Figure 2. a) Slopes of the regional trends at different durations for observed and modelled AM; significant trends (α-level=0.05) are marked; stochastic uncertainty associated with the modelled AM (90% C.I. of the MonteCarlo simulation) is also reported. b) Slopes of the regional trends for the model parameters: scale parameter (λ), shape parameter (κ), and yearly number of storms (n); significant trends (α-level=0.05) are marked.c) Differential impact on the modelled trends of combinations of changes and no-changes in the model parameters; series labels report the parameters which are allowed to change. d) Slopes of the regional trends for some estimated return levels (2, 10, 25, 50 yr); significant trends (α-level=0.05) are marked; note that the 2 yr return levels correspond to the modelled AM.
3.4 Changes in the proportion of convective-like events
The parametrization of our model allows us to formulate hypotheses about the physical processes underlying the detected changes. In particular, the observed changes could be explained by an increased number of intense convective events, which would mainly contribute to the short duration annual maxima. We analyze possible changes in the number of storms occurring in different seasons, and in the seasonal number of convective-like and other types of storms (Figure 3 ). The positive trend in the yearly number of storms reported above is fully explained by the increases in the number of storms in autumn (SON,Figure 3 a) and in winter (DJF). However, examining changes in the types composition shows no distinct increase in convective-like storms during these seasons (Figure 3 b, c).
Conversely, although no trend emerges in the number of storms in summer (JJA), the number of summer convective-like storms in this season increased significantly, while the number of other storms shows no trend (Figure 3 b, c). This implies a significant increase in the proportion of summer convective-like events. Since convective-like storms are generally associated with heavy intensities at short durations, this change in composition could explain the observed increase in tail heaviness at short durations, and thus the observed trends on short-duration AM. This is confirmed when the parameters of the ordinary events distribution are examined considering spring-summer (MAMJJA) and autumn-winter (SONDJF) separately (Figure S2). These results suggest that the significant positive trends found for short-duration extremes are mostly related to changes in summer storms, and that these can be related to changes in the intensity distributions (increasing tail-heaviness) induced by an increasing proportion of heavy convective-like storms in the summer.