#### 4.1. Runoff Volume Series Analysis

Table 2 reports the trend slopes

m_{k} of each individual series along with the corresponding lower and upper confidence boundaries

m_{kl} and

m_{ku}, derived for a confidence interval of 90%. As can be seen, all series show a decreasing trend, with rates varying from −1.16 to −3.33 mm/year. The uncertainty related to these estimates amounts to about ±0.60 mm/year for the longest series (the Adda and Adige rivers) and to ±1.75 mm/year for the shortest ones (the Oglio, Chiese, and Mincio rivers). As regards the shortest series, such uncertainties are significantly high. More precisely, the estimate of the Oglio series trend rate, at −1.16 mm/year, is such that the uncertainty interval includes even positive values.

The Theil test for null slope (i.e.,

m_{k} equal to 0), the Mann–Kendall test, and the Spearman test all consistently evidence statistically significant trends for all series, except for the Oglio river basin. In

Table 2, the

p-values (

α_{0k},

α_{tk}, and

α_{rk} refer to the Theil, Mann–Kendall, and Spearman tests, respectively), for which the null hypothesis of trend absence cannot be rejected, are generally much less than 1.0%. Conversely, for the Oglio river series, a significance value slightly greater than 32% is estimated. It is worth noting that the Oglio river series trend is close to the one from the Adda river. Therefore, this result could be explained by considering that this trend size would need a far longer series to be recognized as significant.

The homogeneous nature of the observed individual trends is supported by the Sen–Adichie test. As can be seen in

Table 2, the

p-value

α_{p}, according to which the null hypothesis of parallelism cannot be rejected, is 54.8%. This evidences a high confidence both in this hypothesis and in the possibility of exploiting the pooled slope as an estimate of a regional trend rate. This value amounts to −1.45 mm/year due the major sample size of the Adda river and the Adige river series in the pooled sample with respect to the others. Finally, the Theil tests were repeated assuming the pooled slope

m_{p} as individual trend slope. Such tests always yield

p-values

α_{pk} larger than 10% for the hypothesis not to be rejected. In particular, the significance of the Oglio river series is slightly larger than 70% due to the similarity of the individual series’ slope with the pooled slope. Therefore, on a regional level, there is evidence of a statistically significant decreasing trend in the annual runoff volumes from the watersheds of the Central Italian Alps.

A visual representation of the limited slope variability of regression lines is supplied in

Figure 2. The time series of the annual runoff volumes are plotted in the period 1934–2018, when there is a high overlap of data availability, along with the corresponding individual trends, computed with respect to the complete series. Intercepts of regression lines plotted in

Figure 2 are estimated according to Equation (3).

The decreasing trends estimated for the annual runoff volumes could be related both to nonstationarity in the annual precipitation depth and to a simultaneous increase in annual hydrologic losses. A combination of increasing air temperatures and extensive variations in land cover is probably responsible for this negative trend. Indeed, other potential factors, such as changes in irrigation practices and in glacier coverage in this area, were taken into account and excluded. The first factor is excluded by considering the scarcity of irrigated crops in the mountains (the only exception being fruit tree cultivations in the Adige and Adda watersheds whose coverages are, however, trivial as compared with the total watershed area; see land cover discussion reported below). The seasonal regulation of lakes is mainly due to the need to satisfy the demand for irrigation water. Nevertheless, irrigation has an impact on freshwater exploitation only in the downstream lowlands, and the analyzed time series are not affected by relevant irrigation uptakes. Moreover, the second factor can be excluded by bearing in mind the well-established evidence of a globally significant trend in glacier retreats in the Alps [

2], which, on the contrary, should partially compensate the negative trend in runoff volumes by supplying additional contributions as a result of the ice melting.

#### 4.2. Precipitation Series Analysis

An analysis analogous to the runoff volume one was conducted for the mean areal precipitation depths of the Chiese and the Adda watersheds, and its results are reported in

Table 3. Plots of the precipitation series and corresponding trends are instead illustrated in

Figure 3 for the period 1930–2018 when both series are available.

It is worth noting that the precipitation trend rate in the Chiese watershed amounts to +1.05 mm/year and is strongly in contrast to the decreasing trend rate of −3.12 mm/year, estimated for the runoff volume series (

Table 2). Moreover, when the Theil test for individual regression slopes is conducted by assuming the trend absence, the

p-value

α_{0k} according to which the null hypothesis cannot be rejected is 40.8%. The Mann–Kendall test and the Spearman test yield analogous results since their coefficients are close to zero and the corresponding

p-values

α_{tk} and

α_{rk} are greater than 40%. When the mean areal precipitations of the Adda watershed are considered, a decreasing trend is found. However, the trend rate is far less than the runoff volume one (

Table 2), and its statistical significance is negligible, as confirmed by the

p-values

α_{0k},

α_{tk}, and

α_{rk} for the trend absence hypothesis not to be rejected, all larger than 20%.

When the two individual series are pooled together, the common trend rate is estimated to be negative and the parallelism hypothesis cannot be rejected with

p-values greater than 22%. Accordingly, the pool trend rate

m_{p} cannot be rejected as an individual slope for large

p-values, even for the Chiese series. All the tests that were performed, therefore, agree that the regional trend is nonstatistically significant since the hypothesis of trend absence cannot be rejected for large significance values. This outcome is consistent with that obtained for the Adige river watershed. A 150-year-long series of mean areal precipitation estimated from the Historical instrumental climatological surface time series of the greater Alpine region (HISTALP) database was investigated for the Adige watershed [

15], and a negative precipitation trend was assessed by the Theil–Sen estimator to be −0.31 mm/year. Despite the length of the series, the null hypothesis of trend absence cannot be rejected for significances larger than 29%, according to all three tests that were used above. On the whole, it can be concluded that the marked regional negative trend in the annual runoff volumes visible in the major rivers of the Central Italian Alps cannot be justified by means of a statistically significant decrease in the annual precipitation depths, and therefore, trends in hydrologic losses must be taken into consideration, in particular the evapotranspiration processes.

Two factors affect such processes: temperatures and land cover. Regarding the first factor, there has been significant progress in the field of temperature studies, with evidence of statistically significant increasing trends having been collected. For instance, it has been demonstrated that the whole Alpine subregion is characterized by uniform trends in temperature [

34]. Over the period from 1886 to 2005, the rate increase in mean annual temperatures was 1.4 ± 0.1 °C/century [

10], with very high confidence in accordance with the Mann–Kendall test, which confirms the increased speed of the global warming trend. In the second half of the 20th century, trend rates are acknowledged to be higher than those observed in the same regions in the previous half.

Nevertheless, increasing trends in temperatures alone cannot explain those in total hydrologic losses. For instance, the increase rate in observed hydrological losses (i.e., the difference between the mean areal precipitation and the runoff volumes) in the Adda watershed is higher than the increase rate estimated for the potential evapotranspiration by accounting for only the temperature nonstationarity [

14]. By using the Thornthwaite method on a gridded surface assuming stationary land cover, the trend rate of mean areal potential evapotranspiration (including evaporation from water surfaces) is assessed to be 36 mm/century. In contrast, the trend rate in the mean areal total hydrologic losses is assessed to be 92 mm/century.

#### 4.3. Land Cover Transformation Analysis

In consideration of the remarkable difference between trend rates of annual losses and evapotranspiration, additional causes must be advocated to explain the more rapid decline in observed runoff volumes. Consequently, land cover transformations must be taken into consideration as an additional reason for the observed regional decrease in the streamflow discharges by using spatial data from the Lombardy Region database. Large portions of the watersheds in this study lie within the Lombardy Region boundary (see

Figure 1); the only exception to this is the Adige river. The analysis is particularly significant for the Oglio river watershed, which is completely covered by the available spatial land cover data; for the Adda river watershed (89% area analyzed); and for the Chiese river and Mincio river watersheds (57% and 28% of the area analyzed, respectively). Overall, the area investigated for land use changes covers 7020 km

^{2}. Seven macroclasses, featuring highly different responses in terms of hydrologic balance, were considered: woodland (including broad-leaved trees, conifers, and mixed woods), bushland (including scrub and/or herbaceous vegetation, except for natural grasslands), cropland (including arable lands but excluding permanent crops and pastures), fruit trees (including permanent crops, such as apple orchards, vineyards, and olive groves), grassland (including natural grasslands and pastures), urban (including all artificial surfaces), and others (including water bodies, wetlands, glaciers, and uncultivated lands).

The regional land cover layers were then clipped by the total watershed divide of the analyzed area and reclassified according to the above-mentioned land cover list. The results are summarized in

Figure 4a, where they are expressed in terms of the area covered by each of these macroclasses, spanning from 1954 and 2018. In

Figure 4b, the percentage variations of area covered by each class, occurring in the three time periods separating the survey years, are reported. Such percentage variations were calculated using the difference between the present and the past coverage areas and dividing it by the past coverage area. Although the results are aggregated with respect to the total study area, they are actually representative of what occurred in all the analyzed watersheds since similar results were found for each of them.

The most impressive increase is associated with the urban land cover, which totally grew by almost 300% from 1954 to 2018 (increase of 250 km

^{2}), even if it remains a minor portion of the total watershed area (less than 5%). This evidence can be explained in the examined mountain context by the widespread urban development as a result of touristic demand for holiday houses and hotels. In particular, inside the study area, the western shore of Lake Garda experienced a dramatic urban sprawl, being strongly attractive for aquatic leisure activities. The urban expansion mainly occurred at the expense of croplands, which were subjected to strong decreases, totally assessed at 73%. Such an abrupt decrease can also be explained by the conversion into fruit tree cultivations, in particular apple orchards, vineyards, and olive groves. Areas devoted to fruit production totally increased by 66%. A demonstration of this is visible in

Figure 5, which represents the land cover transformation that occurred between 1954 and 2018 in a square sample area of 125 km

^{2} across the watershed divide separating the Adda and Oglio rivers. The major urban expansion is that of the Tirano urban settlement, located in the valley, which has progressively spread into the surrounding cropland areas. Furthermore, the substitution of arable land with permanent crops (in this case, apple tree orchards) contributed to the phenomenon of cropland disappearance. Due to the socioeconomic development of the area, permanent crops have progressively become more attractive and remunerative than traditional seasonal crops.

The rest of the land covers show different trends, with figures for woodland increasing (total variation by about 20%) and those for bushland and grassland decreasing (total percentage variation by about 26% both, starting from 600 and 1380 km

^{2}, respectively, in 1954). As a consequence, woodland that covered 37.8% of the total area in 1954 expanded its area by up to 45% in 2018 (

Figure 4a), corresponding to an areal expansion of about 510 km

^{2} in 2018 out of the 2650 km

^{2} covered in 1954. This result is not surprising as an increase in the percentage of land covered by woodland has been documented all over Europe by the Food and Agriculture Organization (FAO) [

35,

36]. The FAO states that woodland areas have universally increased by 1.1 × 10

^{6} km

^{2} since 1990 in nontropical regions, with their highest rate of growth between 2000 and 2010. Overall, the phenomenon is referred to as afforestation and accounts for three processes: the spontaneous regeneration of woodlands, the reforestation of formerly wooded areas, and the afforestation of new areas, which had previously been used for different purposes.

The existence of afforestation trends has also been confirmed in other regions of the Alps in Switzerland and Austria. By comparing georeferenced land registry maps with current satellite images, land cover changes were investigated in some sample areas of the Adige river watershed, and it was found that an afforestation trend can be identified from the mid-19th century onwards, with strong variations in the percentage increases, depending on the sample area location [

15]. The evolution of this phenomenon since 1954 is clearly shown in

Figure 5, selected as an example. It is evident that the increase in woodland coverage mainly resulted from the contraction of bushland and grassland. Afforestation can thus be understood in terms of the abandonment of pasture activities and the reduction of woodland exploitation for firewood harvestings. Forestry has also gained prominence in mountain communities, which has led to better woodland maintenance and promotion of woodland expansion.

Therefore, the land cover transformations assessed in the watersheds analyzed between 1954 and 2018 are consistent with those observed in Europe. Although the urban sprawl is characterized by far larger percentage increases than afforestation, woodlands still account for a large portion of land coverage, as made clear in the areas detailed in

Figure 4a. As opposed to urban sprawl, afforestation can therefore have a noticeable impact on the hydrologic balance by increasing evapotranspiration. Measurements and simulations of evapotranspiration processes conducted on nearby Swiss watersheds at variable spatial scales by using the Penman–Monteith equation [

37] demonstrate that annual evapotranspiration depth increase with the woodland cover percentage. A crucial role is played by the leaf area index of woodlands, which is far greater than that of grassland. The impact of afforestation in decreasing mean runoff volumes is also clearly evidenced by recent studies conducted in Swiss Alps [

38], which simulated the effects of afforestation and deforestation scenarios on streamflow discharges in seven Alpine watersheds and recorded a noteworthy impact when the increase in woodland area was about 20%, with a 1% runoff volume decrease per 10% increase in forested areas, approximately. Potential impacts involve an increase in hydrologic losses, namely, in canopy interception and evapotranspiration. Applying the results reported in [

38], the 20% increase in woodland we identified in the investigated area for the 1954–2018 period would result in a 20 mm decrease in runoff volume, on average, a value that can explain a major part of the 36 mm of runoff volume losses that cannot be explained by claiming the enhanced potential evaporation due to temperature increase only, assessed with a −36 mm/century rate for the Adda watershed in [

14]. Afforestation can therefore be confirmed as a potential concomitant cause for the statistically significant decreases assessed for the annual runoff volumes.

A final remark regards the expansion of water surfaces due to the construction of artificial reservoirs during the last century. It must be considered that their total water area is about 20.6 km

^{2} in the Adda, Oglio, Chiese, and Mincio watersheds, whose overall catchment area measures 9632 km

^{2} (see

Table 1). According to [

39], the evaporation from lakes in Italy is about 1200 mm per year at middle elevations, while the average actual evapotranspiration losses in watersheds similar to those investigated in this paper are about 500 mm per year [

37]. Therefore, the increase rate in annual evaporation losses due to the artificial water surface expansion in the 20th century in the area of these four watersheds can be estimated to be less than 1.5 mm/century. As a consequence, this factor cannot be advocated as a significant concomitant cause for the marked runoff volume decline.