U.S. winter wheat yield loss attributed to compound hot-dry-windy events

Wheat yields, despite global increases, have stagnated or collapsed in some regions, raising concerns for sustainable production under climate change. Prior studies have typically assessed single extreme events (e.g., drought, heat waves, cold events) or examined multiple extremes in conjunction but not as compound events. The IPCC has defined compound events as combinations of multiple drivers and/or hazards that contribute to risk. Few efforts have quantified the impacts of compound events on crops. The simultaneous combination of high temperature, low relative humidity, and high wind—termed hot-dry-windy events (HDW)—poses particular risks, especially in the U.S. Great Plains, historically affected by the 1930s Dust Bowl. Traditional climate–crop analyses often use fixed seasonal calendars, potentially misrepresenting phenology-specific sensitivities. This study instead uses phenology-based climate indices aligned with three winter wheat stages—planting to jointing (PT–JT), jointing to heading (JT–HD), and heading to maturity (HD–MT)—to evaluate the impacts of HDW on yields across the U.S. winter wheat belt. HDW are defined at hourly resolution as co-occurrences of temperature ≥32 °C, relative humidity ≤30%, and wind speed ≥7 m s−1. Given evidence that atmospheric dryness can rapidly induce crop water stress even when soil moisture is adequate, low relative humidity is used to define the dry component. The study focuses on often-overlooked hourly HDW during HD–MT (grain filling) from 1982–2020 in SD, NE, CO, KS, OK, and TX, which produce nearly all U.S. hard red winter wheat.

The paper situates its work within literature on climate extremes and crop yields, noting that most prior research has assessed single extremes (e.g., precipitation-based drought, heat waves, cold events) and that few studies have explicitly evaluated compound events as defined by the IPCC. Prior research indicates HDW-like conditions (high temperature, low humidity, high winds) occur in the Great Plains and that atmospheric dryness can be a dominant driver of plant stress. Historical context includes the 1930s Dust Bowl, highlighting susceptibility of the region. Existing studies have examined temperature-moisture couplings and their exacerbating effects on crops, and have identified global hotspots for compound events, but the specific impact of HDW on hard winter wheat yields in the U.S. had not been assessed prior to this work.

Study domain and data: The analysis covers 339 counties across SD, NE, CO, KS, OK, and TX with >5000 acres of harvested hard winter wheat and at least 20 years of yield data (USDA-NASS, 1982–2020; irrigation fraction <5% on average). Phenology: Planting dates from 230 stations and harvesting dates from 186 stations were compiled; heading dates were documented at three stations in Texas. Maturity dates were inferred as 14 days prior to harvest. Jointing and heading dates were estimated using accumulated growing degree days. Each county-year was partitioned into three phenological stages: PT–JT, JT–HD, and HD–MT. Climate data: Hourly 2-m air temperature and relative humidity, and 10-m wind speed were from ERA5-Land (1982–2020). Daily precipitation was from PRISM. County-level climate variables were taken from nearest grid points to county centroids. Quality-assured HadISD station observations were used to verify spatial patterns of annual HDW during HD–MT. HDW definition: Hourly HDW is counted when T ≥32 °C, RH ≤30%, and U ≥7 m s−1 at 10 m. Analyses also considered individual (H, D, W), bivariate (H&D, H&W, D&W), and trivariate (HDW) events during HD–MT. Statistical modeling: A linear mixed-effects model with county fixed effects (time-invariant heterogeneity) and year fixed effects (technology trends) related yields to climate indices: freezing days (Frez), extreme degree days (EDD; degrees above 32 °C), precipitation (Prcp; quadratic term by stage), and HDW hours (accumulated during HD–MT). Models were run using standardized predictors (z-scores) for comparability of sensitivities and with variables in original units for direct interpretation. Two-way clustering standard errors (county and year) addressed heteroskedasticity and autocorrelation; adjusted p-values were reported. Additional models with quadratic temperature and temperature bins captured nonlinear temperature effects without offsetting HDW parameters. Trends and teleconnections: Trends were estimated by OLS with significance from Mann–Kendall tests; Poisson regression was also tested for HDW counts. Teleconnections between PDO and HDW anomalies (April–June, 1951–2020) were assessed using partial Spearman correlation controlling for mean temperature, with 9-year moving averages for smoothing. Yield trends attribution: Climate-driven yield trends were quantified by multiplying yield sensitivities (from original units models) by trends in corresponding climate indices. Yield shocks: Relative yield change was computed against a 5-year running mean; a yield shock year was defined as falling below the county’s 25th percentile over 1982–2020. Concurrent anomalies in climate indices during shock years (difference from 1982–2020 mean) identified potential drivers. Robustness checks: HDW thresholds were varied across 36 combinations [T: 30–33 °C; RH: 25–35%; U: 6–8 m s−1]. Average HDW yield sensitivity across combinations was −0.09 t ha−1 per 10 h (range −0.06 to −0.13), consistent with the main threshold choice. Temperature-response robustness was evaluated using quadratic and binned temperature models; HDW effects remained robust.

• Spatial patterns and trends: Frequent HDW during HD–MT concentrate in southwest Kansas and the Oklahoma/Texas panhandles. County-level HDW_HD–MT increased significantly across much of the region, with rates up to ~8 hours per decade, and patterns overlap the 1930s Dust Bowl footprint. Partial correlations indicate high temperature is the dominant control on HDW occurrence. • Yield variability explained: The mixed-effects model explains 59% of yield variation (R^2 = 0.59). • HDW impacts: HDW during HD–MT is the dominant driver of yield loss: −3.5% per standard unit (z-score) or −4% per 10 hours of HDW (padj < 0.001). EDD during JT–HD also reduces yields (−2.2% per standard unit; padj < 0.001). Among event types, hot (H), windy (W), hot–windy (H&W), and dry–windy (D&W) are not significant, while hot–dry (H&D) significantly reduces yields; trivariate HDW has the largest negative effect. • Timing within HD–MT: HDW occurring in the middle sub-stage (around flowering) has the largest negative yield impact, followed by the last sub-stage; early sub-stage impacts are smaller. • Climate-driven yield trends: In severely HDW-affected counties, HDW trends during HD–MT are associated with yield losses up to −0.09 t ha−1 per decade, the largest among considered climate drivers. • Teleconnections: HDW anomalies (April–June) are strongly anti-correlated with PDO (ρ = −0.65) over 1951–2020, with decadal variability and an approximate periodicity >40 years; anomalies typically range from −30 to +30 hours. • Yield shocks: Shock years (bottom quartile relative to 5-year mean) align with climate anomalies: elevated EDD during JT–HD across a central diagonal (OK/TX panhandles to NE KS), elevated EDD during HD–MT in western areas, excessive HD–MT precipitation in the eastern Southern Great Plains (linked to waterlogging, disease, lodging), and lower precipitation (drought) in western/northern areas. HDW-related shocks align with the Dust Bowl footprint.

The study demonstrates that compound HDW events during grain filling are the most impactful climate driver of winter wheat yield loss in the U.S. Great Plains, exceeding the effects of single-variable extremes. By aligning climate indices with phenological stages, the analysis clarifies stage-specific sensitivities, especially the high vulnerability around flowering and late grain filling. The strong PDO–HDW anti-correlation suggests an atmospheric bridge modulating decadal HDW variability and provides context for decadal-scale risk. The spatial overlap of severe HDW impacts with the historic Dust Bowl region underscores persistent regional vulnerability. These findings indicate that traditional risk assessments focusing on individual extremes may underestimate risks from compound events. Adaptation strategies should consider concurrent hot, dry, and windy conditions, with attention to variety selection (heat/drought tolerance), phenology management, and agronomic practices to mitigate rapid atmospheric water demand and wind damage.

HDW frequency and intensity during heading to maturity have increased across the U.S. hard winter wheat belt over the past four decades, concentrating in areas overlapping the Dust Bowl region. Compound HDW events are the most influential climate factor for yield reductions, with approximately 4% loss per 10 hours of exposure and climate-driven yield declines up to 0.09 t ha−1 per decade in heavily affected counties. Yield shocks and their spatial patterns correspond to anomalies in HDW and other climate indices, reinforcing the importance of phenology-specific compound hazard assessments. The results highlight that climate change impacts on wheat productivity arise not only from shifts in single-variable extremes but also, critically, from compound events. Future research should refine understanding of HDW frequency and intensity effects by sub-stage, improve representation of atmospheric dryness and wind interactions in crop models, and explore adaptive management and breeding to reduce vulnerability, considering decadal modulation by climate modes such as the PDO.

• Phenological data limitations: Heading dates were available from only three stations in Texas, and maturity dates were inferred as 14 days before harvest for all county-years, introducing uncertainty in sub-stage timing. • Compound exposure complexity: Crop vulnerability under HDW depends on variety-specific heat/drought tolerance, phenology, and management practices, as well as interactions with pests and diseases, which were not explicitly modeled. • Data sources and resolution: Reliance on reanalysis (ERA5-Land) for hourly temperature, humidity, and wind, and gridded PRISM precipitation, may introduce biases relative to local station conditions despite verification with HadISD. • Threshold definition: HDW thresholds capture transient atmospheric dryness; while robustness checks across 36 threshold combinations supported findings, threshold choices can influence event counts and sensitivities. • Model scope: The statistical approach controls for fixed effects and uses standardized indices but cannot fully capture all nonlinear and interacting biological processes beyond those tested (e.g., temperature bins/quadratic), and irrigation, though generally <5%, was not explicitly modeled.