Assessing spatial biases of a meteorological model for real-time flood forecasts in the hydraulic node of Milan

Meteorological forecasts are crucial for mitigating extreme weather impacts, but they always contain sources of uncertainty. One of these arises in the prediction of the location of intense convective precipitation systems: this issue is particularly critical for flood forecasting in small watersheds, where even a slight discrepancy in the predicted rainfall location can lead to significant inaccuracies in flow forecasts.

In this study, we propose a methodology for quantifying the spatial biases of rainfall forecasts produced by the MOLOCH meteorological model on the hydraulic node of Milan (northern Italy), which is a strongly urbanized and sensible territory due to the presence of productive activities, critical infrastructures and more than 4 million inhabitants. The goal is to investigate whether the model exhibits “preferential” directions where it tends to misplace convective precipitation events.

A total of 65 significative convective rainfall episodes were selected during the 2013-2022 period, comparing the rainfall forecasts at day+0 with the observed rainfall field obtained through a merging of radar and rain-gauge stations (“PRISMA” dataset, provided by Lombardy Region’s Civil Protection).

The proposed procedure relies on the definition of a domain around the study area which is suitable to define the eastwards/northwards spatial misplacements of the forecast rainfall field, such domain was defined accordingly to an analysis on the Fractional Skill Score (FSS).

Results indicate that the MOLOCH meteorological model tends to misplace convective rainfall systems towards the North-West and North-East directions, and this outcome could be significant to forecasters operating in the civil protection sector, especially for these river catchments with limited spatial extent.