A study published in 2025 in Nature has stayed with me since I read it. Aaron Hultgren and his colleagues built a model covering six staple crops across more than twelve thousand regions of the world. They used decades of yield data to measure how producers have adjusted to their conditions over time. For rice, all those adjustments together cut projected end-of-century losses by 79 percent. For wheat, the reduction was statistically zero.
I want to be careful with that. It doesn't mean wheat farmers haven't been adapting. They have. It means that when you add up everything wheat producers have done historically, including new varieties, different fertilizer rates, shifted planting dates, and irrigation where they have it, the combined effect doesn't show up in this model as meaningful protection against what's coming.
The study used what's called an econometric approach. It looked at how yields responded across a lot of places and years, and from that it estimated the net effect of whatever people did in response to climate, without watching each thing they did. So it didn't test particular practices and find that specific wheat adjustments fail. It measured outcomes. The authors' own explanation for the gap between the two crops is that wheat producers in the historical data seem to take on more weather risk as their incomes go up. They don't put it down to wheat biology, or to rice having some built-in advantage. The model can't show what's happening agronomically.
Still, the outcome is the outcome. And there's field-level work from one of the country's main wheat regions that makes it a lot more specific.
The work itself
Eastern Washington grows wheat across a rainfall gradient, and where a farm sits on that gradient decides most of how it runs. In the wetter Palouse country you can plant every year. Go west into the intermediate zone around Davenport, which gets twelve to seventeen inches of rain a year, and most farms run a three-year rotation. One year of fallow, where the field stays unplanted so the soil can store water. Then winter wheat. Then a spring cereal. In the driest areas it's wheat and fallow, two years, because the wheat can't get established without the water that builds up during that long empty stretch.
Aaron Esser, an extension agronomist with Washington State University, has worked that intermediate zone for years. He and Derek Appel run a 320-acre research farm near Davenport and write down what they do in a lot of detail: direct seeding, no-till fallow, soil sampling to four feet, fertilizer rates that vary across the field. Their 2024 bulletin records rotations run side by side to see whether you can get past the three-year system and push a fourth crop into the sequence, if the water allows it. It's a research farm, not a commercial one. But the decisions in that bulletin are real decisions, made by somebody standing at one point on the rainfall gradient, looking at soil cores and yield numbers and trying to work out what the water will carry. The answer keeps coming back to the water.
What I noticed, going through the documented practices for this region, is that every adjustment comes with a catch, and somebody went out and measured it.
A 2025 cover crop trial in inland Northwest wheat systems found that cover crops sown in the fall grew useful biomass and could stand in for one herbicide application during fallow. At the low-rainfall site, though, the covers dried out the soil going into fall. They used water that was supposed to be there for the next wheat crop. In a recent Washington state survey, three percent of dryland wheat acres were cover-cropped. I don't think that's because people out there haven't heard of cover crops.
A 2025 stubble management study looked at how much of the last crop's stalks to leave standing during fallow. Taller stubble cut the wind and changed temperatures right near the ground. But over two years at two sites, with four residue treatments and two row directions, there was no statistically significant difference in stored soil water or in wheat yield. The air over the field changed. The water in it stayed about the same.
A regional analysis of 45 winter-wheat varieties found that the variety explained 27 to 58 percent of yield differences within a single site. That's a real choice with real consequences. Picking the right variety for your field matters. Across sites, though, environment explained 84 percent of the variation. The right variety still has to grow wherever it is on the gradient.
In the lowest-rainfall zones, extension guidance says that under no-till fallow, seed-zone moisture in the preferred planting window, late August to mid-September, is almost always too low. Seed-zone moisture is what it sounds like. Whether the top few inches of soil are wet enough for a seed to sprout. When they aren't, farmers plant in October, or they dust the seed into dry ground ahead of the fall rain and wait. Planting late costs seven to twenty bushels an acre. Modern drills and nutrients placed with the seed can take some of that back. They can't put water in the ground that would have let you plant on time.
WSU's 2024 variety trials recorded poor emergence where seed-zone moisture was marginal, winter injury to cold-sensitive lines when a cold snap came with no snow on the ground, and lighter grain fill during a long stretch of July heat. Breeding a wheat that handles dry ground doesn't do anything about frost, or disease, or grain quality, or when you can get it harvested. You work on one exposure and you can make another one worse.
What the drought didn't sort
A 2023 state survey of dryland wheat management in Washington found 72 percent of surveyed acres under reduced-till or no-till. No-till means you don't plow between crops. You leave the old crop's residue on the surface, which holds the soil against erosion and can hang on to some moisture. Eighty-two percent of the people using it said equipment cost was a problem.
The survey also asked about 2021. Ninety-seven people, farming more than 450,000 dryland wheat acres between them, answered questions about what happened that year. Ninety-seven percent reported lower yields and 94 percent reported drier soil. Those numbers held across management categories. Plowed or not, almost everybody lost yield that year.
Washington's statewide winter-wheat yields over the last decade jump around a lot. Seventy-eight bushels an acre in 2016. Forty-two in 2021. An estimated 79 in 2026, according to the USDA's September forecast. Those are statewide averages that mix rainfall zones, wheat classes, and farming systems, so you can't pull a dryland trend out of them or pin any one year on climate change. What the swing from 42 to 79 does show is the main thing moving outcomes in this system, which is water. The adjustments are real and the people making them know what they're doing. They're doing it with whatever water the year hands them.
The starting balance
There's a second study I keep reading alongside the first. Published in Nature Geoscience in 2026, Black et al. mapped places where agricultural drought is emerging under climate change. Western North America is one of them.
The study defines agricultural drought as two things together: unusually low soil moisture at the start of the growing season, and unusually weak buildup of soil moisture during it. Across the northern mid- and high latitudes, the projected change in growing-season soil moisture lined up with the condition of the soil going into the season at a correlation of 0.95. With what built up during the season, the correlation was 0.01.
Projected soil-moisture change correlates at 0.95 with pre-season conditions and 0.01 with what accumulates during the growing season. The water you start with is nearly the entire story.
The reason is that warming increases evapotranspiration, meaning the water that evaporates off the soil and moves out through the plants. That increase outruns whatever extra rain the models project. Over most of northern North America, the period that sets things up is March through May, and the models show spring soil-moisture buildup declining across most northern regions outside the tropics.
This is a large-scale climate result. It isn't a measurement from anyone's field. It doesn't say how many inches a particular Davenport field will lose out of what it stores during fallow. What it does is describe which way the larger system is heading, the one those fields sit inside. And it fits with what the practice-level research already shows, which is that everything farmers do out there needs enough water to work with. A cover crop that uses fall moisture. A fallow year that stores spring moisture. A planting window that needs water in the top few inches. Each one draws on a starting balance, and the regional projections show that balance going down.
We've written before in this section about the difference between total rainfall and the water a crop can actually use, and in a later piece about how the surface can be wet while the deeper soil stays depleted. The Black et al. finding adds timing to that. It's about what's in the ground before anything starts growing.
Beyond wheat
Ninety-seven percent of Washington's wheat acres carry crop insurance. Insurance takes on part of the money side of a bad year. It doesn't add water or save any of the crop. It keeps the farm going while the exposure stays where it was. In Hultgren's model, insurance payments don't count as a yield effect at all.
So there's what keeps an operation running, and there's what a statistical model can see as protection, and they aren't the same thing. I don't think that's only true for wheat.
I've been thinking about it with other kinds of adjustments, too. To a house, or a city, or what you assumed about the place you live. The generator you bought after the freeze. The flood insurance you renewed at a higher premium. The retirement plan you changed after watching your state's summers change. Those are real adjustments and each one works, within its conditions. But the generator needs a grid that holds up well enough for it to fill in. The insurance needs a market that keeps writing policies in your zip code. And the retirement plan assumes your house keeps its value.
I'd want to know what my own adjustments depend on. That's a different question from whether they were good ones, and they probably were. The generator was a smart buy. The insurance is worth carrying. But a generator answers a grid that failed once. It doesn't tell you whether the grid will fail differently next time, or more often, or in a way the generator can't cover. Flood insurance is a price on a risk somebody else modeled. When the models update, the price moves. If it moves far enough, that tells you something by itself. You can make the adjustments. You mostly can't touch what they're sitting on.
Read narrowly and honestly, the Hultgren study says this much: the way wheat producers have historically adjusted doesn't produce a statistically meaningful reduction in the model's global wheat losses. It can't tell you when a particular farm runs out of options. And the eastern Washington work can't confirm or overturn a global number.
Put them next to each other, though, the global model that can't find a wheat adaptation signal and the field research where every practice turns out to have its own water tradeoff, and you get some idea of what this looks like from inside an operation. People doing careful, skilled work, each piece of it good for what it is, all of it drawing on water that comes in at a level nobody out there gets to set.
WSU's 2025 statewide synthesis says that drier ground at seeding could mean more fallow in some dryland places. More fallow means fewer cash crops in the rotation. That's an adjustment farmers have. It costs a year with no crop to sell. And it only makes sense as long as a fallow year can still store enough water to be worth that.
I don't know where the line is. The research doesn't either, not for one farm in one year. What it does show is that the adjustments are real, the limits inside them are measurable, and the water they depend on is, at the regional scale, heading the wrong way. I figured that was worth knowing, even without a clean answer about what comes next.
-
Tail risk at moderate warming: A 2026 Nature study found that adverse drought outcomes in some climate models at 2°C of warming can exceed multimodel-average outcomes at 3°C or 4°C, which means planning around the average projection may miss the scenario that actually breaks a farm or a region.
-
Adaptation that creates new problems: A scoping review of 89 empirical studies across 21 sectors found that maladaptation is generally multi-level and multi-causal, using shoreline armoring as one example of a measure that protects one place while worsening erosion elsewhere — the same conditional-tradeoff pattern visible in dryland wheat practices.
-
Disaster and housing instability: An Australian longitudinal study found that disaster exposure roughly doubled the odds of moderate-to-severe mental disorder among people with pre-existing conditions, with residential instability and housing-payment arrears among the strongest associated factors — a reminder that financial instruments like crop insurance address one kind of stability while leaving others exposed.
-
Hybrid approaches and publication bias: A meta-analysis of 418 adaptation studies across 78 countries reported better outcomes for hybrid approaches than single-domain interventions, though publication-bias adjustment reduced the estimated advantage by 18 percent, suggesting that even the evidence base for what works may overstate the case.
