Design and validation of a risk assessment model for vegetation frost damage

Impact models for various high-risk weather phenomena can be run as downstream models using output of a numerical weather prediction (NWP) model. Using specialised algorithms, the impact models can further refine risk level estimates of potentially dangerous weather phenomena, of which frost events are a prime example. These events, occurring in Europe mostly in April and May, can cause significant damage in the agricultural sector. The vegetation is often activated during a transient period of relatively high temperatures in early spring and then can be heavily damaged during subsequent periods of cold air advection. The impact can be further enhanced by radiative cooling and cold-air pooling at night-time. The damage can be reduced by a multitude of measures, provided a reliable forecast with sufficient spatial resolution is available. Here, we present a new frost events impact model. The impact model estimates real-time development phase of several crops using up-to-date air temperature sums. These phases can be linked to specific temperature thresholds for frost damage. The thresholds are then compared with a forecast of minimum air temperature from a hectometric-scale configuration of an NWP model. This workflow was tested during two frost events – a pronounced event in April 2024 in Spain and a recent April 2025 event in the Czech Republic. The performance of three NWP models was evaluated using minimum air temperature measurements by the Czech Hydrometeorological Institute and the Spanish meteorological service AEMET. While the models agreed well with observations at some cases, elevated bias was found at some regions. Using Meteosat Third Generation imagery, it was revealed that the bias can be partly attributed to overestimated low-level cloud cover. In addition, despite very-high horizontal resolution of the NWP models, some small-scale temperature variability seems to remain unresolved and can also contribute to model inaccuracies. To minimize the bias, the best performing models were identified and are suggested for operational use. The newly developed method can also be used to evaluate historical climatological data to determine areas with elevated frost risk for agriculture.