Theoretical and Applied Climatology

, Volume 100, Issue 3, pp 351–369

Evidence of climate change within the Adamello Glacier of Italy

Authors

    • Deparment I.I.A.R.Politecnico di Milano, C.I.M.I. Section
  • Guglielmina Diolaiuti
    • Earth Science DepartmentUniversità degli Studi di Milano
Original Paper

DOI: 10.1007/s00704-009-0186-x

Cite this article as:
Bocchiola, D. & Diolaiuti, G. Theor Appl Climatol (2010) 100: 351. doi:10.1007/s00704-009-0186-x
  • 205 Views

Abstract

We analyze a daily series of rainfall, snowfall, air temperature, and snow water equivalent at fixed dates from 40 high-altitude stations on the Adamello Glacier area (Italian Alps), for the period 1965–2007. Purposes of the study are (1) to investigate significant variation in time, (2) to evaluate effect of temperature changes on cryospheric water cycle, and (3) to evaluate underlying climate patterns and the most significant variables for climate change studies. We detect the presence of a trend using linear regression, moving window average and Mann Kendall test. Linear dependence of water related variables on temperatures is assessed. We find substantially unchanged atmospheric water input along with increasing temperature and rainfall, decreasing snowfall and snow water equivalent at thaw, and shortening of snow cover extent and duration. We carry out a principal components analysis which highlights patterns of precipitation distribution resulting from local temperature and external forcing. A set of the most representative variables for climate and glacier studies is then assessed. A comparison with three nearby Southern Alpine glacierized areas in Italy and Switzerland shows substantial agreement. In spite of the relative shortness of the series, the results here are of interest and can be used as a benchmark for climate change impact assessment for the Adamello Glacier area and southern Alps.

1 Introduction

The recognized evidence of global warming requires assessment of its impact on the hydrological cycle and water resource distribution within the mountain areas in temperate regions, and upon the Alpine environment (e.g. Barry 2003; Barnett et al. 2005). In Italian mountain ranges, the importance of water from the Alps has emerged clearly during the latest dry summers, most notably in year 2003, when it saved most of the tributaries of the Po river from the severest droughts, and evidence is raising of ongoing variability of European Alpine water resources due to transient climate change (Rohrer et al. 1994; Beniston 1997; Laternser and Schneebeli 2003).

Snow cover extent, duration and dynamics influence vegetal and animal biota in Alpine areas (e.g. Erschbamer 1989; Gottfried et al. 1999; Theurillat and Guisan 2001; Kulakowski et al. 2006), and freshwater availability from cryosphere during spring and summer regulates the hydrological cycle of Alpine basins and lowlands downstream (e.g. Coughlan and Running 1997, McGlynn et al. 1999, Beniston et al. 2003). Transient climate change influences water budget (e.g. Rohrer et al. 1994; Beniston 1997; Singh and Kumar 1997; Braun et al. 2000; Spreitzhofer 2000; Laternser and Schneebeli 2003; Schneeberger et al. 2003), spring and winter stream flows (e.g. Braun et al. 2000; Liu et al. 2003), and dynamics of natural water bodies (e.g. Stefan and Fang 1997).

The hydrology of Alpine basins was studied in different ways. The amount of snow water equivalent (SWE) cumulated in snowpack and available at spring thaw has been investigated (e.g. Bohr and Aguado 2001), and daily SWE distribution has been studied (e.g. Bocchiola and Rosso 2007b), together with snow-pack distribution in time (e.g. Skaugen, 1999). Local evaluation of climatic trends, maybe coupled with scenarios from climatic models was used (e.g. Gellens and Roulins 1998; Gondor et al. 2000; Drogue et al. 2004; Kang and Ramírez 2007) to provide the climatic input for medium- to long-term impact analysis on the habitats of animal and vegetal species (e.g. Gottfried et al. 1999; Fiorese et al. 2005; Keller et al. 2005, Cannone et al. 2007; 2008), on water resources (e.g. Yu et al. 1999; Braun et al. 2000; Arora and Boer 2001; Beniston et al. 2003; Hagg and Braun 2005), hydrological extremes (e.g. Boroneant et al. 2006) and extreme snowfalls (e.g. Bocchiola et al. 2009).

General circulation models (GCMs) and limited area models (LAMs) are physically based tools presently used in predicting climate change effects (e.g. Bardossy 1997; Bates et al. 1998). Although they perform reasonably well in simulating synoptic atmospheric fields, they fall short in estimating precipitation at the spatial scales of interest for hydrological conjectures (e.g. Mearns et al. 1995; Walsh and McGregor 1995; Gangopadhyay and Clark 2005) This is particularly true within mountain areas, where topographically controlled snowfall plays a major role within hydrology and ecology (e.g. Beniston et al. 2003; e.g. Keller et al. 2005; Cannone et al. 2007; 2008). As such, accurate assessment of local climate trends is also necessary to make use of GCMs for climate change impact projections in high altitude areas (e.g. Rotach et al. 1997).

Usually, reference windows for climate change studies are a length of 30 years or so (e.g. 1961–1990, IPCC 2000; Juen et al. 2007; Jiang et al. 2007). Here, we investigate a 42-year window (1965–2007) of observations from a network of 40 meteorological and snow gauging stations inside the Adamello Park (Italian Alps). The Adamello Park nests the Adamello glacier, at 18 km2 of area, the largest Italian glacier, and several smaller ice bodies spreading out across an area of 24 km2 (Fig. 1a). The glaciers located in the Adamello Park are both mountain and valley glaciers, also including debris covered types, thus being representative of the Italian and European Alpine glaciation (Fig. 1b). This study was carried out under the umbrella of the CARIPANDA project, funded by the Cariplo Foundation of Italy, with the intention of evaluating scenarios of water-resources distribution for the Park in a window of 50 years or so (until 2050). We investigate daily series of rainfall, snowfall and air temperature, together with cumulated SWE at fixed dates during the melting season. We address the presence of climatic trends using three different methods including linear regression, moving window average and Mann Kendall test (e.g. Cislaghi et al. 2005; Jiang et al. 2007). Then, we investigate the relationship between rainfall and snowfall precipitation, as well as snow deposition and depletion timing versus temperature, i.e. the main driver of modifications in Alpine areas (e.g. Bultot et al. 1994; Beniston, 1997; Schneeberger et al. 2003; Keller et al. 2005). We then carry out a principal component analysis to unravel the underlying patterns and relationships between climatologic and hydrologic variables of the area and to highlight the most representative variables to be accounted for in climate change study (e.g. Clausen and Biggs 2000; Huth 2000; Alpert et al. 2004). To place our study within the context of existing knowledge on climate change in the European Alps, we carry out a comparison with three nearby Southern Alpine glacierized areas in Italy and Switzerland.
https://static-content.springer.com/image/art%3A10.1007%2Fs00704-009-0186-x/MediaObjects/704_2009_186_Fig1_HTML.gif
Fig. 1

a Study area and position of the park. Reference glacierized areas also reported, together with reference weather stations used for comparison. Acronyms explained in text. b Measurement network. Glaciers of the Adamello group are labeled. For station numbers refer to Table 1

2 Case study area

2.1 The Adamello glaciarized region

The case study area covers the Adamello group, nested (for 23.2 km2 out of 24 km2) within the Adamello Park (Brescia Province, Italy), covering 510 km2 in the northeastern part of Lombardy region. Altitude ranges from 280 to 3,554 m asl (Mount Adamello), with an average value of 1,880 m asl. The park area (Fig. 1a) displays an alpine climate with very cold winter and moderate summer temperatures and considerable solar radiation and high frequency of clear sky conditions, mainly during winter. Precipitation in the Park area is quite low, with values around 1,000 mm year–1. Snowfall is frequent from October to May, and generally persists in time, particularly in the N–W area, in view of the cold air masses from the Pisgana Glacier and snowfall from N–E through the Tonale Pass. The Adamello glacier group covers an altitude belt from 2,530 (east of Pisgana Glacier) to 3,440 (Adamello Glacier) inside the park. At the highest glacierized altitudes, snowmelt starts in April and snow crust may be observed until half June, with subsequent start of ice melt. Total ice volume is estimated in 2 km3, for a ice water equivalent (IWE) of 1.86 km3 (Smiraglia et al. 2004). The Adamello area also represents an important research site and previous authors completed several studies deepening the understanding of the relations between geomorphological evidences and glacier variations (e.g. Baroni and Carton 1987). Several recent studies (e.g. Citterio et al. 2007a; Maragno et al. 2009) indicate consistent evidence of climate warming driven retirement of Adamello group’s glaciers, thus making accurate investigation of climate trends warranted.

3 Data and methods

3.1 Data base and treatment

The water resources of this area feed a number of reservoirs of ENEL hydropower company. Since 1965, trained personnel from ENEL carries out measurements of daily precipitation, P, and snow depth on the ground, HS, together with maximum and minimum daily temperature, Tmax and Tmin, at six sites in the proximity of its reservoirs, spread out between 1,820 and 2,500 m asl (Fig. 1b). Moreover, since 1965, observations of HS are available in 34 sites (Fig. 1b), performed 5 days a year (March 1st, April 1st and 15th, May 1st, June 1st) during the melting season by ENEL personnel (HS5). At these dates, snow density, DS, measurements were also carried out in 14 of these sites, reported and numbered in Fig. 1b, located between 1,820 and 2,800 m asl, with an average altitude of 2,216 m asl, and are quite uniformly distributed into the seven considered reservoirs, Pantano d’Avio, Venerocolo, Avio, Benedetto, Baitone, Salarno and Arno (DS5). Here we use data of HS5 and DS5 from these 14 stations to study the SWE amount, SWE5. This measuring network was established for water resource assessment with particular emphasis on snow cover, and represents the most comprehensive available climatic data set on the Adamello glacier we are aware of. ENEL Produzione of Milano provided data from this network for this study, and it is gratefully acknowledged. In view of the complex environmental conditions (some of the stations can only be reached using a helicopter during winter and early spring and often snow depth is remotely evaluated via hovering and with the use of binoculars) and of personnel availability for the surveys, a lack of data is observed in the data base. A preliminary analysis indicated four best stations in term of data availability, namely Arno, Avio, Salarno and Pantano (see Table 1), which we therefore have used for the assessment of climatic trends. Preliminary analysis indicated consistent patterns from the two remaining sites. In Table 1 we report an index of completeness C (i.e. the ratio of the number of available days of observations out of the total number of days in the measuring period, 0 ≤ C ≤ 1). In the four manual stations, the daily cumulated precipitation P is observed. To separate rain from snow, cross referencing with HS is carried out, and rainfall is conjectured when no (positive) change in HS is observed. We analyze here a number of variables. Both cumulated yearly precipitation (hydrological year, October 1st–September 30th), Pcum and liquid precipitation, Rcum, are considered. Then, snow pack depth HS is taken. The average value of HS, namely HSav is considered (1 October–31 May as from observations, see e.g. Laternser 2002). Then, values of HS over a given threshold are investigated (HS > 0, 5, 10 and 20 cm, while higher thresholds gave little results, compare e.g. Beniston et al. 1997; Laternser and Schneebeli 2003). Particularly, we investigate the first and the last day of snow cover HS above a threshold, B, and E, respectively (e.g. B5, E5, for HS > 5 cm). We initially hypothesized to study new snow depth HN, estimated as the positive difference of HS in two consecutive days. However, the yearly cumulated value of HN, HNcum is strongly correlated (ρ = 0.89) with HSav (as in e.g. Laternser and Schneebeli 2003), so making the assessment of a trend based upon HSav reasonable. Also, we evaluate the number of snowfall days NS (i.e. days with HN > 0). Eventually, we investigate the daily minimum and maximum temperatures, Tmin and Tmax. First, their average was considered for the whole hydrologic year (October 1st–September 30th). Second, we consider the temperature regime during fall (here, October 1st–December 30th) and spring (here, April 1st–June 30th), to evaluate the effect of a change in temperature regime on the snow cover dynamics, i.e. by shifting either the starting of snow deposition period (i.e. changing B), or the snow melting period (i.e. changing E). The data base (HS5, DS5) from the 14 snow-density measuring stations is used for SWE5, while HS5 from the remaining stations is used for cross referencing. For each station i the amount of snow water equivalent SWEi [m] is calculated as
$$ {\text{SWE}} = {\text{H}}{{\text{S}}_{\text{i}}} \cdot \frac{{{\text{D}}{{\text{S}}_{\text{i}}}}}{{{{\text{D}}_{\text{w}}}}} $$
(1)
with HSi snowpack depth (m), DSi (kg/m3) snow mass density and Dw (kg/m3) water mass density (Dw = 1,000 kg/m3). We evaluate for each date a spatially representative value of SWE5 by simple averaging, i.e.
$$ {\text{SW}}{{\text{E}}_{5,{\text{A}}}} = \frac{1}{14}\sum\limits_{{\text{i}} = 1}^{14} {{\text{SW}}{{\text{E}}_{5,{\text{i}}}}}, $$
(2)
where suffix A indicates spatial averaging.
Table 1

Measuring stations and measured variables. C is completeness index (i.e. number of available daily measurements out of the total number of days in the measuring period)

ID

Station

Altitude [m asl]

Reservoir

Measurements

Period

C[.]

1

Ovest Diga Pantano D'avio

2,520

Pantano D'avio

HS5, DS5

01/03/1967–01/06/2007

0.94

2

Est Lago Verso Adamello

2,590

Pantano D'avio

HS5, DS5

01/03/1967–01/06/2007

0.91

3

Diga Venerocolo

2,530

Venerocolo

HS5, DS5

01/03/1967–01/06/2007

0.94

4

Verso Vedretta Venerocolo

2,800

Venerocolo

HS5, DS5

01/03/1967–01/06/2007

0.89

5

La Palazzina

1,940

Avio

HS5, DS5

01/03/1967–01/06/2007

0.92

6

Dosso Prepazzone

2,000

Benedetto

HS5, DS5

01/03/1967–01/06/2007

0.80

7

Malga Lavedole

2,040

Benedetto

HS5, DS5

01/03/1967–01/06/2007

0.89

8

Sud Ovest Laghetto Miller

2,220

Baitone

HS5, DS5

01/03/1967–01/06/2007

0.82

9

Nord Est Lago Baitone

2,315

Baitone

HS5, DS5

01/03/1967–01/06/2007

0.86

10

Santa Barbara

2,070

Salarno

HS5, DS5

01/03/1967–01/06/2007

0.88

11

Destra Diga D'arno

1,820

Arno

HS5, DS5

01/03/1967–01/06/2007

0.91

12

Pozza D'arno

1,950

Arno

HS5, DS5

01/03/1967–01/06/2007

0.92

13

Sud Ovest Diga D'arno

2,040

Arno

HS5, DS5

01/03/1967–01/06/2007

0.81

14

Est Scale Adame'

2,185

Arno

HS5, DS5

01/03/1967–01/06/2007

0.83

15

Arno

1,820

Arno

HS, Tmax,Tmin, P

01/01/1966–28/02/2007

0.90

16

Avio

1,940

Avio

HS, Tmax,Tmin, P

01/01/1966–28/02/2007

1.00

17

Salarno

2,070

Salarno

HS, Tmax,Tmin, P

01/01/1966–9/02/2007

0.83

18

Baitone

2,200

Baitone

HS, Tmax,Tmin, P

01/01/1966–28/02/2007

0.46

19

Pantano

2,300

Pantano

HS, Tmax,Tmin, P

01/01/1965–28/02/2007

0.92

20

Venerocolo

2,500

Venerocolo

HS, Tmax,Tmin, P

01/01/1966–28/02/2007

0.60

3.2 Linear regression versus time LR

A weighted linear regression of the considered variables is carried out with respect to time (i.e. years). The weights are inversely proportional to the yearly values of C. A non null slope of the regression line indicates a trend with time. Significance of the regression is given using p–value with a level α = 5% (e.g. Jiang et al. 2007). Notice that multiple trend could be detected in time series analysis, e.g. by assessing slope changes in regression model or by Bayesian change-point analysis (see e.g. Seidou and Ouarda 2007). However, here focus is cast upon evaluation of the general trend, also in view of the relative shortness of the series, and therefore single regression analysis is carried out.

3.3 Moving window average vs. long-term average, MW vs. LT

A moving window average (MW) over a 10 years period is calculated (e.g. De Michele et al. 1998; Cislaghi et al. 2005). De Michele et al. (1998) proposed use of a moving average analysis with a 30–year window to evaluate the time evolution of maximum annual daily rainfall with a fixed return period for a 90-year-long series. In view of the shortest available series, here a moving window of 10 years is used. Applying the window at a sub-sample of 10-year data, the resulting value corresponds to the last year of the window and all the data have the same weight (e.g. the average of year 1999 is the 1990–1999 sample average). This technique should be called “nowcast moving average”, i.e. the moving average based on previous n years of observations, and allows comparison with the results of Mann-Kendall test, taking all the previous data available at a given time (see next section). This MW average is then compared with the long-term average (LT, 1965–2007). The MW fluctuations are then compared with the confidence interval for the LT, with a significance level α = 5%. The confidence interval is taken under the hypothesis of stationarity. Therefore, fluctuations of MW should be charged to the year to year (sampling) variability. In case the MW settles outside the confidence limits of the LT, the series may be affected by a non stationarity, either monotonic or periodic (e.g. Cislaghi et al. 2005). The confidence interval of LT is (e.g. Kottegoda and Rosso 1997)
$$ \mu - {z_{\alpha /2}}\,\sigma /\sqrt n \leqslant \overline X \leqslant \mu + {z_{\alpha /2}}\,\sigma /\sqrt n, $$
(3)
with μ and σ LT average and standard deviation from the whole sample and n moving window size. \( \overline X \) is the MW average and zα/2 is the quantile of the normal standard distribution with probability 1–α/2.

3.4 Mann-Kendall test, MK

Mann Kendall test (Mann 1945; Kendall 1975) is widely adopted to assess significant trends in hydrometeorological time series (Hirsch and Slack 1984; Gan 1998; Chiew and McMahon 1993; Lettenmaier et al. 1994; Zhang et al. 2000; Yue and Wang 2002; Bocchiola et al. 2008). This is non parametric, thus being less sensitive to extreme sample values, and independent from hypothesis about the nature of the trend, either linear or not (e.g. Wang et al. 2005), so being somewhat complementary to the LR test. Let consider a sample of a random variable, e.g. HS, {HSy, y = 1, 2,. . ., Y} with Y length of the series in years. Let py denote the number of elements of the sample with j < y and HSj < HSy, while τ indicates
$$ \tau = \sum\limits_{y = 1}^Y {{p_{\text{y}}}} . $$
(4)
One can show that τ is asymptotically normally distributed with the mean and standard deviation
$$ \mu \left( \tau \right) = Y\left( {Y - 1} \right)/4;\;\sigma \left( \tau \right) = \sqrt {Y\left( {Y - 1} \right)\left( {2Y + 5} \right)/72} . $$
(5)

The normalized variable \( u\left( \tau \right) = {{\tau \left( {\tau - \mu \left( \tau \right)} \right)} \mathord{\left/{\vphantom {{\tau \left( {\tau - \mu \left( \tau \right)} \right)} {\sigma \left( \tau \right)}}} \right.} {\sigma \left( \tau \right)}} \) has a standard normal distribution, so one can build the associated confidence interval. The Mann-Kendall test verifies the assumption of stationarity of the investigated series by ensuring that the normalized variable u(τ) is included within the confidence interval, for given significance level (for α = 5% range from –1.96 to 1.96). In the progressive form of the Mann-Kendall test, the variables τj and u(τj) are calculated for each element of the sample j, by trading Y for j in Eqs.(4) and (5). The value of τ defines the direction (positive/negative) and magnitude (modulus) of the trend. The same procedure is applied by starting from the most recent values and backward. In this case, pi’ indicates the number of elements of the series of HSy with j > y, and HSj > HSy. By py’ one gets τ’j and u(τ’j). If no trend is present, the chart of u(τj) and u(τj’) versus time (i.e. years) presents several crossing points, while on the other hand, the crossing period is unique and allows the approximate location of the starting point of the trend. Here the MK test was applied to raw data, without pre-whitening, according to Yue and Wang (2002).

3.5 Linear regression versus temperature RT

To evaluate the influence of climate warming upon atmospheric feeding of the hydrological cycle, we carry out a linear regression of precipitation data versus temperature data (RT). For shortness, and in view of the substantially consistent behavior of the different stations, as observed from preliminarily analysis and shown here forward, we use averaged values among the four reported stations. Again, in view of the relative shortness of the series, single regression analysis is carried out.

3.6 Principal components analysis, PCA

Several statistical grouping techniques, including PCA, are currently adopted for analysis, classification and regionalization of climatic and hydrologic variables (e.g. Baeriswyl and Rebetez 1997; Brinkmann 1999; Littmann 2000; Bocchiola and Rosso 2007a; Bocchiola et al. 2008), to identify underlying climatic patterns (e.g. Huth 2000, Alpert et al 2004; Michailidou et al. 2009a) to and highlight the most discriminating core variables (Clausen and Biggs 2000; Michailidou et al. 2009b), to avoid redundancy and maximize information content. Here, we apply PCA to the available set of variables analyzed in the previous Sections, using averaged values from the four reported stations.

4 Results

4.1 Precipitation

We consider here the yearly cumulated total precipitation Pcum and rainfall Rcum (e.g. Cislaghi et al. 2005). First, LR is carried out (as in e.g. Jiang et al. 2007) reported in Table 2. For the three lowest stations from Arno to Salarno, a positive trend of Pcum is observed, while a decreasing trend is observed at the highest Pantano station. The particularly high value at Arno station (10.00 mm year–1) may depend upon its comparatively lower altitude, 1,820 m asl. The related p-val are also reported in Table 2 (p-val ≥ 0.05 indicates non significant trend). During the period 1975–1980 or so the MW is below the LT, and either outside or narrowly inside the 5% confidence limits for the LT (not shown, see MW results in Table 2). The results of the traditional MK test are also reported in Table 2, and indicate non significant trends. Results of progressive MK test for Pcum are reported in Table 3, where a likely starting period for the trend is reported, together with the change in the average of Pcum before and after that period. However, here no significant change seems detected.
Table 2

Results of LR test, MW test and traditional MK test. Coeff. is coefficient (i.e. slope of the regression line). The X symbol indicates failure to match long-term average LT (p-val, α = 5%) for MW test. Stations in increasing order of altitude. At Salarno station no test could be carried for duration D, start B and end E of HS (explained in section Data base and treatment)

Station

Arno

Avio

Sal.

Pan.

Arno

Avio

Sal.

Pan.

 

LR

LR

LR

LR

MW

MW

MW

MW

     

MK

MK

MK

MK

Pcum Coeff. [mm year –1]/MW

10.00a

1.72

2.33

–3.30

X

X

Pcump-val [.] LR/MK

5E-02a

6E-01

6E-01

4E-01

7E-02

8E-01

3E-01

6E-01

Rcum Coeff. [mm year –1] /MW

11.42a

4.40

3.31

5.83

X

X

-

X

Rcump-val [.]LR/MK

2E-02a

1E-01

3E-01

7E-02

2E-02a

2E-01

1E-01

4E-02a

HSav Coeff. [cm year–1] /MW

–0.70

–0.79a

–1.14a

–1.11

X

-

X

-

HSavp-val [.] LR/MK

8E-02

2E-02a

1E-02a

7E-02

5E-02a

4E-03a

4E-04a

3E-05a

NS Coeff. [dd year–1] /MW

–0.56a

–0.32a

–0.81a

–0.73a

X

X

X

X

RL NS p-val [.] LR/MK

1E-04a

3E-03a

1E-04a

2E-04a

2E-04a

5E-03a

9E-04a

3E-04a

B0 Coeff. [dd year –1] /MW

0.00

0.33

NC

0.02

-

-

NC

-

B0 p-val [.] LR/MK

1.E-01

1.E-01

NC

9.E-01

3E-01

1E-01

NC

4E-01

B5 Coeff. [dd year –1] /MW

0.16

0.31

NC

0.11

-

-

NC

-

B5 p-val [.]LR/MK

5E-01

1E-01

NC

7E-01

5E-01

1E-01

NC

3E-01

B10 Coeff. [dd year –1] /MW

0.15

0.63a

NC

0.04

-

-

NC

-

B10 p-val [.] LR/MK

5E-01

2E-02a

NC

9E-01

6E-01

1E-02a

NC

4E-01

B20 Coeff. [dd year –1] /MW

0.91

0.39

NC

0.47

-

-

NC

-

B20 p-val [.] LR/MK

3E-01

1E-01

NC

1E-01

1E + 00

2E-01

NC

9E-02

E0 Coeff. [dd year –1] /MW

0.00

–0.53a

NC

–1.90

-

-

NC

-

E0 p-val [.] LR/MK

7E-01

5E-03a

NC

2E-01

4E-03

8E-01

NC

2E-05a

E5 Coeff. [dd year –1] /MW

–1.51

0.49

NC

–1.50

X

X

NC

X

E5 p-val [.] LR/MK

6E-04a

6E-01

NC

5E-03a

1E-03a

5E-04a

NC

3E-05a

E10 Coeff. [dd year –1] /MW

–1.47a

–1.41a

NC

–1.74a

X

X

NC

X

E10 p-val [.] LR/MK

2E-03a

6E-03a

NC

9E-04a

2E-03a

1E-02a

NC

2E-05

E20 Coeff. [dd year –1] /MW

–1.04a

–0.32

NC

–1.48a

X

-

NC

X

E20 p-val [.] LR/MK

2E-02a

4E-01

NC

8E-03a

1E-02a

9E-02

NC

2E-03a

D0 Coeff. [dd year –1] /MW

0.00

–0.80

NC

–1.11

-

-

NC

-

D0 p-val [.] LR/MK

3E-01

4E-03a

NC

6E-02

5E-03a

7E-01

NC

3E-02a

D5 Coeff. [dd year –1] /MW

–1.74

–1.06a

NC

–1.62a

X

X

NC

X

D5 p-val [.] LR/MK

3E-03

3E-02a

NC

3E-03a

1E-02a

2E-03aa

NC

1E-03a

D10 Coeff. [dd year –1] /MW

–1.68a

–2.11

NC

–1.69

X

X

NC

X

D10 p-val [.] LR/MK

3E-03

3E-03

NC

3E-03a

4E-03a

2E-03a

NC

3E-05a

D20 Coeff. [dd year –1] /MW

–0.86

–0.72

NC

–1.94

X

-

NC

X

D20 p-val [.] LR/MK

2E-01

1E-01

NC

3E-03a

1E-01

2E-01

NC

8E-02

Year Tmin,av Coeff. [°C year –1] /MW

0.01

0.02a

0.08a

0.06a

-

X

X

X

Year Tmin,av p-val [.] LR/MK

7E-01

5E-02a

1E-04a

1E-04a

9E-01

1E-01

3E-07a

3E-05a

Year Tmax,av Coeff. [°C year –1] /MW

0.10a

0.02a

0.09a

0.08a

X

X

X

X

Year Tmax,avp-val [.] LR/MK

1E-04a

4E-02a

1E-04a

1E-04a

3E-06a

3E-02a

8E-08a

2E-06a

Spring Tmin,av Coeff. [°C year –1] /MW

0.12a

0.04a

0.20a

0.11a

X

X

X

X

Spring Tmin,avp-val [.]LR/MK

3E-04a

4E-03a

1E-04a

1E-04a

3E-04a

1E-02a

3E-07a

2E-05a

Spring Tmax,av Coeff. [°C year –1] /MW

0.22a

0.05a

0.20a

0.14a

X

X

X

X

Spring Tmax,avp-val [.] LR/MK

1E-04a

4E-03a

1E-04a

1E-04a

9E-09a

8E-03a

4E-07a

4E-07a

Fall Tmin,av Coeff. [°C year –1] /MW

0.01

0.00

0.02

0.02

-

-

-

X

Fall Tmin,avp-val [.] LR/MK

5E-01

9E-01

2E-01

3E-01

9E-01

4E-01

2E-01

3E-01

Fall Tmax,av Coeff. [°C year –1] /MW

0.07a

–0.01

0.03

0.03

X

-

-

X

Fall Tmax,avp-val [.] LR/MK

4E-04a

4E-01

1E-01

8E-02

4E-04a

3E-01

1E-01

9E-02

aSignificant p-val (α = 5%)

NC not calculated

Table 3

Average values before and after the start of the trends as deduced from progressive MK test, given as anomalies (absolute values and percentage) with respect to the long-term average LT. Stations in increasing order of altitude

Station

Arno

Avio

Sal.

Pan.

Pcum LT[mm]

1,465.92

1,305.83

1,418.35

1,352.02

Pcum start [year]

1999

1975

1975

1985

Pcum before [%]

–0.51

–5.36

–0.63

–2.85

Pcum before [mm]

–7.48

–69.94

–8.97

–38.58

Pcum after [%]

+2.04

+1.75

+0.18

+1.66

Pcum after [mm]

+29.93

+22.8

+2.60

+22.41

Rcum LT[mm]

1,118.28a

920.93

1,014.62

909.55a

Rcum start [year]

1980a

1979

1980

1990a

Rcum before [%]

–14.75a

–2.99

–8.27

–9.06a

Rcum before [mm]

–164.89a

–27.58

–83.93

–82.41a

Rcum after [%]

+8.51a

+1.55

+4.77

+14.16a

Rcum after [mm]

+95.13a

+14.3

+48.42

+128.76a

HSav LT [cm]

52.45a

54.72a

57.00a

95.28a

HSav start [year]

1991a

1988a

1986a

1986a

HSav before [%]

+13.87a

+18.70a

+26.24a

+17.85a

HSav before [cm]

+7.28a

+10.23a

+14.96a

+17.01a

HSav after [%]

–12.13a

–12.51a

–14.96a

–17.01a

HSav after [cm]

–23.12a

–22.86a

–26.24a

–17.85a

Tmin,av LT [°]

–0.80

–3.44

–1.36a

–6.40a

Tmin,av start [Year]

1997

1988

1987a

1990a

Tmin,av before [%]

–8.75

–2.17

–63.8a

–3.18a

Tmin,av before [°]

–0.87

–3.52

–2.13a

–6.61a

Tmin,av after [%]

+32.5

+10.64

+56.9a

+4.77a

Tmin,av after [°]

+0.54

+3.08

+0.56a

+6.10a

Tmax,av LT [°]

9.49a

4.86a

6.74a

1.65a

Tmax,av start [Year]

1979a

1983a

1984a

1987a

Tmax,av before [%]

–14.6a

–10.73a

–14.6a

–46.61a

Tmax,av before [°]

–8.10a

–4.34a

–5.69a

–0.88a

Tmax,av after [%]

+6.8a

+7.93a

+6.85a

+51.52a

Tmax,av after [°]

+10.14a

+5.24a

+7.65a

+2.50a

aSignificant anomalies (MK test, Table 2)

An increasing trend of Rcum is detected by the slope of the regression lines in Table 2. Only Arno station presents a significant p-val, as well as a particularly high slope, but for a higher p-val (e.g. α = 10%), significant regression would be found at all stations but Salarno. The MW test of Rcum is reported in Fig. 2 and Table 2. Hereon, for shortness, we report the charts for the lowest and highest stations of Arno and Pantano, while the remaining stations display in practice a similar behavior.
https://static-content.springer.com/image/art%3A10.1007%2Fs00704-009-0186-x/MediaObjects/704_2009_186_Fig2_HTML.gif
Fig. 2

Moving window (MW) average of 10 years for Rcum, Arno and Pantano

During 1975–1985 or so, the MW is below the LT, but an increase is observed starting from the 90s with values outside the upper confidence limits for the period 2000–2004. The traditional MK test (in Table 2) indicates significant non stationarity for Arno and Pantano, and a less significant (p-val in the order of 10% or so) trend for Avio and Salarno. Progressive MK test for Rcum is reported in Fig. 3 (and Table 3). A critical period may be located in the window 1980–1990.
https://static-content.springer.com/image/art%3A10.1007%2Fs00704-009-0186-x/MediaObjects/704_2009_186_Fig3_HTML.gif
Fig. 3

Progressive MK test for Rcum, Arno and Pantano

4.2 Snow depth

The LR on HSav (Table 2) shows a decrease in time, with a variable slope, from –0.70 cm year–1 for Arno station to –1.14 cm year–1 for Salarno. The MW test for HSav in Fig. 4, indicates that until 1985 or so the expected snow cover depth was greater than the LT, while from 1990 or so, a considerable decrease is observed. MK test indicates significant decrease at all stations, and the progressive form in Fig. 5 indicates a critical window during 1985–1990, and noticeable anomalies of HSav before and after.
https://static-content.springer.com/image/art%3A10.1007%2Fs00704-009-0186-x/MediaObjects/704_2009_186_Fig4_HTML.gif
Fig. 4

Moving window (MW) average of 10 years for HSav, Arno and Pantano

https://static-content.springer.com/image/art%3A10.1007%2Fs00704-009-0186-x/MediaObjects/704_2009_186_Fig5_HTML.gif
Fig. 5

Progressive MK test for HSav, Arno and Pantano

The number of yearly snowfalls, NS is also significantly decreasing (Table 2), at an average rate of Coeff = –0.6 snowfall days per year. Particularly, Salarno and Pantano stations (with the highest altitudes of 2,070 and 2,300 masl) display the greatest decrease. Both MW (Fig. 6) and progressive MK (Fig. 7) tests, indicate a starting trend between 1985 and 1990, depending upon altitude. We then study beginning, ending and duration of snow cover for different thresholds, namely HS > 0, 5, 10 and 20 cm. Days of the year are numbered starting from October 1st. The tests LR, MW and MK were applied to first and last day of snow cover over a given threshold (B, E), and duration therein (D = EB). This analysis was not carried out for Salarno station, because in the most recent years some lack of data was observed during the melting period, so giving less accurate results (however, qualitative screening indicated a similar trend as for the other stations). The LR (Table 2) shows in practice a (non significant) delay of snow cover start B, but there is instead a significant anticipation of snow cover melting, as indicated by a decrease of E for the highest thresholds (i.e. HS > 10, 20 cm). Consistently, duration D decreases. The MW and progressive MK tests (not reported here) display a decrease of E and D, starting from the 90s, more relevant for 5, 10 and 20 cm. The traditional MK test indicates a significant anticipation in ablation and a consequent decrease in duration of the snow cover. For all the tests, the intermediate thresholds (i.e. 5 and 10 cm) display the most pronounced trend.
https://static-content.springer.com/image/art%3A10.1007%2Fs00704-009-0186-x/MediaObjects/704_2009_186_Fig6_HTML.gif
Fig. 6

Moving window (MW) average of 10 years for NS, Arno and Pantano

https://static-content.springer.com/image/art%3A10.1007%2Fs00704-009-0186-x/MediaObjects/704_2009_186_Fig7_HTML.gif
Fig. 7

Progressive MK test for NS, Arno and Pantano

4.3 Temperature

First, we analyze the yearly average values of maximum and minimum daily temperature, Tmax,av and Tmin,av. The LR test displays in practice a significant increase of both, with a slightly greater trend of the former. The MW test for Tmin,av and Tmax,av (not shown here) indicates a moderately increasing trend until 1990 or so, followed by a greater rate of increase thereafter. The MK test indicates significant increase of both Tmin,av and Tmax,av. The progressive MK test (not reported) evidences an increasing trend, with a visible start during the first half of the 80s for Tmax,av, and more recently for Tmin,av, in the decade 1988–1997 (see Table 3). This trend is particularly visible for the highest Salarno and Pantano, and confirmed by LR. We then analyze the average values of daily Tmin and Tmax during fall and spring. The LR indicates that while no significant change is observed in Fall, a significant increase is observed of Tmin,av and Tmax,av spring, i.e. at thaw start, with a rate of increase up to Coeff = +0.2°C year–1 or so. The MW and MK test indicate increase of Tax,av and Tmin,av spring. We report here the MW and MK progressive tests for Tmax,av spring (Figs. 8 and 9), most significant for temperatures, and indicating a clear speeding up of the observed trend during the early 90s. No clear trend is observed in fall temperatures (not shown).
https://static-content.springer.com/image/art%3A10.1007%2Fs00704-009-0186-x/MediaObjects/704_2009_186_Fig8_HTML.gif
Fig. 8

Moving window (MW) average of 10 years for Tmax,av spring, Arno and Pantano

https://static-content.springer.com/image/art%3A10.1007%2Fs00704-009-0186-x/MediaObjects/704_2009_186_Fig9_HTML.gif
Fig. 9

Progressive MK test for Tmax,av spring, Arno and Pantano

4.4 Snow water equivalent

In Table 4, we report the results from LR test on SWE5,A for the five measurement dates. We evaluate the dependence of the rate of decrease of SWE5,i versus altitude. This is reported in Fig. 10, where the absolute value of Coeffi in each stations i (always negative) is reported versus altitude Ai for the dates from April 1st to June 1st. No significant trend is visible for March 1st, so this date was not reported in the chart. For all the dates, an almost constant value of coeffi is seen until 2,000 m asl or so, while a noticeable increase with altitude is observed thereafter. Weighted (on estimation variance) regression of Coeffi versus Ai is carried out, and the results are also reported in Table 4. With the sole exception of March 1st, the loss in SWE in a given site i, SWE5,i noticeably depends upon altitude Ai. Therefore, a more pronounced loss is expected in terms of SWE for the highest stations. The trends become stronger as thaw proceeds, as seen from the increase (in absolute value) of Coeff in Table 4. The MW and progressive MK tests upon SWE5,A are reported in Fig. 11 for April 1st, the most widely adopted date for evaluation of SWE accumulation for water resources assessment (e.g. Ranzi et al. 1999; Bohr and Aguado 2001). The MW falls outside the confidence limits of the LT before 1989 (higher values) and narrowly inside after 1994 (lower values). A pronounced speed up of the decrease process of SWE5,A is observed since 1990 or so (similar results were found for the other four dates), and the traditional MK test indicates again non stationarity of SWE5,A for all the five dates.
Table 4

LR of SWE5,A in time and of SWE5 versus altitude, MW and MK test on SWE5,A. Symbol X indicates failure to match long-term average LT (p-val, α = 5%). Average values of SWE5,A before and after the start of the trends as deduced from progressive MK test reported, given as anomalies (absolute values and percentage) with respect to the long-term average LT

Date

Mar. 1st

Apr. 1st

Apr. 15th

May 1st

June 1

LR SWE5,A Coeff. [cm year–1]

–0.77a

–1.08a

–1.04a

–1.1a

–0.54a

LR SWE5,Ap-val [.]

2E-02a

2E-03a

5E-03a

5E-04a

3E-03a

LR SWE5,i vs. Ai CoeffA [cm year-1km–1]

–0.17

–0.76a

–1.21a

–1.60a

–1.64a

LR SWE5,i vs. Aip-val [.]

5E-01

4E-03a

1E-04a

2E-05a

1E-06a

MW SWE5,A

X

X

X

X

X

MK SWE p-val [.]

4E-04a

5E-05a

3E-04a

5E-06a

8E-05a

SWE5,A LT [m]

0.41

0.47

0.48

0.39

0.14

SWE5,A start [year]

1990

1990

1990

1990

1990

SWE5,A before [m]

0.07b

0.10b

0.10b

0.11b

0.05b

SWE5,A Before [%]

+15%b

+21%b

+21%b

+26%b

+36%b

SWE5,A after [m]

–0.08b

–0.14b

–0.15b

–0.14b

–0.07b

SWE5,A after [%]

–22%b

–30%b

–31%b

–38%b

–50%b

aSignificant p-val (α = 5%)

bSignificant anomalies according to MK test

https://static-content.springer.com/image/art%3A10.1007%2Fs00704-009-0186-x/MediaObjects/704_2009_186_Fig10_HTML.gif
Fig. 10

Rate of decrease of SWE5,i versus altitude

https://static-content.springer.com/image/art%3A10.1007%2Fs00704-009-0186-x/MediaObjects/704_2009_186_Fig11_HTML.gif
Fig. 11

Moving window (MW) average of 10 years and progressive MK test for SWE5,A on April 1

4.5 Regression versus temperatures, RT

Total precipitation Pcum seems not significantly correlated to temperature. However, Rcum seems significantly (positively) correlated to Tmax,av and Tmin,av, indicating increase of rainfall precipitation in response to increase in temperature. This is consistent with the significant decrease of snow cover depth HSav. Particularly, increase of fall (Tmax,av) and spring (Tmax,av, Tmin,av) temperatures lead to thinner snow cover. Notice also that p-val for regression of HSav vs. average yearly temperatures Tmax,av and Tmin,av are in fact quite close to the significance threshold. The number of snowfall events NS also decreases significantly with an increase of temperature. The beginning of the snowfall season (i.e. B0, B5, B10, B20) is significantly correlated to temperatures during fall (Tmin,av, Tmax,av fall), and the higher the latter the later considerable snow cover is observed. Concerning the depletion date, for the intermediate thresholds (E5, E10), a significant correlation is found versus spring temperatures (Tmin,av, Tmax,av spring). Full snow depletion (i.e. E0) seems negatively correlated to fall temperatures (Tmax,av fall, and very close to the significance threshold, Tmin,av fall). Because increasing (especially maximum) fall temperatures also imply delayed snow cover initiation (i.e. increased B0) and thinner snow crust (i.e. decrease of HSav), warmer falls lead to shorter snow-crust duration (as confirmed by significant dependence of D upon fall temperatures). Duration of snow cover for intermediate thresholds (D5, D10) seems considerably controlled by spring temperatures (Tmin,av, Tmax,av spring), much like the depletion date E5 and E10. Analysis of SWE5,A indicates that while spring temperatures (Tmin,av spring) is strongly impacting stored water resources at fixed dates, non negligible control on SWE availability is given by Tmax,av during the fall, and the signature of a warmer fall period is observed in decreased available SWE in spring.
Table 5

Results of RT test. Coeff. is coefficient (i.e. slope of the regression line)

Variables Y/X

Tmin,av Spring [°C]

Tmax,av

Spring [°C]

Tmin,av

all [°C]

Tmax,av Fall [°C]

Tmin,av [°C]

Tmax,av [°C]

Pcum Coeff. [mm °C-1]

10.56

10.72

–31.21

20.07

62.55

56.20

Pcump-val [.]

6.E-01

5.E-01

6.E-01

6.E-01

2.E-01

1.E-01

Rcum Coeff. [mm °C –1]

32.55

29.37

–47.81

16.42

122.61a

107.27a

Rcump-val [.]

1.E-01

5.E-02

3.E-01

6.E-01

7.E-03a

6.E-04a

HSav Coeff. [cm °C –1]

–5.69a

–3.91a

–1.66

–9.25a

–9.73

–6.99

HSavp-val [.]

1.E-02a

3.E-02a

8.E-01

2.E-02a

8.E-02

7.E-02

NS Coeff. [dd °C –1]

–3.73a

–2.81a

–1.63

–4.31a

–6.77a

–5.76a

NS p-val [.]

1.E-04a

1.E-04a

5.E-01

5.E-03a

8.E-04a

1.E-04a

B0 Coeff. [dd X–1]

0.90

0.58

6.49a

5.73a

–0.34

–0.38

B0 p-val [.]

4.E-01

5.E-01

2.E-02a

3.E-03a

9.E-01

8.E-01

B5 Coeff. [dd °C –1]

0.15

0.01

6.94a

6.40a

–3.08

–1.37

B5 p-val [.]

9.E-01

1.E + 00

9.E-03a

7.E-04a

3.E-01

5.E-01

B10 Coeff. [dd °C –1]

0.08

0.12

6.64a

5.56a

–2.83

–1.35

B10 p-val [.]

9.E-01

9.E-01

5.E-03a

9.E-04a

2.E-01

4.E-01

B20 Coeff. [dd °C –1]

2.35

1.65

9.93

10.26a

3.75

3.13

B20 p-val [.]

3.E-01

4.E-01

9.E-02

1.E-02a

5.E-01

4.E-01

E0 Coeff. [dd °C –1]

–1.09

–2.10

–14.19

–10.95a

–4.12

–7.34

E0 p-val [.]

7.E-01

4.E-01

6.E-02

5.E-02a

6.E-01

2.E-01

E5 Coeff. [dd °C –1]

–7.68a

–6.04a

3.91

–3.39

–13.68a

–13.08a

E5 p-val [.]

1.E-04a

1.E-04a

5.E-01

4.E-01

6.E-03a

1.E-04a

E10 Coeff. [dd °C –1]

–8.90a

–6.73a

2.71

–4.97

–14.75a

–12.46a

E10 p-val [.]

1.E-04a

1.E-04a

6.E-01

2.E-01

6.E-03a

7.E-04a

E20 Coeff. [dd °C –1]

–1.13

–1.27

7.75

3.23

–3.69

–5.45

E20 p-val [.]

7.E-01

5.E-01

2.E-01

5.E-01

5.E-01

2.E-01

D0 Coeff. [dd °C –1]

–3.32

–2.80

–12.02a

–10.60a

–4.15

–4.65

D0 p-val [.]

9.E-02

6.E-02

9.E-03a

1.E-03a

4.E-01

2.E-01

D5 Coeff. [dd °C –1]

–7.83a

–6.05a

–3.03

–9.79a

–10.60

–11.71a

D5 p-val [.]

1.E-03a

1.E-03a

6.E-01

3.E-02a

8.E-02

4.E-03a

D10 Coeff. [dd °C –1]

–8.38a

–6.39a

–12.04

–13.27a

–9.21

–13.17a

D10 p-val [.]

2.E-02a

2.E-02a

2.E-01

3.E-02a

3.E-01

2.E-02a

D20 Coeff. [dd °C –1]

–3.20

–2.96

0.70

–3.16

–18.62a

–16.66a

D20 p-val [.]

4.E-01

3.E-01

9.E-01

6.E-01

2.E-02a

3.E-03a

SWE5,A Mar. 1st Coeff. [m °C –1]

–0.04a

–0.02

–0.02

–0.06

–0.07

–0.03

SWE5,A Mar. 1st p-val [.]

5.E-02a

2.E-01

7.E-01

8.E-02

2.E-01

3.E-01

SWE5,A Apr. 1st Coeff. [m °C –1]

–0.06a

–0.03

–0.06

–0.08a

–0.09

–0.04

SWE5,A Apr. 1st p-val [.]

2.E-03a

6.E-02

2.E-01

2.E-02a

8.E-02

2.E-01

SWE5,A Apr. 15th Coeff. [m °C –1]

–0.05a

–0.02

0.01

–0.04

–0.08

–0.02

SWE5,A Apr. 15th p-val [.]

2.E-02a

2.E-01

8.E-01

3.E-01

1.E-01

5.E-01

SWE5,A May 1st Coeff. [m °C –1]

–0.06a

–0.03a

–0.03

–0.08a

–0.09a

–0.04

SWE5,A May 1st p-val [.]

6.E-04a

4.E-02a

5.E-01

2.E-02a

4.E-02a

2.E-01

SWE5,A June 1st Coeff. [m °C –1]

–0.03a

–0.01

–0.03

–0.04a

–0.06a

–0.02

SWE5,A June 1st p-val [.]

6.E-04a

5.E-02

2.E-01

5.E-02a

2.E-02a

1.E-01

aSignificant p-val (α = 5%)

4.6 Principal components analysis, PCA

The results of PCA tests are summarized in Fig. 12. Therein, grouping is reported based upon the first five principal components (PCs), giving a cumulated variability of 78% (not shown for shortness), chosen since the original variables have their greatest (in absolute value) loading, load, therein (for none of the remaining PCs load > 0.4). Figure 12 shows a loading chart of the first two PCs (cumulated variability, 55%). We found 13 out of 23 variables with their greatest load upon PC1, (for most load > 0.8 or so, 38% of total variability), while 7 variables have their greatest load on PC2 (18% of total variability). In the figure, five groups of variables (one for each PC) are shown, each representing a climatic pattern underlying the observed data base. The first group includes variables with highest load on PC1. Positive load on PC1 are borne by snow cover related variables (label 1a), i.e. NS, HSav, and SWE5,A, together with depletion timing E5 and E10. Negative load upon PC1 are instead observed (label 1b) for spring temperatures, Tmax,av spring and Tmin,av spring, and less for yearly temperatures Tmax,av and Tmin,av. This PC represents the winter dynamics of snow cover and its persistence in spring depending upon temperature. The second group of variables has highest loadings on PC2, i.e. snow cover initiation B and fall temperatures, Tmax,av fall and Tmin,av fall. These PC depicts the effect of climate upon noticeable snow cover initiation. Less clear is the meaning of PC3 (10% of total variability). No variable has a highest load there, and the variables most correlated are Pcum load = 0.57), Tmax,av (load = 0.52) and Rcum (load = 0.48). Considering only load > 0.5, we propose a group made of Pcum and Tmax,av (that may include Rcum), which explains the influence of yearly average temperature upon total yearly precipitation (and rainfall, since in Table 5 there is significant correlation between Rcum and Tmax,av). The only variables with highest loadings on PC4 (7% of total variability) are Pcum and Rcum (load = 0.61, and load = 0.53, respectively). Because PCs are independent from each other by definition, this component much likely explains the degree of variability of total (and less of liquid) precipitation in this area given from external drivers of general circulation. Eventually, PC5 (5% of the total variability) is positively correlated with E0 (load = 0.71), and weakly negatively (load = –0.47) correlated with fall temperature Tmin,av Fall. RT analysis indicates significant negative dependence of E0 upon Tmax,av fall, and upon Tmin,av fall with confidence level α = 6%. Therefore, E0 is a representative variable, indicating the timing of full snow depletion and signature of fall temperatures therein. It seems difficult to cluster E20 in any group of variables, as it does not show a reasonably high correlation with any of the PCs (i.e. load >0.5 or so). However, it shows the two highest loadings on PC2 and PC4 (load = 0.49, –0.48, respectively), resulting in diagonal symmetry versus Rcum with respect to PC2 and PC4 (load = –0.52, 0.53, respectively). This may suggest weak ability of rainfall to reduce duration of thick snow crust. Notice also that E20 seems to not be considerably linked to temperatures (Table 5), and may be more weakly related to other climatic variables, thus possessing less informative content for the purpose of climate change studies.
https://static-content.springer.com/image/art%3A10.1007%2Fs00704-009-0186-x/MediaObjects/704_2009_186_Fig12_HTML.gif
Fig. 12

Loadings upon first two principal components. Grouping is indicated, reported in text. Dashed lines indicate grouping based on principal components from 3 to 5, thus not visible in the chart

5 Recent changes in the Adamello glacier area and comparison with three nearby Alpine glaciers

The recent (1983–2003) area variation of Adamello glaciers were investigated by Maragno et al. (2009) who found a decrease from 28.28 km2 to 22.99 km2, i.e. –19%. Considering the three greatest (area > 1 km2) glaciers of Adamello group (see Fig. 1)—namely Adamello (Mandrone plus Pian di Neve glaciers), Pisgana and the debris covered Venerocolo—they underwent considerable reduction in area (18.85–16.66 km2, i.e. –12%, 4.24–3.44 km2, i.e. –19%, and 1.68–1.25 km2, i.e. –25% respectively) during 1983–2003, and generally the smaller the glacier the greater the percentage of area loss (in agreement with e.g. Paul et al. 2004, who analysed changes in area of Swiss glaciers). Loss of mass (water equivalent) is estimated in –0.61, –0.46 and –0.39 m year–1 respectively for these three of the greatest glaciers, and –0.60 m year–1 on average upon the whole Adamello group, during 1932–2003 (CGI 1914–1977, 1978–2005).

Concerning the terminus fluctuations of Adamello glaciers, we refer to the CGI data base (CGI 1914–1977, 1978–2005), reporting data about the behaviour of several Italian glaciers. Most complete records for the Adamello group are those of Mandrone (North lobe of Adamello glacier, see Citterio et al. 2007a) and Venerocolo glaciers. CGI data base reports a 200 m retreat of Mandrone glaciers during 1953–2002 (i.e. –4.0 m year–1), with substantial stationarity during 1977–1983 and again a rapid decrease during 1987–2002 (i.e. –7.5 m year–1). The Venerocolo debris covered glacier displayed constant shortening during the period 1951–2002 (i.e. –3.9 m year–1), with a clear speeding up during 1982–2002 (–6.7 m year–1).

To compare the observed changes in the Adamello group with those experienced by other glaciers within the southern European Alps, we report here a qualitative comparison of our results with those from three glacierized areas along the Italian–Swiss border (see Fig. 1a). The first is the Sforzellina Glacier, SFO (46°20’N, 10°30’E, elevation ranging from 2,850 to 3,100 m asl, presently with area coverage of ca. 0.4 km2), a small, but yet widely investigated benchmark glacier within Valfurva valley, in Lombardy Region of Northern Italy (Catasta and Smiraglia 1993; Rossi et al. 2000; Diolaiuti et al. 2002; Citterio et al. 2007a; Cannone et al. 2008). Approximately 10 km SW of Sforzellina, is the greatest Italian valley glacier, named Forni, FOR (46°23’N, 10°35’E), ranging from 2,600 to 3,670 and showing an area coverage of ca. 12 km2 at present (Smiraglia 1989; Citterio et al. 2007b). The third glacier is Morteratsch MOR, in Engadine (46°24’N, 8°02’E, altitude ca. 2,000–4,000, presently area coverage ca. 16 km2) the largest glacier within the Bernina Massif, in the Northern slopes of Italian–Swiss Alps, also widely investigated (Ohlendorf et al. 1997; Oerlemans 2000; Klok and Oerlemans 2002; Klok 2004; Oerlemans 2007; Nemec et al. 2009).

Cannone et al. (2008), among others, investigated SFO evolution, local climate, and vegetation succession within the glacier’s forefield over the past three decades. They analyzed climate series of mean daily temperature Tmean and total precipitation Pcum for three weather stations (WSs: Uzza, 1,250 m asl, UZZ; Santa Caterina Valfurva, 1,730 m asl, SCV; Forni dam, 2,180 m asl, FOD), representative of climate conditions on SFO and FOR. They studied Tmean during the summer (June–August) finding a slight decrease during 1970–1980, as experienced by the whole Alpine region (Wood 1988), followed by an increase of about +0.5°C during 1988–2006 (Fig. 5 in Cannone et al. 2008). They also detected a decrease of Pcum in the same period (up to –10% at 2,180 m asl). However, they did not carry out any significance test. Citterio et al. (2007a) studied terminus fluctuations of a set of 98 Italian glaciers from 1913 to 2002, including SFO and FOR, versus meteo (Tmean, Pcum) data from a WS in Bormio (1,225 m asl), BOR, at the bottom of Valfurva Valley, for the period 1920–2006. They used MW (5 years) average of Tmean and evidenced a noticeable drop during the 1980s, followed by a substantial increase ever since 1990. Pcum is seen as substantially unchanged, albeit with some oscillation, ever since 1980 or so. Cannone et al. (2007) studied daily snow cover depth HS and snow cover duration D0 at Cancano WS (1,948 m asl, ca. 10 km N–W of Sforzellina glacier, CAN), for the period 1978–2003. They highlighted a decreasing trend of both, with a change in HS of approximately –1 cm year–1 (versus –0.79 cm year–1 as we found at Avio, 1,940 m asl), and of about –1.15 dd year–1 of continuous snow cover (versus –0.8 dd year–1 found at Avio). Cannone et al. (2008) argue that these climate variations caused strong glacier shrinkage of SFO, which considerably reduced in length. They reported an average retreat of approximately –2 m year–1 during 1971–2006, –2.6 m year–1 considering only 1985–1995, and an acceleration up to –5 m year–1 during 1996–2006. In spite of this reduction trend, SFO terminus fluctuations also showed a small glacier advance in the period 1975–1984 (+14.5 m equal to a rate of +1.5 m m year–1). This was followed by a transition phase, when the glacier alternated retreats and small advances, while, starting from 1992, the glacier shrinkage proceeded. Concerning volume changes, Cannone et al. (2008) reported an average loss of –1.1 m year–1 water equivalent during the period 1987–2006. Citterio et al. (2007a) report similar terminus fluctuations for FOR, albeit with different magnitude due to the larger glacier size (Fig. 6b therein). FOR experienced a terminus advance during 1971–1984 (+22 m year–1), and conversely underwent considerable retirement during 1987–2002 (–36 m year–1), with an average value of –24 m year–1 during 1926–2002. D’agata et al. (2002) report area variation of ablation tongue of FOR from 1.43 to 0.23 km2 during 1929–2004, with a mass loss of –0.96 m year–1, whereas the whole FOR passed from 17.80 km2 at the end of the Little Ice Age LIA (~1860) to 11.62 km2 in 2003 (–34.7%).

Morteratsch glacier, MOR, was studied among others by Nemec et al. (2009), who modelled glacier mass balance based upon thickness variation, temperature Tmean and precipitation Pcum at two nearby stations (Samedan Airport, 1,705 m asl, SAM; Segl, or Sils Maria, 1,802 m asl, SMA) for the period 1864–2005. They found a specific mass loss of approximately 44 m, with an almost continuous trend, with short periods of mass gain around 1920, 1935 and 1980, again due to a generally cooler period. Thereafter greater losses are found. They report that the decreasing specific mass balance is mainly associated with increasing summer temperature (compare Figs. 4 and 7 therein). Uehlinger et al. (2003) report a loss of area of about –15% (from 19.3 to 16.4 km2), as well as a decrease in length of –25% (from 8.9 to 6.7 km) since the end of LIA until 2000. Begert et al. (2005) investigated temperature and total precipitation series for 12 weather stations in Switzerland for the period 1864–2000. At SMA they found a significant increase of yearly mean of daily temperature Tmean (+6E–3 °C year–1) as well as of fall and spring temperature Tmean (September–November, +8E–3 °C year–1; March–May, +7E–3 °C year–1), the latter showing a more recent speeding up in the late 1980s, as shown by Mann Kendall progressive test (see Figs. 8 and 10 therein). They report that their findings concerning temperature in SMA well fit those observed in Northern Italy (see Brunetti et al. 2000, p. 75 therein). No significant trend is found for Pcum. Rebetez and Reinhard (2008) analysed temperature trends at SMA during 1975–2004, detecting a change in Tmean of +0.06°C year–1, while Tmean Spring (March–May) varied by +0.08°C year–1. Ohlendorf et al. (1997) compared glacier varves from glacial Lake Silvaplana versus a 127 year climate data series (summer rainfall and temperature, yearly snowfalls NS) from SMA to detect climate change effect on Morteratsch and Roseg Glaciers. For 1966–1990 they show a decrease of NS of about –0.64 dd year–1, very similar to our findings here. They also found significant inverse correlation between NS and ice melt at thaw, which they explain via decreased albedo from new snow. Seiz and Foppa (2007) report a –0.22 cm year –1 of HNcum (strongly linked to HS as reported) for SMA during 1864–2006, whereas Wüthrich et al. (2008) report that the 1864–2005 trend of days with snow pack (D0) reveals a significant decrease for a number of investigated stations in the Swiss Alps, including SMA.

In Table 6, we report numerical comparison between the three chosen benchmark glaciers, and the three greatest glaciers of the Adamello group. We use area and length variation for the reference available periods, together with variation of maximum or mean temperature during spring, more relevant for ice melting, and snow cover depth and duration (where available), for a reference station. Albeit qualitative, the comparison seems to support the robustness of our findings. These are consistent with the present evidence of speeding up of transient warming within the Southern Alpine area between Italy and Swiss, leading to decrease of snow precipitation and enhancement of mass loss from glaciers therein.
Table 6

Comparison with three reference glacierized areas. Glacier altitude is the last observed minimum altitude. TW is time window of statistics below. Percentage variation of glacier area and terminus length, Ag and Lt are reported with respect to initial measured values, Ag0 and Lt0. Representative stations for snow depth and precipitation are also reported, together with rate of variation within the time window. Here Avio station has comparable altitudes as Adamello (but see also Table 2). Pcum not reported as substantially unchanged

Site/Variable

ADA

PIS

VEN

Avio

FOR

SFO

FOD

CAN

MOR

SMA

Altitude [m asl]

2,650

2,565

2,560

1,940

2,600

2,850

2,180

1,948

2,000

1,802

TW [.]

1983

1983

1983

-

1860

1920

-

-

1860

-

2003

2003

2003

2003

2006

2000

-

A0 [km2]

18.85

4.24

1.68

-

17.80

0.56

-

-

19.30

-

Ag[%]

–11.6

–18.9

–0.253

-

–34.7

–57.4

-

-

–15

-

TW [.]

1953

1914

1920

-

1864

1926

-

-

1860

-

2002

2003

2003

2003

2003

2000

Lt0 [m]

5,200

3,072

2,302

-

6,595

1,404

-

-

8,900

-

Lt [%]

–3.8

–21.4

–7.8

-

–28.8

–29.2

-

-

–24.7

-

TW [.]

-

-

-

-

-

-

1988

-

-

1975

2006

-

-

2004

Tmean spring [°C year–1]

-

-

-

-

-

-

+0.03b

-

-

+0.08a

TW [.]

-

-

-

1966

-

-

-

-

-

-

2007

Tmax spring [°C year–1]

-

-

-

+0.05a

-

-

-

-

-

-

TW [.]

-

-

-

1966

-

-

-

-

-

1966

2007

1990

NS [dd year–1]

-

-

-

–0.32a

-

-

-

-

-

–0.64b

TW [.]

-

-

-

1966

-

-

-

1978

-

1864

2007

2003

-

2006

HS(HNcum) [cm year–1]

-

-

-

–0.79a

-

-

-

–1.00b

-

(–0.22)b

TW [.]

-

-

-

1966

-

-

-

1978

-

-

2007

2003

D0 [dd year-1]

-

-

-

–0.80a

-

-

-

–1.15b

-

-

aStatistically significant (when significance test available)

bFrom reference studies with unknown significance

6 Discussion and conclusions

Some indications arise from the present study upon the Adamello area and by comparison with nearby reference glacier areas within the southern Alps. The analysis of total precipitation Pcum seems to indicate in practice an unchanged input of atmospheric water, whereas the signal of rainfall Rcum is more indicative of an increasing trend, starting during 1980–1990. Snow cover depth HSav is decreasing with time, again starting since the early 1980s, and consistently within the southern Alps. Investigation of snowfall days NS indicates a widespread decreasing trend, of about –0.6 dd year–1, which for the period of 42 years here leads to 24 too few snowfall days, on average. Concerning snow-cover duration, the choice of different thresholds is relevant. While for thin snow cover (0 < HS ≤ 5) weak trends are observed, for 5, 10 and 20 cm, stronger evidence is seen. Above 2,000 m asl or so, depletion of snow cover shifts from about half May to about half April for the highest thresholds 5, 10 and 20 cm, while it seems more stationary for 0 cm, consistently with the present literature concerning the European Alps (e.g. Laternser and Schneebeli 2003).

Beniston (1997) studied HS and D for three principal stations in Swiss Alps for 1945–1994 and found that periods with low values of both are linked to persistent high surface pressure fields over the Alpine region during late fall and winter, accompanied by large positive temperature anomalies, correlated with anomalies of the North Atlantic Oscillation index, NAO. Scherrer et al. (2004) investigated snow cover duration D5 for 110 stations in Switzerland an found that control of NAO oscillation is more significant in southern than in northern (Swiss) Alps. Maragno et al. (2009) observed positive correlation (ρ = 0.52) between temperature and NAO oscillation index, and negative correlation (ρ = –0.51) between snow cover depth HS and NAO for three WSs nearby Adamello glaciers, including Avio here, for the winter period (JFM).

Consistently, we observed here a decrease of snowfalls, snow cover depth and duration, linked to increase of air temperature at thaw, particularly since the late 1980s, as well as considerable signature of fall temperatures upon snow cover. In Fig. 13, we report HSav and NAO Winter anomaly (JFM), together with estimated total area and volume of Adamello group glaciers for the period 1983–2003 (four dates, Maragno et al. 2009). A clear trend is visible for this period, with increasing NAO, decreasing HSav and decreasing glaciers’ size, in accordance with the context of existing knowledge upon climate change and glaciers’ dynamics within European Alps.
https://static-content.springer.com/image/art%3A10.1007%2Fs00704-009-0186-x/MediaObjects/704_2009_186_Fig13_HTML.gif
Fig. 13

HSav, NAO anomaly (JFM) and Adamello group’s glacier area and volume for the period 1983–2003

The value of winter stored SWE is decreasing within the Alps. Its rate of decrease (cm year–1) is augmented with altitude and as the melting season proceeds. Rohrer et al. (1994) investigated SWE from some 50 stations in the Swiss Alps for 1947–1992. They do not carry any statistical tests, but eyeball analysis indicates a noticeable decrease in SWE at April 1st and May 1st from 1985 onward (Fig. 5 therein). Because SWE is the potential amount of IWE to be stored every year within ice bodies (see e.g. Jansson et al. 2003), reduction of the former much likely lead to loss of mass by the latter. Yearly average temperatures (min and max) are significantly increasing, but a more pronounced trend is seen in spring than in fall. Increase in temperatures trades snowfall for rainfall and therefore carries a considerable bearing upon modified hydrology of cryosphere.

The PCA procedure allows the identification a set of most relevant variables, the observation of which should provide reasonable coverage of climatic variability. Besides tracking of temperatures, we stress the importance of total precipitation P (with splitting of R), snow cover HS and number of snowfalls NS, the latter also directly affecting glacier dynamics as reported. Concerning timing of snow cover, a most relevant threshold seems that of 10 cm, so that tracking of E10 and B10 (and therefore D10) provides considerable information. Tracking of SWE seems indeed warranted by sampling within the windows April 1st–15th, consistently with widespread use of that period as a boundary between accumulation and melting season in mountain areas worldwide (e.g. Ranzi et al. 1999; Bohr and Aguado 2001; Simaityte et al. 2008). Here, evaluation of time averaged values of SWE5,A indicates greatest values between April 1st and April 15th in the last three decades (not shown, reported in Bianchi Janetti et al. 2008), so corroborating this finding. Eventually, depletion date E0 should be tracked, of utmost importance also in ecosystem dynamics (e.g. Coughlan and Running 1997).

The results here refer to the very localized and yet very important area of the Adamello Glacier, the widest of Italy, but are likely paradigmatic for the southern European Alps, and may be used as a benchmark for future studies. Also, these findings will be useful for future hydrological and ecological conjectures based upon locally valid climate change scenarios. The relative shortness of the series suggest that care is taken in interpretation of the results and in their extrapolation for the future. Notwithstanding, the proposed results seem to indicate a reasonably clear trend of the past and, likely, for the future, and surely pave the way for future investigation concerning this area, and, in general, the Alpine area.

Acknowledgements

The present paper reports work carried out under the umbrella of the CARIPANDA project, funded by the CARIPLO foundation of Italy (http://www.parcoadamello.it/progetti/caripanda.htm) under the direction of the ADAMELLO Park authority, which is here acknowledged also for supporting with logistic aid. ENEL Produzione is acknowledged for providing snow and precipitation data from their stations and for helping with logistic aid. The research presented in the present paper was also partially supported by the European Community, through the EU projects AWARE (EC contract 012257), and IRASMOS (EC contract 018412).

Copyright information

© Springer-Verlag 2009