6.1. Methodological Considerations
Assessing the sensitivity of the glacio-hydrological system of the study area based on past observations, instead of future projections, enabled to reduce the main source of uncertainty in sensitivity studies based on future climatic scenarios, whose prediction shows a considerable spread e.g., [
13]. This was feasible thanks to the rare hydro-meteorological time series available in the study area.
Past climatic observations, however, are not free from problems. In particular, they are subject to inhomogeneities and measurement errors. The setup of a solid meteorological input was an essential prerequisite for obtaining meaningful results in our sensitivity analysis.
The homogenization technique adopted in this work was primarily intended to preserve as much as possible the original data. For this reason, we avoided using precipitation data from outside the study area to homogenize the raw series, taking in consideration the high spatial variability of precipitation in mountain areas [
46]. In addition, we avoided correcting smaller inhomogeneities, in particular for temperature series, considering possible impacts in the solid vs. liquid partitioning of precipitation. Inhomogeneities in temperature measurements are indeed expected to be larger under stable weather conditions (high radiation, low wind speed), and smaller during precipitation events. Therefore, applying a correction to inhomogeneous periods in temperature series is expected to have a higher impact (excessive temperature correction) during precipitation events, further amplified by the application of precipitation correction factors.
Some assumptions have been made while pre-processing the meteorological data before modeling, mainly regarding the time/space invariance of precipitation correction factors and gradients of temperature and precipitation. Precipitation correction factors are likely different for different instruments and prevailing meteorological conditions during precipitation events. These differences should be higher at high altitude, were the fraction of solid precipitation and the wind speed are larger. Carturan [
29] and Carturan et al. [
47] investigated this problem at the high-altitude weather station of the Careser Diga (2605 m), finding negligible difference in precipitation correction factors between the old manual rain gauge, used until 1991, and the new tipping-bucket automatic heated rain gauge, in use since then. No information exist for other rain gauges of the study area, but given the smaller undercatch of precipitation at lower altitude, we think that the assumption of unchanged correction factors between the old manual instruments and the new automatic ones is reasonable.
Vertical gradients of precipitation and temperature exhibit a significant variability in space and time [
48]. Monthly vertical gradients in temperature and precipitation were calculated in periods with overlapping data from available weather stations located at different elevation: from 1940 to 1984 for precipitation, and from 2003 to 2014 for temperature. In absence of data from the same stations outside these periods, we have calculated a mean annual regime for temperature and precipitation vertical gradients, applying it unchanged in all the modeled time windows. This simplification, that does not consider the inter-annual variability of vertical gradients, was imposed by the availability of meteorological data.
The horizontal gradients of precipitation are accounted for using the spatial interpolation procedure described in
Section 4.2. The horizontal variability of temperature, deriving for example from valley winds, warm and cold advection, local adiabatic heating, warming or cooling of different surfaces, is highly complex and difficult to model. The glacier cooling effect is a dominant factor of temperature variability, with relevant implications for glacier sensitivity to climatic changes over a long period of time [
49,
50]. However, multi-glacier fully-distributed modeling of the glacier cooling effect is not trivial and still in development [
50,
51,
52]. For this reason, and because we have decided to focus on short time windows assuming unchanged glacier extent, the glacier cooling effect was not explicitly modeled, and was accounted for through calibration of melt parameters.
Depending on future climatic conditions, future glacier extent modeling is also subject to large uncertainty, further complicated by the existence of feedbacks that are still under investigation. Available reconstructions of past glacier extent enabled to investigate the hydrological response in our study area using constraints from actual measurements of glacier change.
While the Current extent and surface topography of glaciers are very well documented by recent LiDAR surveys, there are uncertainties in the reconstruction of glacier bedrock topographies (i.e., the Noglac scenario) and of surface topographies for the LIA glacier extent. The calculation procedure employed provided errors in single point ice thickness calculations of 20%, which is in line with previous works [
32,
33], and is acceptable considering the modeling simplifications.
One of the main issues of this work was finding information on glacier geometry as close as possible to the oldest hydro-meteorological data available (i.e., for the 1920s–1940s). We used the maximum Little Ice Age extent of glaciers, because reconstruction in this period is much more reliable, thanks to the well preserved and widespread landforms left by the glaciers [
17], compared to more recent fluctuations in the late 19th Century or in the first half of the 20th Century. Assuming that in the 1920s–1940s the glacier extent was close to that of the LIA is allowable. Based on available reconstructions in the study area [
53,
54], the La Mare and Careser glaciers were only 9–10% smaller in the 1920s–1930s compared to the maximum LIA extent (see for example
Figure 9). Reconstructions available for the Lobbia [
55], Pisgana [
56], and Solda glaciers [
57], which are close to the study area, further support this assumption, as well as the comprehensive work of Desio [
58]. The choice of focusing on single decades, keeping unchanged the glacier extent during these periods, was done because continuous modeling of glacier dynamics and geometric adjustment was outside the aims of this work.
The glacio-hydrological model that has been used in this work is a combination of a fully distributed module for the simulation of snow/glacier mass balance, coupled to a semi-distributed module for runoff routing. The choice of the modeling tools, and of the spatial domain to which they have been applied, was primarily dictated by the modeled processes. Because glacial processes typically show a high spatial variability, at the scale of tens of meters, they require sufficiently high spatial resolution to be taken into account with enough accuracy [
59,
60,
61]. This is even more true for investigations on the sensitivity of snow and ice to climatic fluctuations, as in this case.
Hydrological processes also show high spatial variability, but conceptual and semi-distributed approaches are commonly used for modeling snow and ice hydrology in mountain catchments e.g., [
62,
63,
64,
65]. This is justifiable in this work with the fairly lower climatic sensitivity of the processes controlling runoff routing, compared to cryospheric processes. We had also to consider the availability of measurements to be used for constraining model parameters. In this work, the internal consistency of the snow/glacier mass balance module could be ensured by detailed measurements of snow and ice mass balance, at various spatial scales, whereas runoff routing parameters could be constrained using lumped data, i.e., outlet streamflow measurements.
6.2. Considerations on the Sensitivity Analysis Results
The climatic conditions in the 1940s were rather similar to those in the 2000s, in particular for the warm summer temperature and for the seasonal distribution and amount of precipitation. The heavy precipitation in November 2002, November and December 2008 and November 2010 make the 2000s more snowy, on average, at the glaciers’ altitude. The 1970s were significantly more favorable for glaciers (leading to their observed expansion between the 1970s and 1980s, [
66]). Temperature was ~2 °C lower than in the 1940s and 2000s between March and September, and precipitation was higher, in particular during winter and spring. The solid fraction was also larger, principally in summer, with important effects on the summer balance of glaciers [
53,
67].
The similarity between the 1940s and the 2000s is remarkable, as well as the particular conditions measured in the 1970s. This behavior is likely affected by multi-decadal patterns of oceanic thermal anomalies, like the Atlantic multidecadal oscillation (AMO) and the Pacific decadal oscillation (PDO), further modulated by continental-scale circulation patterns such as the North Atlantic oscillation (NAO) and the East Atlantic pattern (EA), which influence the temperature and precipitation distribution in winter and summer across Europe [
68,
69,
70]. Most of these patterns favored positive glacier mass balance (i.e., positive storage) in the 1970s, and negative mass balance (i.e., negative storage) in the 1940s and 2000s.
These analogies/differences among the three periods are clearly visible in the hydrological response of the four analyzed catchments (
Figure 5,
Figure 6 and
Figure 7). The 2000s and 1940s behave very similarly, and with distinct differences when compared to the 1970s. As found e.g., by Huss, [
5], smaller catchments with higher glacierized area fraction are comparatively more affected by changes in glacier storage, while larger catchments are more affected by changes in the precipitation regime and seasonal snow accumulation/ablation. Climatic conditions during the 1970s are emblematic. There is significantly higher runoff in mid-summer during the 1970s in the larger catchments, and differences from the 1940s and 2000s increase from LIA to Current to the absence of glaciers (
Figure 7). Therefore, these differences are not attributable to enhanced glacier runoff, but to specific climatic characteristics of the 1970s. In particular, the larger snow accumulation from October to May ensures sustained runoff from seasonal snow during summer, also in case of absence of glaciers. In the smaller catchment there is an opposite behavior, with higher mid/late summer runoff due to glacier melt in the 1940s and 2000s, and lower runoff in the 1970s due to positive glacier storage.
A considerable change in the hydrological regime from June to October was detected in response to decreasing glacier cover. This change consists mainly in a strong decrease of unit discharge after the seasonal snow is melt, in the second half of summer, with a maximum decrease in the month of August (
Figure 5). There is also the tendency to an anticipation of the runoff peak from August to July in the smallest and most glacierized basin, whereas this effect is not visible on larger catchments where seasonal snow dominates the warm season runoff regime. Under the 2000s meteorological conditions, the peak is in July also for the LIA glacier extent, due to the warmer spring temperature compared to the 1940s and 1970s, in combination to higher snow accumulation outside the glaciers in autumn and early winter. Variations in summer runoff among glacier cover conditions are much more evident for the warm 1940s and 2000s meteorological periods, compared to the cooler 1970s.
These results highlight increased hydrological sensitivity towards warmer climatic conditions, reinforced by glacier decay, and suggest impacts for downstream catchments, in spite of their very small percent glacierization. According to the literature, macroscale transboundary catchments, as for example the Po valley, benefit from glacier runoff during the second half of summer in a significant way. Even if they are far from glaciers, the runoff from glacier melt is relevant because in the plains the summer is much warmer and drier, with strong evapotranspiration. Huss, [
5], for example, demonstrated that glacier contribution to August runoff at Pontelagoscuro and Piacenza reaches 15–20% on average, with peaks of 30% during warm and dry summers, like in 2003.
In the 2000s, the glacier contribution in August (i.e., the negative glaciers storage change from the water balance equation) ranges from 43% to 84% with the LIA glacier cover, and from 26% to 78% with the Current glacier cover, with percent contributions that are inversely proportional to catchment area (
Figure 10). The decrease in glacier contribution with increasing catchment area is not linear. The largest decrease is observed between the smaller catchments, in our case between the Pian Venezia and Vermiglio basins, and becomes much slower for larger catchments. We do not observe the increasing importance of glacierization, described for macroscale catchments by Huss [
5], because our larger catchment (Tassullo) is rather small in comparison, and because in the study area the climatic conditions are rather homogeneous.
Glacier contribution can strikingly increase during extreme conditions, like in summer 2003.
Figure 10 shows that in August 2003 the percent glacier contribution is nearly twofold for the three larger catchments analyzed, compared to average meteorological conditions in the 2000s, reaching 58% at Tassullo. The increase in the headwater catchment is of only 15% because glacier runoff already dominates under average meteorological conditions. Based on these results, relevant impacts from glacier vanishing are expected during warm and dry years like 2003, not only in headwater catchments but also on downstream catchments. Similar results were obtained for example for the Skykomish River, Washington, by Pelto [
10].
When compared to glacio-hydrological experiments that use future modeled atmospheric warming scenarios [
71,
72], these results confirm a strong decrease in the capability of glacierized catchments in damping the runoff fluctuations caused by precipitation/temperature variability. This decrease is related to the progressive melt out of residual ice masses, but is preceded by a period of increased runoff derived from negative glacier storage change, with a peak that is expected between 2020–2040 for highly glacierized catchments [
7,
73,
74]. It is therefore interesting to assess ‘at which point’ our study area is in this transition. To do that, we have quantified the residual damping effect (RDE) of Current glaciers, compared to that of the LIA glacier extent
where
DE is the damping effect, calculated as
where,
qgc is the unit discharge of August with the LIA or Current glacier cover, and
qga is the unit discharge of August in absence of glacier cover. Results are displayed in
Table 5, and show that in the smaller and most glacierized catchment RDE is around 50%, whereas in the larger catchments it is about 30% (24% for the Tassullo catchment with the 2003 meteorological conditions). This means that three/quarter of the LIA damping effect are already vanished from the study area, and only in headwater catchments it still reaches half of the original value.
The summer of 2003 has been frequently referred to as a possible example of future climatic conditions during summer in the Alps, under sustained CO
2 emissions scenarios [
75,
76]. To reveal if the peak in runoff due to glacier wastage still has to come, or has already passed in the study area, we compared the runoff in August 2003 with the runoff under ‘average’ conditions during the 2000s, with different glacier cover. Results show that the LIA glaciers would have ensured higher unit runoff on all four analyzed basins in August 2003, compared to average 2000s meteorological conditions (
Table 5). In contrast, the Current glacier cover enables increased unit runoff only in the Pian Venezia basin.
These results are not conclusive because a complete assessment would have required continuous modeling of climate, glacier mass balance, glacier dynamics, geometric adjustments, and runoff for a sufficiently long time period, spanning the latest decades and the next two or three decades. However, the values displayed in
Table 5 strongly suggest that the peak in late-summer runoff under warming climate, due to glacier wastage, has already passed in Val di Sole, with the exception of few glacierized headwater catchments. Because more than half of the glacier area has been lost from most of the watersheds (
Table 1), it would require more than a doubling in summer ablation compared to the 2003–2012 mean to be offset, which has a low probability of occurrence.
Koboltschnig and Schöner, [
9], came to similar results, finding that only Austrian catchments with glacierization larger than 10% were able to provide larger runoff in August 2003, compared to mean long-term August runoff. Other works in the recent literature agree with our findings and indicate that the peak in runoff under warming scenarios is expected earlier for catchments with lower initial glacierization and smaller/thinner glaciers [
7,
74,
77], or has already passed [
10,
78,
79,
80].
It is interesting to note that, also without glaciers, there is a small increase in unit runoff during August 2003 at Pian Venezia, compared to 2000s average meteorological conditions. This increase is brought by the melt of snow accumulated at high altitude in previous years (impending formation of small glaciers during model initialization). However, this marginal increase in unit runoff and the formation of small glaciers is observed when 2003 represents an exceptionally warm summer, and are unlikely to occur when 2003 is assumed to represent average meteorological conditions.