Orographic influence on the spatial variability of rainfall extremes in Italy

Italy is a mountainous country with peculiar exposure to severe rainstorms. With the aim of improving the rainstorm hazard assessment over ungauged areas, the spatial variability of the relationship between rainfall and elevation is investigated over Italy, using local linear regression analysis. The analysis focuses on annual precipitation extremes in 1 and 24 hours. Data are taken from the recently released Improved Italian – Rainfall Extreme Dataset (I2-RED), a collection of rainfall measurements acquired from 1916 until now by 5265 rain gauges. In this analysis, only the series with at least 10 years of data are considered, for a total of more than 3700 time series. Starting from a knowledge base of general multivariate modeling, we have addressed local simple linear regression models between elevation and the average of extremes for any of the 1-km size pixels that cover Italy. The analysis is carried out by selecting, for each pixel, rain gauges available within a radius ranging from 1 to 50 km, with a progressive increase of 1 km. Regression constraints are set so to include only cases with at least 5 rain gauges and > 100 m of difference in elevation. Statistical significance is set at the 5% level. To avoid excessive extrapolation effects in high elevations (i.e., where snow is often recorded in place of rainfall), only the cases in which a tolerance of +/- 100 m between the estimation pixel elevation and that of the considered rain gauge stations are retained. Further constraints, as a higher minimum search radius or a constraint on the persistence of the significance of the regression as function of the radius, have been also tested. The results obtained for the 1-hour duration highlight a different sign of the “orographic effect” between rainfall and elevation in different areas: an inverse effect is noticed over the Alps, Liguria region and central Apennines, where rainfall decreases with elevation. For 24-hours rainfall the spatial distribution of the trend is different: for the most of the Italian territory the 24-hours extremes increase with elevation. The advantages of the high spatial detail granted by the proposed approach are finally demonstrated by the error assessment analysis, showing low absolute bias and very limited error clustering effects.