Mapping microclimatic spring frost risk through high-resolution spatial modeling in STYRIAN fruit- and winegrowing regions

In light of climate change, spring frost damage may persist or increase despite fewer frost days because earlier budbreak increases crop exposure. The targeted selection of sites for fruit and vine production is therefore a crucial element of climate change adaptation strategies. Existing temperature grid datasets for Austria, however, lack the spatial resolution to provide meaningful information for this purpose. To support decision makers and growers in implementing adaptation measures, we developed a high-resolution spring frost risk map for the fruit- and winegrowing regions of Styria. This map is based on a spring frost risk index calculated from weather station time series data, which was then spatially distributed using a high-resolution digital elevation model and derived topographic predictors. These predictors serve as proxies for cold-air drainage under radiation-frost conditions with limited atmospheric mixing. Calibration used observations from 151 WegenerNet stations covering diverse topographic settings. We investigated nine spatial-statistical and machine-learning configurations and evaluated them using repeated fivefold cross-validation with ten repetitions. The best performance was achieved with regression kriging combined with a heterogeneous stacking regressor and RidgeCV meta-learner (R 2 = 0.71, RMSE = 0.46). Previous studies primarily mapped minimum temperature or crop-specific frost thresholds using regression kriging, individual machine-learning models, or single-learner residual-kriging hybrids. In contrast, our framework combines a multi-threshold radiation-frost exposure index, a heterogeneous stacking regressor, and kriging of the remaining spatial residuals in one 50-m climate-service workflow. The resulting map provides locally resolved information for screening potential sites for fruit and vine production. Practical implications The resulting frost risk map provides detailed spatial information that can support climate-adaptation measures in Styrian fruit production. Practical applications from a long-term perspective may include deliberate site, crop, and variety selection and long-term planning under future warming scenarios. From a short-term perspective, information on spring frost exposure at a specific site may support fruit- and winegrowers in assessing the economic viability of technical measures against damage by radiation frost. Although the FRI framework was developed for long-term spatial risk zoning rather than real-time forecasting, it could support operational climate services. Areas of high modeled frost exposure could inform the placement of temperature sensors and the spatial targeting of early-warning systems. Combining this static 50-m risk baseline with real-time observations, short-range weather forecasts, and crop- and phenology-specific thresholds could support targeted frost mitigation but would require prospective operational validation.