Regionalisation, Bias Correction and Downscaling of ERA5-Land Snow Variables by Means of Local Observations Recorded in Central Italy

The snow depth and the increase of snow depth after three consecutive days of snowfall, hereinafter referred to as d_s and DH3gg, respectively, are typically chosen for avalanche protection and avalanche hazard assessment purposes. With specific reference to the Central Apennines (Central Italy), the preferable provider of observations for avalanche related applications is MeteoMont, which supplies d_s observations at 34 manual stations, measured between 1978 and 2023. The area of interest is also covered by ERA5-Land, over a period of 73 years, from 1950 to 2023. In terms of temporal, spatial and quantitative availability of snow information, ERA5-Land consists in a more appealing choice as most manual weather stations set up in the Central Apennines are located at lower altitudes compared to where avalanches are likely to occur. Moreover, data recorded at manual stations appears to be incomplete, especially during extreme snowfall events. However, it is necessary to stress that ERA5-Land is affected by biases (e.g. underestimation or overestimation of extremes) and the use of uncorrected data in all applications might lead to unreasonable results. Therefore, in order to overcome the listed limitations, the suggested approach consists in the regionalisation of both ERA5-Land and MeteoMont d_s and DH3gg and in the subsequent bias correction and downscaling of the regionalised ERA5-Land variables by means of the regionalised MeteoMont ones. With regards to ERA5-Land, 51 nodes have been considered as their grids intersect recorded and reconstructed avalanche paths in the Abruzzo Region (extracted from the Avalanche Record and the Map of Probabilistic Location of Avalanches provided by the Abruzzo Region). This ensures that the selected nodes are solely representative of areas where avalanches are most likely to occur. The regionalisation of both ERA5-Land and MeteoMont d_s and DH3gg is performed by applying the index value regional method before the bias correction and the downscaling of ERA5-Land data as, in terms of computational efforts, only 2 bias corrections and downscalings for each couple of best-matched ERA5-Land and MeteoMont homogeneous areas would be required instead of 102 (2 for each couple of nodes and stations). The bias correction and downscaling of the ERA5-Land regionalised variables are then performed by means of a statistical transformation based on the assumption that said variables are described by one of the distributions belonging to the GEV family. This work is of particular relevance as, on the one hand, it overcomes the limited availability of snow information in the Central Apennines, especially in relation to avalanche related applications. In fact, it provides a tool that quantifies d_s and DH3gg quantiles at elevations and sites that are not supplied with observations. On the other hand, it provides realistic initial and boundary conditions for simulating avalanche dynamics, drawing up hazard and risk maps, and designing active and/or passive defence structures.