Risks of synchronized low yields are underestimated in climate and crop model projections

The study investigates how meandering mid-latitude jet stream patterns (quasi-stationary Rossby waves with wavenumbers 5 and 7) influence concurrent extreme weather and synchronized crop yield losses across major breadbasket regions. With global warming increasing the frequency and intensity of extremes and their simultaneous occurrence, synchronized crop failures pose systemic risks to global food security and supply chains. Prior assessments largely used statistical approaches or focused on modes of climate variability; however, in Northern Hemisphere summer, circumglobal teleconnections linked to high-amplitude wave-5 and wave-7 patterns can promote simultaneous heat and rainfall extremes. A key gap is the quantification of wave-pattern impacts on yield co-variability between regions and the evaluation of whether current climate and crop models reproduce these high-impact events. The study aims to: (1) assess biases in climate models in reproducing upper-tropospheric jet patterns and associated surface weather anomalies; (2) evaluate whether crop models driven by reanalysis and bias-adjusted CMIP6 simulations reproduce observed regional yield anomalies related to these waves; and (3) quantify how wave events modulate concurrence of low yields across regions and compare observed signals to model outputs.

Previous work identified the jet stream and circumglobal teleconnections as drivers of concurrent mid-latitude extremes, with quasi-stationary Rossby waves (wavenumbers 5 and 7) implicated in major recent events. Risks of synchronized breadbasket failures have been assessed statistically and linked to large-scale climate variability, but impacts of specific wave patterns on interregional yield co-variability remained unquantified. Luo et al. (2021) showed that climate models underestimate the surface imprint of wave patterns despite reasonable phase representation, primarily in CMIP5-era models; bias-adjusted CMIP6 and linked climate–crop model performance had not been thoroughly evaluated. There is also limited assessment of historical wave-induced concurrent low yields in coupled climate–crop simulations and little quantification of future changes in wave patterns, their surface anomalies, and associated yield co-variability.

Data and models: ERA5 reanalysis (1960–2014) provided daily 2 m temperature, precipitation, and 250 hPa meridional winds, regridded to 1° and detrended (linear trend removal and seasonal cycle removal). W5E5 (1979–2014) bias-adjusted ERA5-based dataset was used to drive crop simulations. CMIP6 daily fields (historical 1960–2014, SSP5-8.5 2045–2099) from 12 models were regridded to 1°, detrended similarly; precipitation flux converted to m per hour. Four ISIMIP3 bias-adjusted and downscaled CMIP6 models (IPSL-CM6A-LR, MPI-ESM1-2-HR, MRI-ESM2-0, UKESM1-0-LL) were used for climate-driven crop simulations. Wave detection: Following Kornhuber et al., weekly averages of mid-latitude (37.5–57.5°N) 250 hPa meridional winds (JJA, 13 weeks) were decomposed using Fourier analysis. High-amplitude wave events for wave-5 and wave-7 were defined as weekly amplitudes exceeding 1.5 standard deviations above the JJA climatological mean per dataset. ERA5 yielded 54 wave-7 and 62 wave-5 events for 1960–2014. Regions: Five major Northern Hemisphere mid-latitude crop regions were defined using a 25% harvested-area threshold (Ray et al. 2005): North America (USA, Canada), Western Europe (France, Switzerland, Spain, Portugal, UK, Belgium, Netherlands, Germany), Eastern Europe (Greece, Bulgaria, Moldova, North Macedonia, Ukraine, Romania, Serbia, Albania, Russia), India, and Eastern Asia (China, Mongolia). An “all regions” grouping and global total were also computed. Crop modeling and yields: Global gridded LPJmL simulations at 0.5° for maize and wheat were driven by W5E5 (1960–2014) and the four bias-adjusted CMIP6 models (2045–2099). Observed national yields from FAOSTAT were aggregated to regions by production and harvested area. Modeled and observed yields were detrended using singular spectrum analysis (SSA) to isolate interannual variability relative to 10-year moving windows. Composite analysis: For each tier—Obs/Obs (ERA5 waves × FAOSTAT yields), Obs/Model (ERA5 waves × LPJmL driven by W5E5), and Model/Model (waves × LPJmL driven by bias-adjusted CMIP6)—composite yield anomalies were computed for years with ≥1 wave event vs years with 0 events. Regional yield values were standardized by dividing by the mean of the three years before and after event years. Uncertainty was quantified via 500 bootstrap resamples (~90% subsamples) to form distributions. Likelihood Multiplication Factor (LMF): The LMF quantifies how wave events change the likelihood of concurrent below-trend yields in two regions relative to no-event years: LMF = P(y1<0 ∩ y2<0 | w>1) / P(y1<0 ∩ y2<0 | w=0). Analogous calculations were done for concurrent positive anomalies. Poor/good years were defined by combined maize+wheat yields below/above the multiyear trend. Evaluation metrics: Composite fields of meridional wind, temperature, and precipitation anomalies during wave events were compared between models and ERA5 using Pearson correlation and coefficient of determination (R²) over 38–58°N. Historical vs future changes in anomaly patterns were assessed. Timing of wave events within JJA was compared across datasets.

• Atmospheric circulation: Bias-adjusted CMIP6 models reproduce the phase and intensity of upper-tropospheric wave-5 and wave-7 patterns, but substantially underestimate associated near-surface temperature and precipitation anomalies. Meridional wind anomaly fields show high agreement with ERA5 (correlations ~0.82–0.92; R² ~0.76–0.92), whereas temperature correlations are low (~0.03–0.51; R² ~0.02–0.28), and precipitation lower still. Bias adjustment did not markedly improve spatial correlations or anomaly magnitudes relative to raw CMIP6 for high-amplitude wave events.
• Future circulation anomalies: No consistent increase in overall wave amplitudes is projected under SSP5-8.5, but regional amplification of ridges/troughs leads to larger positive temperature anomalies over Western North America and East Asia for wave-7 and over Western North America and much of Eurasia for wave-5. Pattern correlation with observed temperature anomalies increases in future simulations partly due to amplified temperature responses beyond mean warming.
• Surface anomalies in crop regions: Most climate models underestimate wave-related temperature and precipitation anomalies over key breadbasket regions. Discrepancies are largest where regions align with strong wave-induced anomalies (e.g., WEU for wave-7; NA, WEU, EEU for wave-5). Reanalysis shows NA is hotter during wave-5 and drier during wave-7; WEU is wetter/colder during wave-5 and hotter/drier during wave-7; EEU is warmer during wave-5 and wetter during wave-7; IND and EAS are drier during wave-7.
• Observed regional yield impacts: Years with multiple wave events are associated with significant combined maize+wheat yield reductions: up to −7% (EAS), −6% (NA), and −3% (EEU) for wave-7, with regional means ~−2% to −3%. For wave-5, observed anomalies are −1% to −3% overall, with largest in EEU (−3.7%) and NA (−2%).
• Model yield impacts: CMIP6-driven LPJmL underestimates wave-7 impacts globally (multi-region average near 0%, about 3% lower than observations). For wave-5, the multi-region average matches observations but masks regional biases (e.g., +3% overestimation in NA offsetting underestimation elsewhere). Historic simulations particularly underestimate impacts in EAS (by ~9%) and EEU (by ~3%). Driving LPJmL with W5E5 reanalysis reduces discrepancies in NA, EEU, and IND but not uniformly (WEU, EAS remain biased), implicating climate-model surface anomaly biases as a key source of crop impact underestimation.
• Timing biases: Models misplace the intra-seasonal timing of wave events relative to observations (e.g., wave-5 skewed earlier in JJA in models), which may contribute to crop impact biases given crop calendars.
• Concurrent low yields (LMF): Wave events increase the likelihood of concurrent below-trend yields for several regional pairs in observations, especially those including NA; they tend to decrease the likelihood of concurrent positive yields. Reanalysis-driven crop simulations capture these features; historical CMIP6-driven simulations show fair agreement in sign, particularly for wave-5. For future projections, models increasingly suggest co-occurring positive yields for wave-7, with only limited pairs showing increased co-occurring negative yields (e.g., NA×EAS for wave-7; NA×EEU and IND×WEU for wave-5). However, inter-model disagreement in future LMF changes is high, limiting confidence.
• Overall: Climate models’ underestimation of surface anomalies during high-amplitude wave events propagates to underestimation of regional crop yield losses and deep uncertainty in projections of synchronized harvest failures.

The study demonstrates that high-amplitude Rossby wave patterns (wave-5 and wave-7) materially increase the risk of regional and concurrent crop yield losses in major Northern Hemisphere breadbasket regions. While climate models accurately reproduce upper-level wave structures, they substantially underestimate surface temperature and precipitation anomalies that directly affect crops, especially during wave events. This underestimation leads to muted crop yield impacts in linked climate–crop simulations compared to observations. Reanalysis-driven crop simulations align more closely with observed yield responses, underscoring the role of biased surface anomalies and the importance of correct sub-seasonal timing relative to crop phenology. The findings imply that current climate risk assessments may understate the probability and severity of synchronized harvest failures tied to meandering jet regimes. Although future projections show regional amplification of temperature anomalies (e.g., Western North America, Eurasia, East Asia), models disagree on changes in concurrent low-yield likelihood, reflecting deep uncertainty. These results highlight the need for improved model representation of extreme events, more effective bias adjustment that preserves inter-variable coherence during high-amplitude circulation regimes, and decision-making frameworks that account for model blind spots and low-probability, high-impact risks.

This work quantifies how meandering jet stream patterns (wave-5 and wave-7) elevate the likelihood of simultaneous low crop yields across key breadbasket regions and evaluates the fidelity of state-of-the-art climate and crop models in reproducing these impacts. The main contributions are: (1) empirical mapping of observed wave-related concurrence of low yields; (2) demonstration that climate models reproduce upper-level wave patterns but underestimate associated surface anomalies; (3) evidence that these biases propagate to underestimation of regional yield losses and to uncertainty in concurrent failures; and (4) initial future projections showing regional amplification of temperature anomalies without consistent increases in wave amplitudes, with mixed signals for concurrent low-yield risks. The study underscores high-impact blind spots in current climate risk assessments and calls for improved process-based modeling, bias-adjustment methods that preserve extreme-event structure, integration of physically constrained machine learning, and decision-centric approaches (e.g., storylines, expert elicitation) to better capture deeply uncertain, compounding hazards. Future research should examine interactions among uncertainties (e.g., CO2 fertilization, hydrologic changes) and refine representation of sub-seasonal timing and crop phenology in impact models.

• Climate model biases: Underestimation of surface temperature and precipitation anomalies during high-amplitude wave events in both raw and bias-adjusted CMIP6 outputs; bias adjustment may not target extreme-event subsamples.
• Crop model limitations: LPJmL uses static harvested area and may not capture interannual changes; regional variation in responsiveness to wave events persists even with reanalysis forcing; uncertainties in CO2 fertilization and hydrologic cycle changes affect projections.
• Event timing: Differences in intra-seasonal timing of wave events between models and observations can misalign with crop growth stages, affecting yield impact estimates.
• Future projections: High inter-model disagreement in changes to concurrent low-yield likelihood (LMF) under SSP5-8.5; no consistent increase in wave amplitudes, leading to deep uncertainty in future risk quantification.
• Data constraints: Reanalysis coverage and differences (ERA5 vs W5E5), and limited number of bias-adjusted CMIP6 models used for crop simulations may affect generalizability.