*3.2.1 Data sources and preprocessing*

Maps, aerial orthophotos and satellite images were analyzed to calculate the area and volume changes of Lys Glacier and changes in the area of the proglacial lakes. The available cartographic sources include large-scale maps from the Val d'Aosta region, produced in 1975 and 1991 at a nominal scale of 1:10000 (**Table 1**). The maps were available in digital form as rasters, projected in the UTM32N coordinate


#### **Table 1.**

*List of datasets used in this study and their usage; the acquisition date (whenever available) is reported as dd/mm/yyyy.*

system based on the ED50 datum; contour lines and elevation points were digitized as shapefiles, and a DEM was produced from the 1991 map with a pixel spacing of 10 m using ArcMap Topo to Raster utility.

Aerial orthophotos from 1988 to 2012 (**Table 1**) were obtained from Geoportale Nazionale (http://www.pcn.minambiente.it/mattm/); these have a pixel size of 0.5 m and were downloaded in the UTM32 coordinate system based on the WGS84 datum. Most images were acquired in late summer months (August and September; **Table 1**), with minimum snow cover, while the orthophotos from 1988 and 2012 were acquired in early summer (June and July) and show evidence of residual snow, which however does not affect the glacier tongue. In addition to the aerial orthophotos, we obtained a stereo pair acquired from the Pleiades satellite constellation from the European Space Agency. The stereo pair was imaged on 1 September 2014 and provided at level 1B processing stage; we further processed it to obtain a DEM and an orthorectified image using NASA's AMES stereo pipeline (ASP). Eleven ground control points were selected from those available from Val d'Aosta regional authority (http://geonavsct.partout.it/pub/GeoNavITG/monografie.asp) to improve the geolocation accuracy of the DEM through bundle adjustment in the software; the DEM was then produced using the semi-global matching algorithm available in ASP [33], and the raw multispectral images were projected onto the DEM for orthorectification. Both the Pleaides DEM and multispectral image have a final resolution of 2 m.

### *3.2.2 Calculation of area and volume changes*

Based on the available planar data (**Table 1**), the glacier and proglacial lake outlines were estimated over multiple years: in the 1975 and 1991 technical maps, they were already drawn on the map, while in the other datasets, they were manually digitized as part of this study. The spatial resolution of the orthophotos allowed us to clearly identify the outlines and to distinguish the debris-covered parts of the tongue from the proglacial areas. Before digitization, all available data were reprojected to the UTM32N coordinate system based on the WGS84 datum for consistency. We further evaluated the accuracy of the manual delineation using the buffer method proposed by Paul et al. [34], by allowing the glacier and lake outlines to grow and shrink with a buffer of 2 pixels.

The analysis of elevation and volume changes was based on the comparison of the DEMs from 1991 and 2014. Before performing this comparison, the DEMs were co-registered using the approach developed by Berthier et al. [35] and applied by Fugazza et al. [36]. In this approach, one DEM ('slave') is iteratively shifted with respect to a reference DEM ('master'), to minimize the standard deviation of elevation differences over stable areas located outside the glacier (σΔh). We selected the oldest DEM as the reference and resampled both DEMs to a common resolution of 10 m. By applying the co-registration, we obtained a residual σΔh of 4.18 m; the average elevation difference over stable areas was 7 m, which was subtracted from the reference DEM. Based on the elevation difference on the glacier surface, we then calculated the volume change as:

$$
\Delta \mathbf{V}\_{\text{m}} = \sum\_{\mathbf{k}=1}^{K} \mathbf{d}\_{\text{k}} \mathbf{A}\_{\text{k}} \tag{1}
$$

**173**

**3.3 Meteorological data**

*Variations of Lys Glacier (Monte Rosa Massif, Italy) from the Little Ice Age to the Present…*

<sup>σ</sup>∆h\_cor = √

leads to larger errors [38], particularly with the oldest DEM from 1991.

on the correlation length (i.e. the distance at which two pixels are correlated). We calculated the correlation length using an empirical semivariogram in R software

\_\_\_\_\_\_\_\_\_\_\_

\_ Acor 5 ∙ A1991

σ∆h 2 ∙

where A1991 is the glacier area in 1991. The uncertainty calculation can in our case only be applied to the glacier tongue, as at altitudes above the glacier equilibrium line the presence of snow with little contrast deteriorates DEM reconstruction and

To characterize the evolution of debris cover on Lys Glacier, we adopted the methodology proposed by Azzoni et al. [9] for glaciers in the Ortles-Cevedale Group. Supraglacial debris was mapped by employing a maximum likelihood (ML) supervised classification approach. Four classes were included in the classification: debris, ice, snow and shadow. Although we did not aim at mapping the other surface types, a higher number of classes was chosen to permit an improved mapping of debris cover; the shadow class was also included as some images are affected by topographic shadow; this occurred mostly in the accumulation basins except for the orthophoto from 2003, where shadows occur over the whole glacier; thus, this orthophoto was excluded from the analysis (**Table 1**). We also excluded the maps from 1975 and 1991, as the information on the presence of debris was not available. For each of the satellite/aerial orthophotos, we independently selected 10 to 15 training areas. To estimate the accuracy of the classification approach, we performed a validation test on the 2006 and 2012 orthophotos, by selecting 100 random points for each test and comparing the results of ML classification against those from a manual classification. The 2 years were chosen because they represent best (2006) and worst (2012) conditions for debris cover mapping, in terms of image quality, presence of snow and shadows. For both years, we calculated overall accuracy (the ratio of the number of correctly classified points to the number of total points), as well as producer's accuracy (PA) and user's accuracy (UA), for the different classes. PA denotes the ratio of the number of correctly classified points to the total (manually classified) point for a class and is the complement of omission errors; UA denotes the ratio of the number of correctly classified points to the number of predicted points for a class and is the complement of commission errors.

In order to interpret the glaciological data of the study area, we analyzed meteorological variables recorded by the weather station installed in 1927 at Gressoney d'Ejola, 1850 m a.s.l., at a distance of 3.7 km from Lys Glacier (**Figure 1**). This long dataset allows investigating the recent climate behaviour, which influences the evolution of the glacier. At this weather station, manual observations of air temperature, liquid precipitation, total snow depth and thickness of fresh snow, atmospheric pressure, relative humidity, wind speed and velocity and cloud type and cover were collected three times a day (at 8 am, 2 pm and 7 pm) from 1927 to 2012. The data record is almost uninterrupted except for a short period from 1962 to 1970, when the station was temporarily moved downvalley, at 1730 m a.s.l. in the Orsia village. In 2002, an automatic weather station was installed at the same site, and it replaced the manual weather station after its dismissal in 2012. The automatic

. The uncertainty in volume

(2)

*DOI: http://dx.doi.org/10.5772/intechopen.91202*

change σ∆h\_cor is then expressed as:

*3.2.3 Mapping debris cover and its changes*

as 200 m and the effective correlated area as 0.12 km<sup>2</sup>

where k = 1 … K are the pixels of the reference surface, Ak is the area of each pixel and dk is the pixelwise difference between the 1991 and 2014 DEMs.

To express the uncertainty of volume changes, we used the approach described by Fischer et al. [37]: the uncertainty in elevation change (σ∆h) is not considered as entirely correlated but scaled to account for the effective correlated area Acor, based *Variations of Lys Glacier (Monte Rosa Massif, Italy) from the Little Ice Age to the Present… DOI: http://dx.doi.org/10.5772/intechopen.91202*

on the correlation length (i.e. the distance at which two pixels are correlated). We calculated the correlation length using an empirical semivariogram in R software as 200 m and the effective correlated area as 0.12 km<sup>2</sup> . The uncertainty in volume change σ∆h\_cor is then expressed as:

related area as 0.12 km $^{4}$ . The uncertainty in volume 2 ed as:

$$
\sigma\_{\Delta \text{h\\_cor}} = \sqrt{\sigma\_{\Delta \text{h}}^{2} \cdot \frac{\text{A}\_{\text{cor}}}{\text{5} \cdot \text{A}\_{\text{1991}}}} \tag{2}
$$

where A1991 is the glacier area in 1991. The uncertainty calculation can in our case only be applied to the glacier tongue, as at altitudes above the glacier equilibrium line the presence of snow with little contrast deteriorates DEM reconstruction and leads to larger errors [38], particularly with the oldest DEM from 1991.

#### *3.2.3 Mapping debris cover and its changes*

*Glaciers and the Polar Environment*

final resolution of 2 m.

*3.2.2 Calculation of area and volume changes*

to grow and shrink with a buffer of 2 pixels.

calculated the volume change as:

10 m using ArcMap Topo to Raster utility.

system based on the ED50 datum; contour lines and elevation points were digitized as shapefiles, and a DEM was produced from the 1991 map with a pixel spacing of

Aerial orthophotos from 1988 to 2012 (**Table 1**) were obtained from Geoportale Nazionale (http://www.pcn.minambiente.it/mattm/); these have a pixel size of 0.5 m and were downloaded in the UTM32 coordinate system based on the WGS84 datum. Most images were acquired in late summer months (August and September; **Table 1**), with minimum snow cover, while the orthophotos from 1988 and 2012 were acquired in early summer (June and July) and show evidence of residual snow, which however does not affect the glacier tongue. In addition to the aerial orthophotos, we obtained a stereo pair acquired from the Pleiades satellite constellation from the European Space Agency. The stereo pair was imaged on 1 September 2014 and provided at level 1B processing stage; we further processed it to obtain a DEM and an orthorectified image using NASA's AMES stereo pipeline (ASP). Eleven ground control points were selected from those available from Val d'Aosta regional authority (http://geonavsct.partout.it/pub/GeoNavITG/monografie.asp) to improve the geolocation accuracy of the DEM through bundle adjustment in the software; the DEM was then produced using the semi-global matching algorithm available in ASP [33], and the raw multispectral images were projected onto the DEM for orthorectification. Both the Pleaides DEM and multispectral image have a

Based on the available planar data (**Table 1**), the glacier and proglacial lake outlines were estimated over multiple years: in the 1975 and 1991 technical maps, they were already drawn on the map, while in the other datasets, they were manually digitized as part of this study. The spatial resolution of the orthophotos allowed us to clearly identify the outlines and to distinguish the debris-covered parts of the tongue from the proglacial areas. Before digitization, all available data were reprojected to the UTM32N coordinate system based on the WGS84 datum for consistency. We further evaluated the accuracy of the manual delineation using the buffer method proposed by Paul et al. [34], by allowing the glacier and lake outlines

The analysis of elevation and volume changes was based on the comparison of the DEMs from 1991 and 2014. Before performing this comparison, the DEMs were co-registered using the approach developed by Berthier et al. [35] and applied by Fugazza et al. [36]. In this approach, one DEM ('slave') is iteratively shifted with respect to a reference DEM ('master'), to minimize the standard deviation of elevation differences over stable areas located outside the glacier (σΔh). We selected the oldest DEM as the reference and resampled both DEMs to a common resolution of 10 m. By applying the co-registration, we obtained a residual σΔh of 4.18 m; the average elevation difference over stable areas was 7 m, which was subtracted from the reference DEM. Based on the elevation difference on the glacier surface, we then

> ∆V = ∑ k=1 K

pixel and dk is the pixelwise difference between the 1991 and 2014 DEMs.

where k = 1 … K are the pixels of the reference surface, Ak is the area of each

To express the uncertainty of volume changes, we used the approach described by Fischer et al. [37]: the uncertainty in elevation change (σ∆h) is not considered as entirely correlated but scaled to account for the effective correlated area Acor, based

dk Ak (1)

**172**

To characterize the evolution of debris cover on Lys Glacier, we adopted the methodology proposed by Azzoni et al. [9] for glaciers in the Ortles-Cevedale Group. Supraglacial debris was mapped by employing a maximum likelihood (ML) supervised classification approach. Four classes were included in the classification: debris, ice, snow and shadow. Although we did not aim at mapping the other surface types, a higher number of classes was chosen to permit an improved mapping of debris cover; the shadow class was also included as some images are affected by topographic shadow; this occurred mostly in the accumulation basins except for the orthophoto from 2003, where shadows occur over the whole glacier; thus, this orthophoto was excluded from the analysis (**Table 1**). We also excluded the maps from 1975 and 1991, as the information on the presence of debris was not available. For each of the satellite/aerial orthophotos, we independently selected 10 to 15 training areas. To estimate the accuracy of the classification approach, we performed a validation test on the 2006 and 2012 orthophotos, by selecting 100 random points for each test and comparing the results of ML classification against those from a manual classification. The 2 years were chosen because they represent best (2006) and worst (2012) conditions for debris cover mapping, in terms of image quality, presence of snow and shadows. For both years, we calculated overall accuracy (the ratio of the number of correctly classified points to the number of total points), as well as producer's accuracy (PA) and user's accuracy (UA), for the different classes. PA denotes the ratio of the number of correctly classified points to the total (manually classified) point for a class and is the complement of omission errors; UA denotes the ratio of the number of correctly classified points to the number of predicted points for a class and is the complement of commission errors.

#### **3.3 Meteorological data**

In order to interpret the glaciological data of the study area, we analyzed meteorological variables recorded by the weather station installed in 1927 at Gressoney d'Ejola, 1850 m a.s.l., at a distance of 3.7 km from Lys Glacier (**Figure 1**). This long dataset allows investigating the recent climate behaviour, which influences the evolution of the glacier. At this weather station, manual observations of air temperature, liquid precipitation, total snow depth and thickness of fresh snow, atmospheric pressure, relative humidity, wind speed and velocity and cloud type and cover were collected three times a day (at 8 am, 2 pm and 7 pm) from 1927 to 2012. The data record is almost uninterrupted except for a short period from 1962 to 1970, when the station was temporarily moved downvalley, at 1730 m a.s.l. in the Orsia village. In 2002, an automatic weather station was installed at the same site, and it replaced the manual weather station after its dismissal in 2012. The automatic weather station measures air temperature, total precipitation (by a heated gauge) and snow height (by an automatic webcam) every hour. The contemporary presence of manual and automatic instruments has allowed ensuring the homogeneity and continuity of the series. Although a weather station also exists at a higher elevation and closer proximity to Lys Glacier (Alpe Courtlys, 1992 m a.s.l., 2.5 km distance), this station was installed in 2001 and does not permit a long-term analysis of climate variables. In this study, we calculated monthly, seasonal and annual averages of air temperatures and monthly, seasonal and annual cumulated liquid/ solid precipitation for the analysis of climatological conditions.
