Yield reduction under climate warming varies among wheat cultivars in South Africa

The study investigates how extreme heat affects dryland wheat yields in South Africa and whether impacts vary among cultivars. Wheat is a key supplementary staple in South Africa and the broader Southern African region, with production vulnerable to increasing temperature extremes and climate variability. Prior work often relies on controlled experiments, small plots, or crop models, with a paucity of long-term, multi-location, in-field empirical analyses in Southern Africa. The purpose is to quantify yield responses to heat using daily weather and extensive field trial data, estimate yield changes under uniform warming scenarios (+1 to +3 °C), and assess heterogeneity across cultivars to inform adaptation strategies, breeding priorities, and food security policy.

Previous research has documented negative effects of high temperatures on wheat and other staples, largely outside Africa or via simulations. Southern Africa faces projected increases in temperature (often 3–4 °C by mid-to-late century) and uncertain precipitation changes, with recent droughts highlighting food security sensitivity to weather shocks. South Africa’s wheat sector features significant public and private breeding efforts, yet empirical, multi-year, multi-location open-field studies are scarce. Literature suggests potential annual yield declines under drier conditions and rising irrigation demand, while regional wheat market dynamics link South African production to broader Southern African prices and food security. Existing meta-analyses project notable wheat yield declines in Southern Africa by 2050, underscoring the need for empirical evidence from in situ field trials and attention to cultivar-level responses.

Data: 18,881 dryland wheat field trial observations from ARC-SGI across 17 locations and 71 cultivars (1998–2014). Trials were matched to nearby weather stations (within 75 km) with daily max/min temperatures and precipitation (NASA GSOD via GSODR). Planting and flowering dates are observed; harvest dates inferred as 30 days after flowering (validated with robustness checks and alternative maturity lengths). Temperature exposure: Daily max/min temperatures were interpolated sinusoidally to estimate intra-day temperature duration within 5 °C bins, aggregated over each season. Eight bins were used, with all temperatures above 30 °C grouped in a top bin. Seasonal cumulative precipitation entered as a quadratic. Statistical model: Preferred specification regresses log yield on cultivar, location, and year fixed effects; quadratic cumulative precipitation; and the eight temperature exposure bins. Standard errors are clustered by province-year to address spatial correlation. Moran’s I analyses show the regression substantially reduces spatial autocorrelation in residuals. Warming impacts are simulated by uniformly shifting observed daily temperatures by +1, +2, and +3 °C and recomputing exposure bins; yield impacts are calculated via the estimated coefficients using the Delta Method. Cultivar heterogeneity: A multilevel model introduces a random slope for the >30 °C exposure bin across cultivars to estimate cultivar-specific heat effects. Relationships between cultivar release year, mean yield, and heat sensitivity are examined. Robustness checks: Alternatives include using seasonal average temperature quadratics (which underperform and mask heat-threshold damages), varying bin widths/thresholds, adding pre-season precipitation, cubic precipitation polynomials, low-precipitation indicators and interactions with heat, stage-specific (vegetative/flowering/grain-filling) weather effects, breeder and growth-habit interactions, restricting to cultivar-years with >30 °C exposure, expanding weather stations with distance-weighted interpolation, substituting CHIRPS precipitation, optimizing maturity length, and simulating earlier planting dates. Results are qualitatively consistent across checks.

- Extreme heat strongly reduces yields: an additional 24 hours of exposure above 30 °C is associated with a 12.53% decrease in wheat yield (t(30)=40.26, p=0.000). - Uniform warming scenarios yield nonlinear, increasingly negative impacts: +1 °C reduces average yield by 8.5% (Delta Method=-3.21, p=0.001); +2 °C by 18.4% (Delta Method=-3.68, p=0.000); +3 °C by 28.5% (Delta Method=-4.16, p=0.000), averaging roughly 9% loss per °C. - Heat vs. precipitation: Interaction between extreme heat (>30 °C) and low precipitation (below 10th percentile) is positive but not statistically significant (t(30)=0.79, p=0.38); warming estimates are similar with or without the interaction. - Precipitation effects: A one standard deviation decrease in seasonal rainfall is associated with a 9.6% yield reduction; very low rainfall (10th percentile) corresponds to an approximate 18% reduction. - Cultivar heterogeneity and trends: Newer cultivars have higher mean yields (~0.7% annualized relative gain) but larger negative heat effects. The ratio of heat effect to mean yield has trended less negative over time, though not statistically significant (p>0.1). - Adaptation potential via cultivar choice: Switching from a cultivar with the largest heat effect to one with the smallest reduces warming impacts by about 5.5, 11.5, and 17.6 percentage points under +1, +2, and +3 °C, respectively. - Breeder and growth-habit differences: Breeder-specific heat effects differ statistically but magnitudes are small; spring/facultative/winter types do not differ significantly in heat effects. - Planting date adaptation: Shifting planting 14 days earlier reduces warming impacts modestly (~1% at +1 °C; ~4% at +3 °C). - Robustness checks consistently support threshold-driven heat damages and simulated warming impacts; models using seasonal average temperatures fail to capture observed heat-threshold yield losses.

Findings demonstrate that short exposures above 30 °C drive substantial declines in dryland wheat yields, and projected uniform warming leads to progressively larger losses. While moderate warmth (25–29 °C) can be beneficial, exposures beyond the 30 °C threshold dominate, producing net negative impacts. The results address the research question by quantifying the magnitude of heat damages using in-field multi-year, multi-location evidence in South Africa and by showing significant heterogeneity among cultivars. This heterogeneity offers a pathway for adaptation through targeted breeding and optimal cultivar selection. However, a tradeoff emerges: more recently released cultivars, while higher yielding on average, tend to exhibit greater sensitivity to extreme heat, likely reflecting historical breeding priorities (e.g., yield and quality) over heat tolerance. The study’s precipitation analyses indicate that heat damages are not solely artifacts of drought conditions, and policy or breeding strategies that incorporate heat tolerance traits could mitigate warming impacts. The results complement climate model projections and can inform crop modeling, breeding strategies, and policy for enhancing resilience and food security in Southern Africa.

This study provides empirical, field-based evidence that extreme heat (>30 °C) substantially reduces wheat yields in South Africa and that warming impacts increase nonlinearly with temperature rise. It quantifies yield penalties under +1 to +3 °C scenarios and reveals significant cultivar-level variation in heat sensitivity, indicating scope for adaptation through breeding and cultivar selection. Breeding programs should prioritize integrating heat tolerance traits into newer, high-yielding cultivars, alongside exploring agronomic adaptations (e.g., adjusted planting dates) and potential irrigation strategies. Future research should incorporate CO2 effects, extend empirical datasets across Southern Africa, examine irrigation offsets and management adaptations, and integrate statistical findings with process-based crop models to evaluate climate-resilient pathways.

- Soil moisture is not directly observed; precipitation measures and interactions serve as proxies, with the heat-by-low-precipitation interaction not statistically significant. - CO2 fertilization effects are omitted, limiting inference on long-term climate change tradeoffs between biomass and yield. - Weather station data may not perfectly match field conditions, especially for precipitation; robustness checks with expanded stations and CHIRPS precipitation indicate similar results. - Harvest date is inferred (30 days post-flowering); alternative maturity lengths and optimized maturity analyses yield similar conclusions. - Spatial correlation exists; clustering by province-year and Moran’s I analyses indicate residual spatial dependence is reduced but not eliminated. - Results are specific to South African cultivars, locations, and dryland systems; representativeness for other Southern African regions may vary. - Some cultivars had limited exposure above 30 °C in certain locations; robustness checks restricting to exposed cultivar-years confirm core findings.