Climate variability and perennial fruit crop yields: insights from Trentino-Alto Adige, Northern Italy

• Climate-yield relationships assessed for apples and grapes in alpine orchards. • Linear regression and interpretable machine learning approach applied. • Apple yield shows more consistent sensitivity, with frost as main constraint. • More complex and buffered climate-yield relationships for grapes. Effective adaptation planning for perennial crops, particularly in mountain contexts with climate-sensitive agroecosystems, requires a robust understanding of yield responses to climatic variability. Based on long-term data and a methodological framework covering both linear regression and machine learning techniques, the present study investigates the influence of interannual variability in agro-climatic indices on apple and grape yields in Trentino-Alto Adige, an alpine region in northern Italy. Results reveal that apple yields are more consistently influenced by climate variability than grape yields, with frost occurrence and heat-related indices emerging as key predictors. The machine learning approach, through variable importance metrics and individual conditional expectation plots, provides insights into nonlinear yield responses to critical climatic thresholds, such as sharp declines beyond a certain number of frost days or plateauing gains under sustained heat accumulation. Conversely, grape yields exhibit more heterogeneous and buffered responses, reflecting more complex interactions with climatic conditions. Overall, the study highlights the added value of data-driven approaches with physical interpretability for capturing intricate climate–yield relationships. In regions increasingly exposed to climate pressures, such as the alpine valleys, these tools can support the development of targeted strategies to sustain long-term crop productivity.