Near-term decadal climate predictions are valuable for decision-making in climate-sensitive sectors like agriculture. The agricultural sector is significantly impacted by climate variability and extremes, such as droughts and heatwaves. Climate risk management for adaptation and mitigation benefits greatly from decadal predictions, offering a temporal coherence with observations and allowing quality assessment through retrospective prediction evaluation. This is particularly advantageous compared to long-term climate projections. Wheat, a globally crucial crop, is highly susceptible to drought and heat stress, affecting both yield and quality at various growth stages. The EU H2020-MED-GOLD project highlighted the need for reliable decadal climate information for strategic decisions in the wheat sector, including global market risk estimates, infrastructure investments, supply chain planning, and breeding programs. This study aims to assess the skill and reliability of predicting the Standardized Precipitation Evapotranspiration Index (SPEI) and the Heat Magnitude Day Index (HMDI) on a multi-annual timescale to inform climate services for the wheat sector.

The introduction cites several studies highlighting the importance of decadal climate predictions for various sectors (e.g., agriculture, energy) and the existing limited research on their application to specific sectoral needs. The need for climate risk management in agriculture and the advantages of using decadal predictions over long-term projections are discussed. Previous research on the impacts of heat and water stress on wheat yield is reviewed, emphasizing the sensitivity of yield and quality to these stresses at different growth stages (flowering and grain filling). The study is placed within the context of the EU H2020-MED-GOLD project, which focused on co-designing climate services based on user needs in the durum wheat sector. Advances in decadal climate prediction are noted, emphasizing the remaining need to translate predictions into usable information for specific sectors like wheat production.

The study uses decadal hindcasts from the Community Earth System Model Decadal Prediction Large Ensemble (CESM-DPLE), which includes 40 ensemble members at 1-degree resolution. The analysis focuses on the first five forecast years. Two agro-climatic indices were used: SPEI6 (standardized precipitation evapotranspiration index accumulated over six months before harvest) and HMDI3 (heat magnitude day index over three months before harvest). The SPEI6 calculation involves estimating potential evapotranspiration (PET) using the Thornthwaite approach, with sensitivity tests conducted using Hargreaves and modified Hargreaves methods. HMDI3 is calculated based on daily maximum temperatures exceeding the 90th percentile. The study compared adjusted and unadjusted indices. Calibration employed a variance inflation technique to adjust the interannual variance of the forecasts to match observations. Forecast skill was assessed using Fair Ranked Probability Skill Score (FRPSS) and Fair Continuous Ranked Probability Skill Score (FCRPSS). Reliability was assessed using reliability diagrams. The impact of initialization was evaluated by comparing decadal hindcasts to uninitialized historical simulations. The analysis was performed globally and regionally, focusing on key wheat-producing regions. The crop calendar data from MIRCA2000 was used to define wheat harvesting periods.

The assessment of drought index (SPEI6) showed that calibrated forecasts are more skillful and reliable than unadjusted forecasts, particularly for below-normal and above-normal categories. The skill in predicting SPEI6 is linked to the skill in predicting individual climate variables (precipitation and PET) and the ability of the forecast system to capture their influence on SPEI6. Regionally, reliability varied, with the Asian region showing reliable predictions for all categories, unlike other regions where reliability was lower for the 'normal' category. For the heat stress index (HMDI3), the calibrated forecasts also showed improved skill and reliability, especially in areas with initially low skill. Similar to SPEI6, the skill in predicting HMDI3 is associated with the forecast system's ability to predict maximum temperature. Comparing initialized decadal forecasts with uninitialized historical simulations revealed that initialization significantly enhances the skill and reliability of predictions for both indices in many key wheat-producing areas. An example climate service product was developed presenting forecasts of the most likely tercile category for SPEI6 and HMDI3, demonstrating the practical application of the findings to specific locations in Italy.

The results demonstrate the value of calibrated decadal climate forecasts for providing reliable information on drought and heat stress in wheat-producing regions. The calibration step proves crucial for generating trustworthy climate services. Initialized forecasts consistently outperform forecasts based on climatology, improving risk reduction and adaptation strategies. These results suggest significant opportunities to support stakeholders in decision-making across various timescales. The methodology is adaptable to other crops and sectors. While promising, the study is based on a single decadal forecast system and future multi-model studies with larger ensembles are needed. Further research will explore the influence of natural climate oscillations on predictability. Future efforts should focus on co-designing climate services to optimize product refinement, visualization, and dissemination.

This study shows that calibrated decadal climate forecasts are skillful and reliable in predicting multi-annual drought and heat stress indices for several wheat-harvesting regions. This makes them valuable for informing decisions by farmers, breeders, and policymakers. Future research will expand the analysis to include multiple models and investigate the role of natural climate oscillations. Co-designing climate services with stakeholders will further improve the uptake of decadal climate information.

The study's analysis is based on a single decadal forecast system (CESM-DPLE). While the study included sensitivity tests for PET estimation and addressed ensemble size issues using fair skill scores, a multi-model assessment would strengthen the results. The impact of slow, internally generated natural climate oscillations was not explicitly analyzed. The reliability of the 'normal' category is generally lower compared to other tercile categories. Finally, while the presented climate service product provides a useful example, thorough stakeholder engagement is still required to tailor products to specific needs and ensure effective communication.