Robust negative impacts of climate change on African agriculture

Picture a farmer in Sub-Saharan Africa standing at the edge of a maize field sometime around 2050. The rains came, the seeds went in, and now the stalks are shorter than they should be. The harvest will be smaller. By how much? According to Wolfram Schlenker and David Lobell, the answer — backed by forty-five years of data and sixteen climate models — is about twenty-two percent less than what that field would have yielded today. That number isn't a worst-case scenario. It's the median. And for four of the five staple crops they studied, there's a ninety-five percent probability that losses will exceed seven percent. That kind of precision is new. For a long time, estimates of climate impact on African agriculture ranged so widely that they were almost useless for planning. Some studies warned of yield drops as high as fifty percent by 2020. Others were more optimistic. The disagreement wasn't because scientists were careless — it was because the problem is genuinely hard. Crop simulation models, which try to mechanistically model plant growth, require detailed data on soil properties and farm management practices that simply don't exist for much of Sub-Saharan Africa. So those models produced point estimates — single best-guess numbers — with almost no information about how uncertain they were. Purely statistical approaches fared little better, because the historical agricultural records for Africa are thin and patchy, which made confidence intervals enormous. The result was a scientific literature where almost any claim about African crop losses could find some support, and where policymakers trying to prioritize adaptation investments had little to anchor decisions to. Schlenker and Lobell respond to this with a panel analysis — a method that follows the same countries over many decades and watches how their yields move in response to weather. Their data come from Food and Agriculture Organization yield and harvested area records from 1961 to 2006, matched to gridded weather data. The five crops they focus on are the staples that feed the continent: maize, sorghum, millet, groundnut, and cassava. Each country becomes its own long-running experiment, and the key move is using what statisticians call country fixed effects — a mathematical way of stripping out everything about a country that stays constant over time. Soil quality, average management practices, cultural factors — all of that gets absorbed into a country-specific constant and removed from the equation. What's left is the signal: how does yield in a given country go up or down in response to that year's weather? The weather variable they settle on is degree days, and it's worth pausing on what that actually measures. A degree day is not an average temperature. It's a running total of heat exposure above a threshold. Schlenker and Lobell use two variables: cumulative exposure between ten and thirty degrees Celsius, which captures the warmth crops need to grow, and cumulative exposure above thirty degrees Celsius, which captures the heat stress that damages them. Think of it as two meters running simultaneously through the growing season — one measuring productive warmth, one measuring harm. This specification, they note, tends to give conservative — lower — damage estimates compared to other formulations, which makes the results that emerge from it all the more striking. To generate their mid-century projections, they apply future climate scenarios to those historical weather series using sixteen different climate models under the A1B emissions scenario, targeting the period from 2046 to 2065. They also run one thousand bootstrap samples of their yield-response parameters. Combine those two sources of uncertainty and you get sixteen thousand simulated outcomes — a full distribution of what the future might look like, not just a single number. The results from that distribution are consistent and alarming. Median projected changes in total Sub-Saharan African production by mid-century are minus twenty-two percent for maize, minus seventeen percent for sorghum, minus seventeen percent for millet, minus eighteen percent for groundnut, and minus eight percent for cassava. The probability framing matters as much as the medians. For all crops except cassava, there is a ninety-five percent probability that damages exceed seven percent — meaning a small decline is almost certain. And there is a five percent probability that losses exceed twenty-seven percent — meaning a catastrophic outcome is not a tail risk to be dismissed, but something planners need to hold in mind. Cassava is a partial exception throughout: its smaller projected loss of eight percent likely reflects its flexible growing season and the fact that the weather variables explain less of its yield variation. Temperature is the dominant driver. Precipitation changes matter, but the bulk of the projected damage traces back to warming — specifically to the accumulation of days with temperatures above thirty degrees Celsius. That finding shapes what comes next, and it leads to the most counterintuitive result in the paper. You might expect that the countries best positioned to absorb climate shocks are the ones with the most advanced agriculture — the ones that have invested in fertilizer, modern seed varieties, and irrigation. Schlenker and Lobell find the opposite. Countries with the highest average yields have the largest projected losses. They separate a high-fertilizer subsample — South Africa and Zimbabwe, which have the highest fertilizer use in Sub-Saharan Africa — and find that the weather coefficients for this group are statistically different from the rest of the continent, with a p-value below five percent. The high-fertilizer countries have higher baseline yields, but they are also more susceptible to temperature increases. Modern, high-input crop varieties appear to be optimized for the climate conditions they were bred under, and when temperatures push above their tolerance thresholds, they fail harder than lower-yielding traditional varieties. The implication cuts against a common assumption in development policy. Expanding fertilizer use and improving seeds raises productivity — but under projected warming, it may simultaneously increase vulnerability. Schlenker and Lobell are explicit: higher fertilizer rates will tend to increase yield vulnerability to warming even while raising overall average yields. The low-fertilizer subsample still shows significant losses — the ninety-fifth percentile of the damage distribution remains negative for most crops — but the gradient is real. Better farms, as currently constituted, lose more. This matters enormously for how development and adaptation investment gets allocated. If modern agricultural inputs increase heat sensitivity, then the standard pathway of agricultural development — push yields up through better inputs, and resilience will follow — doesn't hold under climate change. The case for parallel investment in heat-tolerant and drought-tolerant varieties, and for breeding programs that target high temperatures rather than just high yields, becomes much stronger. What makes Schlenker and Lobell's contribution durable isn't just the specific numbers — it's the structure around those numbers. The fact that they report full percentile ranges, drawn from sixteen thousand simulated outcomes, means decision-makers can treat the results as a risk envelope rather than a single prediction. Knowing there's a ninety-five percent chance of at least a seven percent loss, and a five percent chance of a twenty-seven percent loss, is more actionable than knowing someone's best guess is fifteen percent. The two sources of uncertainty — climate model spread and statistical uncertainty in the yield response — turn out to be roughly comparable in magnitude. That itself is useful: it tells you that improving the climate projections and improving the yield-response estimates would both pay off in terms of narrowing the range. There are real limitations. Coarse weather grids miss local variation. The panel models capture how farmers respond to short-term weather shocks, not necessarily how they'd respond to a permanent shift in climate. Carbon dioxide fertilization effects — the possibility that higher carbon dioxide concentrations boost plant growth — are not accounted for, and the evidence for those effects in African field conditions is sparse. The authors are candid that these gaps could mean their estimates are conservative: the actual losses might be larger. But that's the tension the paper leaves open. Schlenker and Lobell have given the field something it didn't have before — a robust, empirically grounded set of projections with explicit uncertainty bounds, derived from nearly half a century of observed yield and weather data. The crops that feed Sub-Saharan Africa are under serious threat from warming. The most productive farms may be the most exposed. And the probability of significant losses, even under conservative assumptions, is high enough that treating this as anything other than a near-certainty would be a mistake. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.