The study, led by Zijie Zhao of the University of California Irvine and the University of Tasmania’s Institute for Marine and Antarctic Studies, together with Yanxuan Du of the University of Melbourne, relied on the ERA5 reanalysis covering 1979 through 2019. Rather than averaging rainfall over broad regions or assuming a canonical bimodal annual cycle, the team applied a detection method at the level of individual grid cells, identifying when and where midsummer drought conditions actually occurred across the domain. Each detected event was then fed into a K-means clustering analysis, a machine learning technique that partitions data points into groups based on their similarity—in this case, the temporal structure of the rainfall reduction. The result was a four-cluster classification that captures the true diversity of midsummer drought expressions, from the classic two-peaked rainfall curve of the Pacific slope of Central America to asymmetric and unimodal variants concentrated over the Caribbean basin and adjacent oceans.

This diversity matters because the midsummer drought—known regionally in Spanish as the canícula or veranillo—is far more than a curiosity of climatology. The mid-season reduction in rainfall coincides with critical stages of the agricultural calendar across Mesoamerica, where maize, beans, and other staple crops are planted and cultivated during the boreal summer wet season. A longer, deeper, or earlier drought can stress crops at their most vulnerable stages, and previous research has linked variations in the drought’s timing to large-scale oscillations including the Madden–Julian Oscillation. Farmers across Mexico and Central America have long organized their planting decisions around the expected pause in the rains, so any framework that mischaracterizes the drought’s structure risks misinforming both seasonal forecasts and long-term adaptation planning. The new cluster-based approach provides a vocabulary for distinguishing, for example, a drought that unfolds symmetrically around a July–August minimum from one whose decline and recovery follow a different rhythm entirely.

To uncover the physical mechanisms behind this diversity, the researchers turned to composite analysis, a technique in which atmospheric and oceanic fields are averaged over all events belonging to a given cluster, isolating the conditions that systematically accompany each drought type. They examined sea surface temperature anomalies, cloud fraction, low-level winds, and moisture flux convergence—the rate at which horizontally transported water vapor accumulates in a column of air, a key precursor to precipitation. The analysis revealed that low-level atmospheric circulation, rather than the ocean surface, plays the central role in shaping the midsummer drought’s many faces. In particular, the Caribbean low-level jet and the Chocó low-level jet, along with wind variations over the eastern tropical Pacific, emerged as the leading actors.

The Caribbean low-level jet, a fast ribbon of easterly wind that channels moisture across the Caribbean Sea toward Central America, has long been recognized as a linchpin of regional hydroclimate. When the jet strengthens, it can reorganize moisture transport and cloudiness in ways that suppress rainfall over parts of the isthmus while enhancing it elsewhere. The Chocó jet, by contrast, flows along the Pacific coast of Colombia and feeds some of the rainiest conditions on Earth over the far eastern tropical Pacific and western Colombia; its fluctuations modulate the cross-isthmus moisture exchange that sustains Central American rainfall. By showing that these jets—and eastern Pacific surface winds—systematically differ across the four drought clusters, the study provides a dynamical explanation for why the drought looks so different from southern Mexico to the Caribbean islands.

Perhaps the most provocative result concerns sea surface temperatures. For decades, hypotheses about the midsummer drought’s origin have invoked ocean-atmosphere feedbacks: warming sea surfaces in the eastern tropical Pacific, the northward migration of the intertropical convergence zone, and shifts in the North Atlantic subtropical high have all been proposed as drivers. Yet the composite analysis found that eastern Pacific sea surface temperature anomalies exhibit only weak and inconsistent associations with the identified drought clusters. This suggests that sea surface temperature–cloud feedbacks, often emphasized in earlier mechanism studies, play a secondary role compared to the direct dynamical control exerted by low-level winds and the moisture fluxes they carry. The finding does not render the ocean irrelevant—temperature anomalies can still modulate drought intensity on interannual timescales—but it repositions circulation as the primary architect of the drought’s structural diversity.

To dig deeper into the moisture budget, the researchers decomposed moisture flux convergence into its zonal (east–west) and meridional (north–south) components. This decomposition showed that the combined action of zonal and meridional moisture transport is the primary driver of the drought’s bimodality—the characteristic two-peak rainfall structure. Notably, meridional moisture fluxes proved especially important for drought events occurring over the Caribbean basin, where the north–south exchange of moisture between the tropical Atlantic and the surrounding continents governs whether the mid-season dry spell forms at all. This level of mechanistic detail is precisely what the cluster-based framework makes possible: by separating events into physically coherent groups, the analysis avoids smearing together distinct dynamical regimes that would otherwise cancel or blur one another in a region-wide average.

The study then tested whether state-of-the-art global climate models could reproduce this diversity. The team evaluated 33 models participating in the sixth phase of the Coupled Model Intercomparison Project (CMIP6), the international modeling effort underpinning the latest assessments of the Intergovernmental Panel on Climate Change. The verdict was mixed. Most models managed to capture the overall frequency of midsummer drought occurrence across the domain—a reassuring baseline. But the models systematically underperformed when it came to reproducing the asymmetric precipitation structures identified by the cluster analysis, with the largest failures concentrated over the Caribbean. In other words, models can simulate that a dry spell happens, but they struggle to get its shape right, particularly in the very region where meridional moisture transport dominates and where the study identified the most distinctive drought signatures.

This model shortfall carries real weight for climate projections. Previous work, including studies published in Climate Dynamics nearly two decades ago, has suggested that the midsummer drought may extend and intensify under greenhouse warming, with significant consequences for Central American water resources and agriculture. If models misrepresent the spatial and temporal structure of the drought today, projections of its future evolution inherit that bias. The authors argue that midsummer drought diversity should be incorporated into model evaluation frameworks—moving beyond simple checks of mean rainfall toward tests that ask whether models reproduce the full family of drought expressions and the circulation features that produce them. Such tests would give model developers a sharper diagnostic and give climate impact assessments a firmer footing.

The study’s methodology is itself a contribution. The objective, grid-cell-based detection algorithm—whose code the authors have made publicly available on GitHub—bypasses the subjectivity that has plagued midsummer drought research, where different definitions of onset, duration, and intensity have led to divergent estimates of how often the drought occurs and how it is changing. This definitional sensitivity was highlighted in earlier hydrological assessments showing that the very occurrence and recent trends of the Mesoamerican mid-summer drought depend on how it is defined. By anchoring the classification in data-driven clustering rather than prescribed regional averages, the new approach offers a reproducible standard that other researchers can adopt. Combined with the openly available ERA5 and CMIP6 datasets, it lowers the barrier for follow-up studies examining drought behavior under warming scenarios.

For Mesoamerica, the stakes are high. The region’s summer rainfall regime is influenced by a tangle of interacting systems—the Caribbean and Chocó jets, the intertropical convergence zone, the North Atlantic subtropical high, the eastern Pacific warm pool, and remote drivers such as the El Niño–Southern Oscillation and the Madden–Julian Oscillation. Untangling which of these governs which flavor of midsummer drought is a prerequisite for reliable seasonal prediction. A forecast system that understands, for instance, that Caribbean drought events hinge on meridional moisture fluxes can target its predictors accordingly, rather than treating the entire region as a monolith. The same logic applies to adaptation: agricultural extension services, water managers, and disaster-preparedness agencies all stand to benefit from forecasts that distinguish drought types rather than issuing a single regional outlook.

The research, supported by the Australian Research Council Centre of Excellence for Climate Extremes, ultimately reframes a familiar phenomenon. The midsummer drought that farmers in Oaxaca, Guatemala, and Costa Rica have observed for generations is real, but it is not a single entity. It is a spectrum of events governed by wind systems that thread moisture through and around the Central American isthmus, with the ocean playing a subtler role than often assumed. Recognizing that spectrum, the authors contend, is the foundation for more trustworthy seasonal forecasts and more credible climate change assessments in one of the world’s most rainfall-sensitive agricultural regions. As global models continue to improve, the four-cluster framework offers a demanding but achievable benchmark: not merely to predict that the rains will pause in midsummer, but to predict which pause will occur, where, and why.