Land-atmosphere feedbacks contribute to crop failure in global rainfed breadbaskets

Ensuring food security for an increasing population is a major challenge. Dry and hot spells can cause crop failure, and their potential aggravation due to global warming heightens risk. Approximately one-third of global inter-annual crop yield variability is attributed to fluctuations in weather and climate, though the relative importance of local climate variability remains debated. While irrigation can mitigate water and heat stress on crops, it also pressures freshwater resources; thus rainfed agriculture—responsible for ~75% of global croplands and ~60% of global food production—remains essential for sustainable food supply and for smallholder livelihoods. Understanding climate vulnerability of global rainfed breadbaskets is therefore critical for adaptation planning. Heat stress can directly damage plants and increase atmospheric aridity and soil moisture deficits. Local land-atmosphere feedbacks play a role: as soils dry, evaporation decreases and sensible heating increases, enhancing temperatures and aridity, which can reduce yields. These local feedbacks are conditioned by large-scale synoptic systems and climate oscillations (e.g., ENSO, NAO), which influence local climate and can induce synchronous crop losses, sometimes via Rossby wave configurations that elevate heatwave exposure across major agricultural regions. Such atmospheric patterns alter the flow of heat and moisture, producing anomalies in imports to breadbasket regions that translate into local temperature and precipitation anomalies and can be amplified or attenuated by local feedbacks. Land-atmosphere feedbacks may also propagate dry/hot anomalies downwind by altering moisture and heat advection and potentially affecting larger-scale circulation. Although recent studies have analyzed local land-atmosphere feedbacks on crops, the dependency of yields on upwind climate conditions is poorly understood. This study examines 75 global rainfed breadbaskets for maize, wheat, soybean, and rice, relates low-yield years to local climate anomalies, and, using a novel atmospheric Lagrangian framework combined with satellite observations, investigates sources of heat and moisture during low-yield years, partitioning contributions from oceans, upwind land, and local land. The study evaluates the influence of local and upwind land-atmosphere interactions on anomalous moisture and heat transport and their role in crop failures.

The paper situates its research within evidence that climate variability explains a substantial portion of global crop yield variability (~one-third). Prior work documents the impacts of extreme heat and compound hot-dry events on yields across major crops and regions (e.g., US maize, European drought/heat, global analyses). Studies highlight stronger temperature–moisture coupling exacerbating warming impacts on yields, and the combined influence of soil moisture and atmospheric evaporative demand on prediction accuracy. Large-scale modes and synoptic dynamics (ENSO, NAO, Rossby wave amplification) have been linked to synchronous or widespread crop failures by modulating heat and moisture transport and preconditioning local land-atmosphere coupling. Land surface processes can propagate droughts and heatwaves downwind through reduced evaporation and enhanced sensible heating, and land-use change (e.g., deforestation, irrigation) can alter moisture recycling and regional climates. However, despite advances on local land-atmosphere couplings affecting productivity, the explicit role of upwind source regions and en route feedbacks in shaping yield outcomes remained underexplored, motivating this study.

Study domain and crops: The authors identified 75 rainfed breadbaskets for maize (29), wheat (25; spring and winter combined), soybean (12), and rice (9) on a 1° global grid. A region qualifies as a rainfed breadbasket if it includes ≥18 contiguous 1° grid cells (~150,000 km²), rainfed fraction >75% in each pixel (from MIRCA2000), and uniform sowing/harvest months (from global crop calendar). Yield data and detrending: Grid-cell yields (GDHY, 1981–2016) were aggregated to breadbaskets using area-weighted rainfed production. Time series were detrended using LOWESS to remove long-term technological trends, yielding relative yield anomalies (η). Crop failure events were years with η below the 25th percentile (1983–2015), typically nine years per breadbasket. Local climate characterization: For precipitation (MSWEP v2.2) and 2 m temperature (ERA5), anomalies were computed over an extended growing season (one month prior to sowing through harvest). Evaporation and potential evaporation were from GLEAM v3.5a (land) and OAFlux (ocean). Aridity index Ep/P (annual climatology, 1983–2015) classified breadbaskets as water-limited (Ep/P>1) or energy-limited (Ep/P<1). Statistical differences were tested with Mann–Whitney U tests. Heat and moisture source attribution (HAMSTER³): The study employed the HAMSTER framework based on FLEXPART v9.0, driven by ERA-Interim (1°; three-hourly), to track 2 million equal-mass air parcels globally (1979–2019; analysis 1983–2015). For each breadbasket and each day from one month before sowing to harvest, air parcels residing over the breadbasket were identified and traced back 15 days. Parcel properties (position, height, specific humidity, density, temperature) were used to diagnose where parcels were moistened (surface evaporation) and warmed (surface sensible heat flux), producing source–receptor relationships for precipitation (moisture imports) and atmospheric energy (heat imports). Trajectories accounted for rain-out and nighttime cooling. Source contributions were bias-corrected using MSWEP precipitation, GLEAM and OAFlux evaporation, and ERA-Interim sensible heat. Partitioning of sources and anomalies: For each breadbasket-year, total imports of moisture and heat were partitioned into ocean (o), upwind land (u), and local land (l) contributions: P_import^i = P_import^l + P_import^u + P_import^o; H_import^i = H_import^l + H_import^u + H_import^o. Relative anomalies (%) in imports from each origin were computed with respect to the total import for that breadbasket and season. Spatial anomalies in source regions during crop failures were mapped by aggregating events and summing overlapping source regions. Analysis of land–atmosphere feedbacks: The authors assessed associations between yield anomalies and joint anomalies in heat and moisture imports from each origin, separately for water- and energy-limited breadbaskets, using two-dimensional probability densities (crop-loss-weighted Gaussian kernels). They defined concurrent local and upwind land feedback events (CLF) as crop failure years with simultaneous negative moisture and positive heat import anomalies from both upwind land and local land. Differences in yield deficits between CLF and other events (no-CLF) were quantified and tested (Mann–Whitney U).

• Crop failure events (n=675 across 75 breadbaskets, 1983–2015) show spatially varying average yield deficits: wheat exhibits the largest average loss (~15%) and rice the lowest (~7%); individual breadbaskets can reach up to ~50% loss in certain years.
• Water-limited breadbaskets generally suffer higher losses than energy-limited ones (p<0.05), consistent with higher interannual climate variability and dependency on precipitation in water-limited regions.
• Most crop failures coincide with precipitation deficits and higher temperatures during the growing season, especially in extra-tropical and water-limited regions. Some energy-limited tropical regions experience above-normal precipitation during failures (potential waterlogging and reduced radiation effects).
• Source attribution reveals widespread anomalies in imports during failures: nearby oceans typically supply less moisture than usual; strongest negative moisture and positive heat anomalies occur over land within and near breadbaskets. Heat anomalies almost always originate over land.
• En route amplification: Negative oceanic moisture anomalies intensify over continents while upwind land imposes positive sensible heat anomalies as air travels toward breadbaskets, evidencing upwind land–atmosphere feedbacks that amplify initial synoptic anomalies.
• Ocean contribution statistics: In water-limited breadbaskets, ~60% of crop failures occur with below-normal oceanic moisture imports across all crops; for water-limited wheat, 76% of failures coincide with below-normal ocean moisture imports.
• Upwind land contributions: During failures, most events exhibit moisture deficits from upwind land and disproportionately positive heat anomalies (“upwind feedback”), especially in water-limited maize and wheat (38% and 37% of failures, respectively, show both negative moisture and positive heat anomalies from upwind land).
• Local land contributions: Upon arrival over breadbaskets, local feedbacks further increase heat anomalies and decrease moisture imports, shifting distributions toward the lower-right (dry–hot) quadrant; for water-limited maize and wheat, 47% and 65% of failures, respectively, occur with both local drying and warming (p<0.05).
• Concurrent feedback amplification (CLF): When both upwind and local land exhibit concurrent negative moisture and positive heat import anomalies, yield deficits are substantially larger: • Water-limited regions: average deficits 37% higher overall; maize +43% (15.2%→21.6%, p<0.05), wheat +25% (16.9%→21.2%, p<0.05). • Energy-limited regions: maize +48% (7.4%→11.1%, p<0.05); wheat +29% (11.4%→14.7%, not significant). • Across all breadbaskets and events: crop failures are on average 42% larger under concurrent local and upwind land feedbacks.
• Notable exception: increased moisture contribution from Amazonia to maize breadbaskets in Venezuela and Colombia during failures, consistent with specific tropical dynamics.

The study addresses the research gap regarding the role of upwind land–atmosphere processes in crop failures. By tracing heat and moisture sources with a Lagrangian framework and linking them to yield anomalies, the authors show that both upwind and local land feedbacks amplify precipitation deficits and heat anomalies during low-yield years. This mechanism explains part of the severity and spatial extent of failures, particularly in water-limited breadbaskets where reliance on advected moisture is high. En route feedbacks from ocean-to-land trajectories highlight that synoptic-scale circulation anomalies are intensified by land feedbacks as air masses cross continents. The findings underscore the importance of upwind land conditions and management for downwind agricultural productivity. Deforestation or land degradation upwind can reduce evaporation and moisture recycling, diminishing precipitation in breadbaskets and elevating sensible heating, thereby increasing failure risk. Conversely, monitoring and managing upwind soil moisture and vegetation conditions have potential to mitigate risks. Implications include improved seasonal crop forecasting by assimilating upwind land surface states (with soil moisture/vegetation memory) from satellite observations, enabling earlier warnings for at-risk breadbaskets. The demonstrated propagation of droughts and heatwaves downwind via reduced evaporation and enhanced sensible heat offers a mechanistic basis for forecasting compound dry–hot events affecting agriculture. While statistical associations align with physical interpretations from trajectories, causality at the yield level remains subject to confounders; nevertheless, the integrated framework strengthens the link between atmospheric source anomalies, feedbacks, and agricultural outcomes.

This work maps the spatio-temporal origins of moisture and heat for 75 global rainfed breadbaskets and demonstrates that concurrent upwind and local land–atmosphere feedbacks substantially amplify crop failures—on average by ~42%—with the strongest effects in water-limited regions and for maize and wheat. The results highlight the critical role of upwind land conditions in modulating precipitation and temperature anomalies over agricultural regions and establish en route amplification of synoptic anomalies by land feedbacks. These insights support adaptation strategies that consider upwind–downwind dependencies, including land management interventions, conservation of upwind ecosystems, and incorporation of upwind soil moisture and vegetation state monitoring into seasonal forecasting and early warning systems. Future research should: refine source attribution with improved observational constraints (e.g., water isotopes), quantify causal pathways between upwind land management changes and downwind yields, assess dynamic shifts in breadbasket extents under climate and management change, and integrate coupled land–atmosphere–crop models to test mitigation strategies.

• Lagrangian trajectory uncertainties: HAMSTER/FLEXPART-derived moisture recycling ratios may be on the upper end of model uncertainties; trajectory errors and representation of subgrid processes affect source attribution.
• Bias correction and observational sparsity: Limited ground observations constrain validation of evaporation, sensible heat flux, and precipitation source estimates; bias corrections rely on reanalyses and satellite products with their own uncertainties.
• Statistical, not strictly causal, yield relationships: While physical trajectory analysis supports mechanisms, the statistical associations between climate anomalies and yields are not necessarily causal at the yield level.
• Static breadbasket delineation: Breadbasket extents are held constant (year-2000 configuration), ignoring agricultural expansion/contraction, irrigation adoption, or poleward shifts under climate change.
• Management and technological trends: Detrending removes long-term changes, but interannual management decisions and technological variations are not explicitly modeled, contributing to unexplained yield variance.
• Limited event samples for some crops/regions: Fewer soybean and rice breadbaskets and CLF events reduce statistical power for those categories.