5.1 Model validation and scaling factor estimate
The results of model validation indicated a precipitation overestimation by the models compared to the ground data, as reported in Table 2. Marked difference between model outputs and ground data are observed especially for the extreme event class. The pairs EC-EARTH-RACMO22E and MPI-ESM -CCLM4 provided the worse comparisons, with the highest Root Mean Square Error (RMSE) in the extreme event class (59.56 mm for the former and 56.11 mm for the latter), while the other GMC/RMC pairs gave RMSE that are no larger than 25.59 mm. In Table 2, the GCM/RCM pairs that are more biased in northern Italy are highlighted in grey. Table 3 reports the maximum and minimum temperature validation. In all cases, there was a clear bias of the model outputs to higher temperature values. This phenomenon is noticeable in the extreme cold and extreme warm classes, where the models significantly overestimated the temperature with RMSE values that range between 2.00 °C and 2.48°C. The analysis of temperature differences did not identify GMC/RCM pairs that are more biased than others.
Figure 2a shows the spatial distribution of the scaling factors between ground data and GCM/RCM pairs for maximum and minimum temperature and for precipitation (further details on the performance of each model pair are given in Annex 1). For precipitation, values close to 0 indicate overestimation by the climate models, values close to 1 indicate agreement between simulations and ground data, and values larger than one represent underestimation by the GCM/RCM simulations. The precipitation scaling factors reported in Figure 2a indicate a good agreement between simulations and ground data in the Po Plain and for the lowest parts of the Alpine chain. The standard deviations calculated for the ground data and for the models are also comparable in this area (Figure 2b), and the scaling factors are no larger than the 0.3 times the GCM/RCMs standard deviation (Figure 2c). However, for higher elevations in the Alps (Julian and Carnic Alps and western Alps) scaling factors close to zero are detected. For the Ligurian region and the southern portion of the Po Plain, on the other hand, scaling factors larger than 1 are present.
For temperature, values close to 0 °C indicate agreement between simulations and ground data, negative values indicate overestimation by the GCM/RCM simulations while positive values imply underestimation by the GCM/RCM simulations. In figure 2a the spatial distribution of the scaling factors shows that the Po Valley and the foothills have the best agreement between simulated and measured temperatures. For maximum and minimum temperature, the scaling factors are between -2°C and +2°C. Figure 2b also shows that the standard deviations calculated on the simulated and ground data are comparable in this area; the scaling factors are up to 5 times the GCM/RCMs standard deviation (Figure 2c). Model overestimation is observed for the southern part of the study area with scaling factors for maximum temperature of about -2°C and -4°C. In a few areas along the western Ligurian coast, scaling factors between -4°C and -6°C were detected. The underestimation is mainly located in the eastern Alps (Julian and Carnic Alps) and western Alps, with scaling factors from about 4°C to 6°C for maximum and minimum temperature.
5.2 Precipitation and temperature anomalies
Figures 3 shows the ensemble average for the near (2021-2050) and far (2071-2100) future mean annual temperature and precipitation anomalies in northern Italy for the RCP 4.5 and 8.5 scenarios (further details on the performance of each model pair are provided in Annexes 2,3 and 4). An increase of minimum and maximum temperature is evident for the whole study area, with a more complex behaviour of precipitation anomalies.
For the near future, both scenarios indicate that the Alpine chain will be significantly affected by global warming, with annual maximum temperature anomalies between 3°C and 4°C and annual minimum temperature anomalies from 2°C to 3 °C (figure 3a). In the same period, the Po Valley and coastal areas will experience no more than 2°C of temperature increase. For precipitation, figure 3c shows that for RCP 4.5 most of the study area is expected to be interested by very small precipitation changes (anomalies close to 0 mm), while the Ligurian coast and the eastern Alps display positive precipitation anomalies of about 40 mm. A stronger precipitation increase is instead obtained for RCP 8.5 in the western and Julian Alps (anomalies between 80 mm and 120 mm).
For the last 30 years of the XXI century, the modelled temperature anomalies display the same spatial behaviour observed for the near future (Figure 3b) with annual maximum temperature anomalies between 4°C and 6°C and annual minimum temperature anomalies between 3°C to 4°C in the Alps. In the Po Plain, and especially along the Ligurian coast, the temperature anomalies will not exceed 3°C. For precipitation, the eastern portion of the Alps (Carnic and Julian Alps) is expected to experience a precipitation increase of about 40 mm, while the western Alps will be affected by a decrease of precipitation (anomalies of -40 mm). The RCP 8.5 scenario shows an intensification of this tendency, with negative annual precipitation anomalies in the entire Alpine chain, and anomalies of -80 mm in the western Alps (Figure 3d). Overall, these results indicate an unchanged, or slightly increasing, precipitation in the Alps for the near future, followed by a significant precipitation decrease at the end of the century. The RCP 4.5 and RCP 8.5 scenarios provide qualitatively analogous results, with the RCP 8.5 case showing enhanced changes.
5.3 Drought estimates
Figure 4 shows the statistics of extreme drought episodes for the baseline (1971-2000), near (2021-2050) and far (2071-2100) future 30-yr periods as identified by SPEI and SPI at the 12-month time scale, for RCP 4.5 and RCP 8.5. The plain-coloured histograms represent the SPEI results, while the line pattern marks the SPI outcomes. For each combination of RCM and GMC, the figure reports the total number, the mean percentage of affected area, and the mean duration of the drought events for the considered 30-yr period. Overall, figure 4 indicates that the SPEI and SPI detect the same number of drought events, with somehow different results in the percentage of affected area and drought duration. The drought analysis based on SPEI indicates that for the baseline period (1971-2000), 3 to 5 extreme drought events are identified. In the period 2021-2050, an increase of extreme drought events is present for the RCP 8.5 scenario, with 5 to 8 extreme drought events (Figure 4b). For the far future (2071-2100), the number of severe drought episodes is expected to further increase for the majority of the reliable model pairs and for both scenarios.
The percentage of area affected by droughts is reported in figure 4, showing that in the 1971-2000 period the maximum percentage of area involved by an extreme drought event ranged between 32% and 47%. For the concentration pathway RCP 4.5 in the 2021-2050 period, the maximum percentage of area interested by an extreme drought event is expected to be comparable with that of the baseline period. On the other hand, an increase of the drought-affected area is expected for the last thirty years of the century, with a maximum of 55% of interested area during a single episode as simulated by the pair HadGEM2-RACMO22E for the RCP 4.5 scenario. This percentage is even larger in RCP 8.5, with 68% estimated by the same pair.
The maximum duration of the main drought events ranges from 13 weeks to 25 weeks for the 1971-2000 period (Figure 4). An increase of drought duration is detected for the near future in the RCP 4.5 scenario, with maximum duration ranging from 15 consecutive weeks for the CM5-ALADIN52 pair to 29 weeks for EC-EARTH-HIRHAM5. A further increase is expected for the last 30 years of the century; for the period 2071-2100 the maximum duration will be between 21 and 32 weeks, respectively estimated by CM5-ALADIN52 and EC-EARTH-HIRHAM5 in the RCP 4.5 scenario. Similar results, with larger changes, are obtained for RCP 8.5.
The comparison between the two drought indices calculated at 12-, 24- and 36-month time scale, for the baseline, near and far future periods is reported in figure 5. Positive values stand for SPEI-identified episodes that are more extended than those identified by SPI. In 1971-2000, for all GMC/RCM pairs there is a tendency of SPEI to detect drought events that are more extended than for SPI; at 36-months SPEI detects events that are 25% and 30% more extended than those estimated by SPI (Figure 5a). Consistently, in 2021-2050, for all GMC/RCM pairs, SPEI detects more severe drought events than SPI (Figure 5b). This happens for all the timescales of integration, in particular at 36-months the events detected by SPEI are 17% and 25% more extended that for SPI. For the 2071-2100 period, the difference between SPEI and SPI is more varied, and it was not possible to detect a consistent difference between the two indices (Figure 5c).
5.4 Drought trends and spatial behaviour
The analysis of the most intense drought events identified an increase of severity in the 2071-2100 period, in terms of both the duration and percentage of drought-affected area. Figures 6 reports the 12-month SPEI trends and their statistical significance in the 2021-2100 period for RCP 4.5 and RCP 8.5, for the four most reliable GCM/RCM pairs. For RCP 4.5 (figure 6a), HadGEM2-CCLM4, HadGEM2-RACMO22 and CM5-ALADIN52 show a significant intensification of droughts along the Alpine chain. In this area, the drought index is expected to decrease between -1.2 and -1.8 ΔSPEI/30years. In other parts of northern Italy, the three model pairs indicate no significant and close to zero trend for the 2021-2100 period, while EC-EARTH-HIRHAM5 indicates a significant increase of droughts also for the central portion (North of the Po Plain, -1.8 ΔSPEI/30years). For RCP 8.5, the models estimate an increase of drought severity (Figure 6b). For HadGEM2-CCLM4 and HadGEM2-RACMO22, the analysis showed significant trends for the whole study area, which becomes especially intense in the Alps (2.4 and -3 ΔSPEI/30years).
The change in the spatial extent of drought episodes, compared to the baseline condition, was investigated considering also how droughts develop in case of a global warming of +2 or +3 °C above preindustrial levels, in keeping with the recent IPCC approach. Figure 7 shows the spatial extent of heavy and extreme drought events according to SPEI and SPI at 12-months. In northern Italy, heavy drought events are homogenously distributed in the 1971-2000 period with duration between 6 and 8 consecutive weeks. Longer extreme drought episodes are detected by SPI in the Alps (8 to 10 weeks, Figure 7a). Both indicators reveal that, for a global temperature increase of +2°C, heavy drought events will be present in the study area, with duration from 5 to 15 consecutive weeks. In a few sectors along the Alps, extreme drought events with a duration of 20 weeks are also foreseen (Figure 7b). Moving to a +3°C global temperature increase, no relevant enhancement of heavy drought events is observed. On the opposite, extreme drought episodes show a strong increase. In particular, the southern portion of the Po Plain is expected to experience extreme droughts with duration of 30 consecutive weeks (Figure 7c). Once more, the difference in climate change impacts between +2 °C and +3 °C is striking.