Climate service driven adaptation may alleviate the impacts of climate change in agriculture

Agricultural productivity is highly vulnerable to climate change, with European assessments indicating potential yield gains in the north and losses in the Mediterranean, and substantial risks from variability and extremes. There is an urgent need to design and test effective adaptation strategies. While advances in ensembles of process-based crop models have improved impact assessments, uncertainties and the difficulty of quantifying the added value of climate services persist. Through stakeholder engagement in the EU-H2020 MedGOLD project, varietal selection at sowing emerged as a key decision point for climate service support. This study investigates an idealised, tailored agro-climate service that provides annual advice on optimal durum wheat cultivar choice at sowing for the Euro-Mediterranean region. Using climate projections for 2021–2040 under RCP8.5 and crop modelling, the research quantifies the benefits of service-driven adaptation on mean yield and interannual yield variability (and resilience), while intentionally holding other management practices constant to avoid confounding effects. The idealised service is represented by a specified prediction skill and applied to simulated future yield outcomes to evaluate potential benefits and trade-offs.

Prior work documents significant climate-change impacts on European crops, including wheat, with north–south contrasts and increased risks from droughts and heat waves. Early calls for agricultural adaptation highlighted knowledge gaps, while more recent literature shows progress in impact modelling and adaptation assessment but cautions about model limitations and uncertainty. The evaluation of climate services remains challenging due to limited integrated frameworks and computational hurdles, though emerging initiatives (e.g., Earth’s Digital Twin) promise advances. For durum wheat, historical breeding trends increased harvest index and grain number while reducing plant height, with slower progress since 1980. Studies have examined climate impacts on wheat, water scarcity risks, and temperature effects on global production. The role of seasonal climate forecasts for agriculture and the potential of climate services to support decisions have been explored, yet comprehensive, quantitative assessments of their added value for specific on-farm choices like cultivar selection are limited. Variety mixtures have been proposed to buffer climate risks, and resilience metrics have been developed to integrate mean performance and variability.

- Climate data: Five EURO-CORDEX regional climate models (0.11°) under RCP8.5 from 2006 onward were used. Daily Tmax, Tmin, and precipitation were bias-adjusted via quantile mapping; global solar radiation was evaluated model-by-model. - Crop model: ECroPS (European Commission Joint Research Centre), based on WOFOST, simulates water limitations, heat stress at flowering and grain filling, and includes CO2 fertilisation effects. Nutrient limitations and pests/diseases were not modelled. Model inputs (soils, parameters, crop calendars) followed the EU-JRC MARS system. - Ideotypes: To represent current genetic diversity, 191 durum wheat accessions from 9 field experiments in 4 Mediterranean countries were analysed. Three families of heading thermal requirements were identified, from which 18 ideotypes were sampled (random without replacement) capturing multimodal distributions. Thermal requirements to flowering and maturity were used to drive ECroPS. The shortest-cycle ideotype reaches flowering at 1040 degree days (base 0 °C) and the longest at 1507 degree days. Sowing dates were fixed per grid, taken from the MARS system. - Simulation experiments: Baseline period 1986–2005 and future period 2021–2040 were simulated for all ideotypes and climate models across the Euro-Mediterranean region. Other management (e.g., sowing date changes, irrigation, fertilisation) was held constant to isolate varietal effects. - Statistical assessment: Differences in mean yield responses (future vs baseline) were tested using a two-sample Anderson–Darling test. Resilience index defined as μ²/σ², where μ and σ are mean and standard deviation of simulated yield. - Idealised climate service: Implemented in post-processing using a Bernoulli hit-rate approach. For each grid and year in 2021–2040, with probability p (service prediction skill or hit rate), the best-performing ideotype (highest simulated yield among 18) is selected; otherwise, a non-optimal selection occurs. Ensembles integrating the service were generated across all RCMs and baseline runs. Skills explored ranged from 10% to 70%. Assumptions include no spatial/temporal dependency (e.g., peer influence), universal annual access to a representative pool of varieties, and no contractual restrictions. - Data and code: Bias-adjusted RCMs available via EC-JRC Data Catalogue; ECroPS model open-source on GitHub; simulated wheat data available via EC-JRC Agri4cast portal or by request.

- Without climate-service-driven adaptation, mean durum wheat yield declines in 2021–2040 relative to 1986–2005 are estimated between −7.8% and −5.8%, despite CO2 fertilisation. Interannual yield variability increases by 7% to 12%. - Ideotype effects: Shorter-cycle ideotypes are less impacted, with more homogeneous spatial responses and unimodal distributions of yield changes. Longer-cycle ideotypes show larger mean losses and more heterogeneous, often tri-modal spatial distributions, reflecting greater exposure to late-season heat and drought. - Regional heterogeneity: Optimal ideotypes to minimise mean yield losses vary by region and climate model ensemble (e.g., longer cycles may be optimal in parts of Ukraine and northern Italy, while shorter-to-average cycles in southern Italy), indicating precipitation sensitivity and the need for local solutions. - Variability vs mean trade-off: Ideotypes optimal for mean yield are not necessarily optimal for minimising interannual variability. Using a resilience index (μ²/σ²) improves model coherence but still shows regional differences (e.g., longest cycles often resilience-optimal in much of Ukraine; mixed signals in France). - Idealised climate service benefits: Integrating the service improves mean yield outcomes beyond any single ideotype strategy. At 40% prediction skill, mean yield losses are offset. At 70% skill, mean yield gains reach about +5.3% versus baseline. - Variability impacts of service: For skills below 50%, interannual variability increases by roughly 25% even as mean yields improve. At higher skills, variability increases are much smaller while maintaining yield gains. - Resilience outcomes: Only services with prediction skill ≥70% achieve a favourable overall trade-off in resilience, combining mean yield gains with acceptable variability. - Variety mixtures implication: Positive effects at lower skills suggest that rotating or mixing varieties over time can mitigate climate impacts.

The study demonstrates that dynamic adaptation via climate service-informed cultivar selection can substantially reduce, and even reverse, projected mean yield losses for durum wheat in the Euro-Mediterranean by 2021–2040. However, improvements in mean yield may come with increased interannual variability, potentially exacerbating market volatility. This creates a policy and management trade-off: services with moderate skill can deliver mean gains but may necessitate stabilisation mechanisms to buffer increased variability. Services with high skill (around or above 70%) provide both yield gains and acceptable variability, aligning with end-user expectations identified in co-design processes. The heterogeneity in optimal ideotypes across regions underscores the importance of localised, flexible strategies and supports the potential of temporal variety mixtures to manage risk. The findings advocate for expanding modelling frameworks to larger, multi-model ensembles and multi-trait ideotype spaces to better capture complexities and inform robust, scalable climate services.

Targeted, climate service-driven cultivar selection at sowing can meaningfully alleviate negative climate change impacts on durum wheat, turning projected losses into gains under near-future conditions when prediction skill is sufficient. While lower-skill services still provide benefits in mean yield, they may increase interannual variability, highlighting the need for complementary risk management (e.g., market stabilisation, insurance) or improved forecast skill. Future research should: expand to very large climate–crop ensembles; incorporate multiple plant traits and their interactions; consider broader management adaptations; and employ integrated farm system models to evaluate climate services across the full range of farm decisions. Successful adaptation will require dynamic, iterative strategies co-developed with end-users, with continuous monitoring and refinement over time.

- Single-trait ideotype construction focuses on thermal requirements; other key traits and trait combinations were not explored. - Only five EURO-CORDEX RCMs under RCP8.5 were used; precipitation variability and model spread contribute to regional uncertainties. - Management practices (sowing dates, irrigation, fertilisation) were held constant; nutrient limitations and pest/disease pressures were excluded. - The idealised climate service assumes a simplified Bernoulli hit-rate selection with no spatial/temporal dependence, universal annual access to all varieties, and no contractual constraints. - Fixed sowing dates per grid may not capture adaptive sowing strategies. - Results hinge on the representation of CO2 fertilisation and model parameterisations; further sensitivity analyses and calibration could refine outcomes. - Potential market impacts of increased interannual variability are not modelled; implications are discussed qualitatively only.