usage, or, shall we say, finding better words to describe something, that definition itself leaves something to be desired. Which is an indication of where I’m actually going with this; perhaps my mind resurfaced the fact that “targeting” incurs a costly penalty in football because it runs counter to the idea that in most other modern arenas, including in most other cases in football itself, accurate and precise targeting tends to be a good thing. Of course, the fact that more accurate and precise targeting tends to be a good thing doesn’t mean all players, whether quarterbacks in football or climate tech startups working in reasonably novel fields, have sufficient, let alone optimal, targeting capacity.

It’s OK to start off target

The initial absence of requisite targeting capacity and technology in various industries doesn’t stop people, companies, or researchers from getting on with their work. Nor should it. For instance, the computer on which I’m typing this and the semiconductors that facilitate its computation are a product of an industry that didn’t let suboptimal targeting capacity get in its way in its earlier days. In the 1960s, early semiconductor manufacturers couldn't precisely control where dopant atoms— foreign atoms intentionally added in small quantities to a base material to alter its fundamental electrical or optical properties—actually landed on silicon wafers during manufacturing (that is, before ion implanters came along). Said differently, they lacked the capacity to target dopant atom placement. They also lacked effective methods for inspecting defects in real-time. Hence, compared to today, yields were abysmal. 

Glossing over a massive amount of innovation and history, thanks to ion beam technologies, which offer atomic-level precision in placing dopants exactly where needed, and automated optical inspection systems, which provide immediate feedback on what is actually happening at the nanoscale, yields have risen from 10% to over 90% as of today in terms of reliability in ion implantation. That step-change improvement is no small part of what’s made Moore’s Law hold for decades. The graphic below charts some of the history of improvements for those interested. TL;DR: Targeting technology for semiconductor manufacturing has improved significantly.

If the above analogy didn’t track super well for you, consider mRNA vaccines as another case study in step changes in targeting technologies. mRNA vaccines were a tantalizing technological fantasy for a long time before the right targeting technology emerged to solve a pernicious problem, namely, how to get fragile RNA molecules, which include the requisite disease-fighting genetic instructions, past the body's defenses and into the right cells? Lipid nanoparticles, i.e., tiny fat bubbles that can protect mRNA and deliver it precisely, ended up being a key solution. Solving that targeting (and delivery problem) has probably saved tens or hundreds of millions of lives worldwide already.

You already live in a cloud-seeded world

The header above is not intended to trigger some descent into tin-foil-hat-wearing conspiracies. Recent events—such as the tragic floods in Texas earlier this summer, unfortunately overlapped with cloud seeding trials in the area (or at least, trials conducted by several preceded the floods by a few days), thus unleashing a separate storm of online vitriol and misinformation suggesting the (actual) storms were a product of geoengineering-–have already brought cloud seeding into the discourse, albeit largely in unproductive fashion. That said, one thing I’d frame as a net positive from the wide range of topics discussed by social media conspiracy theorists is the recognition that cloud seeding is already happening. 

Because it is. No conspiracies needed. Here are specifics to further frame the full picture:

China's comprehensive weather control department and efforts are the world's largest, employing 30,000+ people. A quick Google search can dredge up, for instance, this research paper based on “27,000 cloud seeding operations conducted since 2014” in China alone. And if that’s the publicly available data out of China, it’s possible you could append a zero or two to the figure for data that’s private. Similarly, the UAE has conducted hundreds of seeding missions annually for years, with significant investments in cloud seeding from the National Center of Meteorology and the Gulf Cooperation Council (GCC).  

Even countries you wouldn’t expect, per se, are in on the effort, as well as on attempts to prove cloud seeding can do more than just make more rain. Romania, Bulgaria, and Greece are all exploring and investing in hail suppression programs to protect vineyards. In Romania, I have it on good authority that there are teams of meteorologists who work in tripartite shifts to offer 24/7, around-the-clock coverage, staring at radar screens in a scene I can only imagine feels eerily reminiscent of Cold War thermonuclear warfare deterrence. What they're looking for is the perfect moment to launch rockets into clouds to deliver silver iodide, which acts as an ice-nucleating agent, thereby facilitating the growth of ice crystals in supercooled clouds. 

Of course, don’t just take it from me. As reported by the BBC:

And, last but not least, in the U.S., plenty of people are working on cloud seeding too. According to the U.S. Government Accountability Office, as of 2024, nine states had active cloud seeding programs: California, Nevada, Idaho, Utah, Wyoming, Colorado, New Mexico, Texas, and North Dakota. Insurance companies in Canada fund hail suppression programs to protect crops from damage. In Texas, cloud seeding operations span millions of acres, with the Panhandle Groundwater Conservation District operating programs that cost about 5 cents per acre. Even utilities are in the game—PG&E and Idaho Power have cloud seeding departments to fill reservoirs for hydroelectric generation.

Perhaps the most public-facing player in the Western Hemisphere in cloud seeding is Rainmaker, the CEO of which has been on our podcast (a good cloud seeding primer if you’re jonesing for it) and which—I imagine but can’t confirm—is actively cloud seeding in the Western U.S. this week and most weeks where and when conditions are good. A lot of what’s being tested and developed domestically and internationally is the hardware to actually deliver silver iodide into clouds, as well as equipment to monitor not just meteorological conditions to determine when, as the Romanians track, the weather is primed for experiments or full-scale cloud seeding attempts, but also to try and determine whether more rain actually precipitated than otherwise (itself a big challenge).

Cloud seeding will have its own 10x better targeting moment

Picture again those Romanian researchers and meteorologists gathered in a room. When it comes to precise and optimized targeting, I suspect that by 2050, we’ll look back on cloud seeding in 2025 as a relatively analog practice. Which isn’t an indictment; again, consider what twenty-five years of additional investment and development of targeting technology did for the semiconductor industry or mRNA vaccines. Nor am I suggesting that no one is thinking about this—if I’m arriving at this conclusion and I spend maybe an hour a week thinking about cloud seeding, the people working 996 on cloud seeding have certainly thought of it. I’m just here to say that better targeting, i.e., identification of where and when to deliver silver iodide into supercooled clouds, will probably end up being as important as the delivery hardware.

When cloud seeding works—when silver iodide or other seeding agents hit supercooled liquid water in clouds—it can boost rainfall by 15-20%. Nor is that speculation on my part. This stuff has been validated plenty. However, replication and especially attribution in complex real-world systems, such as constantly changing weather and ecosystems, are hard. Attribution aside, better targeting will help produce more reliable results. I feel like I should say something else to justify that conclusion, but it simultaneously strikes me as self-explanatory and quasi-tautological. If operators don't know which clouds to target with a payload of silver iodide, let alone where within certain clouds to do so or how much seeding material to use, success and efficiency rates will be artificially capped from what they could otherwise be.

Returning to the concept of cloud seeding, the critical factor, at best, is supercooled liquid water (SLW). Cloud seeding works by converting these suspended water droplets into ice crystals. What makes an ideal target is a) a cloud featuring a high amount of supercooled water and b) a cloud in which that supercooled water is still liquid. Clouds that feature a lot of supercooled water as ice, however, aren’t ideal. Not that you can divine any of that from the ground. 

At least not yet. Based on interviews with people working on this problem, I understand that much of the fledgling industry relies on numerical physics-based models. One challenge is that these physical models, which attempt to represent and simulate real-world phenomena tangibly (as opposed to, say, diffusion models), are being built in the absence of a sufficiently robust understanding of cloud physics. Hence, even with increasing computational power, traditional modeling efforts feature significant uncertainty. This targeting limitation is also why most programs focus on rain enhancement, as opposed to other potential applications of cloud seeding, such as hail suppression, all of which, alongside other goals like flood prevention and maybe even typhoon diversion, are technically feasible. But without knowing how to target appropriately, these remain laboratory possibilities rather than operational realities.

I wouldn’t go so far as to say the cloud seeding industry operates on a “spray and pray model,” as it’s quite sophisticated in many ways. But 10x better targeting will come, as it did for ion implantation in semiconductor manufacturing.

As an addendum, verification and attribution for cloud seeding will also require similar step-change-type improvements. For this to be a revenue-generating industry long-term, people paying the money for seeding will want good proof that it’s, you know, responsible for driving increased yields. Rain, after all, will fall at some point, often quite precipitously in a warming world. And if you fire enough rockets into clouds, you might convince yourself you’re doing the thing even if you’re not doing it well or at all. Today, most cloud seeding programs rely on longitudinal studies that compare seeded regions to control regions over multiple years. But on any given day? Shrug. That’ll have to change.

In addition to verification being the linchpin to the longer-term business case for cloud seeding, wherein buyers of cloud-seeded rain will invariably demand a reasonable degree of confidence that their dollars are actually driving additional precipitation, the boon of solid verification would also be significantly better data to plug into models being used to develop a more robust targeting and accurate and precise delivery architecture. Once additional rain can be reliably attributed to specific seeding operations and mapped back to increasingly targeted delivery within clouds, the loop to iteratively improve models based on real-world and correlated feedback should shorten and quicken considerably.

High chance of rain (and silver iodide)

When considering challenges like this, I always like to evaluate, at least at a high level, what confluence of technologies has evolved sufficiently in recent years to make previously pernicious problems more approachable today. Here are three I’d identify in this scenario: 

1. Machine learning + AI: Modern ML models excel at precisely the pattern recognition needed to identify supercooled liquid water conditions. Instead of relying on years of human training to develop intuition—as we’d say in German, Fingerspitzengefühl—the advanced “AI” that’s often referenced in a hand-wavy way elsewhere is very real and valuable in this case, insofar as it can analyze huge datasets of atmospheric conditions to uncover subtle signatures that may indicate or correlate with optimal seeding opportunities.

2. Higher-resolution sensing: New instrumentation—from dual-polarization ceilometers that distinguish ice from water to drone-mounted sensors that take measurements inside clouds—can ideally provide more granular data to feed into #1 above. The ideal would be some type of miniaturized sensors that can reliably measure supercooled liquid water directly within clouds, collecting training data needed to build models that infer 3D distributions of SLW from real-time radar. This type of hardware should be close to ready; a lot of it exists, but hasn’t been systematically deployed for weather modification.

3. Computational weather modeling: High-resolution numerical weather models can already simulate cloud physics at scales relevant to cloud seeding. Combined with real-time observational data, these models could provide a more accurate estimate of the additional precipitation to expect based on what’s actually executed in the field.

OK, that’s almost it. This is where I put the pen down and say, “Someone go build this!”

As to the Overton Window challenge, i.e., what the general public thinks about cloud seeding and other weather modification efforts, I won’t pay it too much attention here. There was a House hearing on geoengineering last week, with Rep. Marjorie Taylor Greene vocally leading efforts to stimy research and deployment of geoengineering efforts (which would be a patently ridiculous thing to do, considering the most active geoengineering stems from the millions of tons of greenhouse gas emissions pumped into the atmosphere every day) and restrictive bills are advancing in state legislatures as well, including ones specific to cloud seeding. 

Whatever the logic or absence thereof, the policy environment belies that public perception of weather modification efforts ranges from unaware to deeply distrustful, to put it mildly. Frankly, that’s simply not my concern here, though. Considering the extent to which some of this piece charted how active the global cloud seeding environment is, press on, I say. If the U.S. makes another unforced policy area, startups and researchers will simply end up moving elsewhere.

Finally, it’s worth noting that cloud seeding doesn’t really meet the definitional criteria for classifying as geoengineering as is. The discrepancies are temporal, for one: geoengineering describes human-induced activities that could alter the Earth’s climate over centuries and millennia. Cloud seeding doesn’t fit that criterion. Plus, you can make arguments for discrepancies in terms of the physical area of influence as well. Geonengineering calls to mind efforts that are by nature global, or at least potentially quite widespread, in their impact. Cloud seeding isn’t that. Which hits home why developing the requisite targeting and verification technologies for it is so vital. 

Hasta la vista,

— Nick