The Arithmetic of Rainmaking

AFew technologies have been promised so often, and demonstrated so seldom, as the deliberate modification of weather. The principle has been understood since the winter of 1946, when laboratory work showed that a cloud held below freezing but containing no ice — a supercooled cloud — could be persuaded to release its water if it were given something for ice to form on. Two techniques descend from that discovery. Glaciogenic seeding introduces silver iodide, whose crystal lattice resembles that of ice, into cold clouds; hygroscopic seeding disperses salt particles into warm ones so that droplets collide and grow heavy enough to fall. Neither creates a cloud. Both merely redirect water that is already suspended overhead, and this modest ambition is routinely lost in public discussion of the subject. By 2020 the World Meteorological Organization was listing operational programmes in more than fifty countries, most of them aimed at snowpack in mountain catchments, at rainfall over farmland, or at the suppression of hail.

BThe difficulty is not the chemistry but the arithmetic of proof. To show that seeding worked on a given day, a scientist must state how much precipitation would have fallen had nothing been done, and that quantity can never be observed. Natural variability in mountain precipitation is large; two storms of similar temperature and moisture may differ in yield by a factor of three. The standard remedy is randomisation — a computer decides, storm by storm, whether the generators are switched on — but the noise is such that a trial must run for many winters before an effect of the size expected can be separated from chance. Several programmes begun in the 1970s were abandoned before they reached that length, and their inconclusive results were widely reported as failures rather than as trials of insufficient duration.

CMarit Dahlberg, an atmospheric physicist at the Kirkwall Institute, has spent much of her career trying to run a trial long enough to settle the matter. Between 2012 and 2019 her group randomised 431 eligible storms over the Tarrow Range, seeding 218 of them from ground generators, and measured the outcome with precipitation gauges and with a radar that tracked the seeded material itself. The treated storms yielded 8 per cent more precipitation than the untreated ones, a difference her team put at odds of roughly forty to one against chance. Dahlberg is nonetheless exact about what the figure covers. A storm counted as eligible only if its cloud tops lay between minus eight and minus twenty-two degrees Celsius and carried measurable supercooled liquid water; outside that window the generators stayed off, because the physics predicts nothing there. "The number applies to the storms we chose to treat," she has written, "and to no others."

DEmeka Osundu, a hydrologist at the Marlow Water Policy Unit, does not dispute the 8 per cent. His objection begins where her caution ends. In the Tarrow catchment, storms meeting her temperature and liquid-water criteria delivered only about a third of the winter's total precipitation, so a gain of 8 per cent on that third comes to under 3 per cent across the season — smaller than the swing the same catchment shows from one year to the next. Worse, he argues, what fills a reservoir is not snowfall but runoff, and in a warming climate a rising share of new snow is lost to sublimation and to soils left dry by autumn. Osundu's conclusion is not that seeding fails. It is that a technique verified at the scale of a storm is now being sold at the scale of a drought.

EThat gap between what is demonstrated and what is promised has consequences beyond the laboratory. Districts lying downwind of seeding operations have repeatedly claimed that rain owed to them was being intercepted, and in 2018 a dispute of this kind reached a federal court. The physics offers little support for the charge, since a seeded cloud releases only a small fraction of the water it carries, and observational studies have found no shortfall downwind of long-running programmes. Yet Dahlberg herself notes that the absence of a detected shortfall is not the same as proof of no effect, because the signal being looked for is smaller than the measurement error of the networks available. Public confidence, meanwhile, is shaped less by such distinctions than by timing: programmes are commissioned in dry years, when rain is least likely to arrive, and are then judged by whether it does.

FTwo further questions are raised often and settled rarely. The first is chemical. Silver is toxic to aquatic organisms at high concentrations, and monitoring in seeded basins has found silver in snow at a few nanograms per litre, far below drinking-water thresholds; but the sampling has concentrated on watercourses rather than on soils, where accumulation would be slower and harder to reverse. The second is economic. Hail suppression, which aims to turn damaging stones into harmless ones by multiplying the particles that compete for the same water, has been funded by insurers in Alberta since 1996 on the strength of claims data rather than of randomised trials, and the mechanism remains unproven even where the savings are counted in real money. Both arguments turn on the same asymmetry. Seeding is cheap, and the cost of continuing a programme that does nothing is small; the cost of establishing that it does something is not.