3.1. Glacier mapping

Glacier outlines were extracted from the six Landsat images in a standard semi-automatic way, using a ratio of TM bands 3 and 5 and an additional threshold of band TM1 to correct misclassified shadow regions automatically (e.g. Reference Paul and KääbPaul and Kääb, 2005). The threshold values for the ratio and band TM1 for the three years (in parentheses) are: 3.0 and 75 (1985), 2.4 and 65 (2000) and 3.0 and 55 (2011), respectively. As the method is unable to capture debris-covered ice (it has the same spectral properties as the surrounding terrain) but often includes parts of water surfaces (which have a similar ratio), these misclassified regions were edited manually using contrast-enhanced versions of the original satellite images in the background. Due to the different image acquisition dates, the glacier area covered by shadow is variable from scene to scene and correct mapping is sensitive to the selected band 1 threshold (see fig. 3 in Paul and others, in press). In the 1985 scene, shadow was most extensive and had partly to be manually corrected by visual inspection using contrast-enhanced images. In part, the high-resolution images available for this region in Google Earth also helped with the identification.

The largest challenge, however, was the proper identification and removal of seasonal snow off-glacier in all three scenes. This is a typical problem when glacier inventories are created in maritime regions (e.g. Reference Paul and AndreassenPaul and Andreassen, 2009; Reference Gjermundsen, Mathieu, Kääb, Chinn, Fitzharris and HagenGjermundsen and others, 2011), as seasonal snow cover hides glacier perimeters at high elevations year-round. Manual correction of seasonal snow is very time-consuming because of the large amount of small snowpatches and the difficulty deciding in each case whether a glacier lies underneath or not. To reduce the workload for the required corrections, a multi-step approach was developed. In a first step, a size filter of 0.05 km² is applied to remove most of the small snowpatches, i.e. larger than in other studies but the same as applied by Reference Rastner, Bolch, Mölg, Machguth, Le Bris and PaulRastner and others (2012) for the same reason. Of course, this also leads to the removal of glaciers that are smaller than this threshold, i.e. the real glacier number and area covered is higher.

In a second step, the amount of manual editing work was reduced by taking advantage of the multitemporal character of the study. Due to different meteorological conditions in each year (e.g. wind direction), the area covered by snow and its spatial distribution varies among the scenes while glacier extents at high elevations hardly change. Of the three scenes the 2000 scene has the best snow conditions in this regard. Based on the evidence that glaciers have mostly retreated over the entire period (e.g. as obvious from flicker images and previous studies), the corrected glacier extents for 2000 are used as a mask to remove seasonal snow from the 2011 glacier map (cf. also Reference Bolch, Menounos and WheateBolch and others, 2010). This substantially reduced the manual correction workload, but now no glacier can be larger in 2011 than in 2000. A careful inspection of both images (using an animated display) revealed that this was actually not the case and a few glaciers had to be corrected back to a larger extent. The manually digitized yellow lines in Figure 3 illustrate these corrections for a couple of glaciers located just southeast of Glaciar Inexplorado.

Fig. 3.

Not all glaciers lost area from 2000 to 2011. To the east of Monte Inexplorado some glaciers advanced and their outline was manually corrected back to their larger 2011 extent (marked yellow).

In 1985, glaciers were generally larger than in 2000, so snowpatches had to be identified and removed differently. After applying the size filter we carefully checked a selection of small polygons, namely those larger than 0.05 km² that did not touch a glacier boundary of 2000. These isolated polygons were identified with a neighbour-hood analysis and mostly removed after visual inspection. This step and the additional multitemporal overlay of 2000 on the 1985 glacier mask in shadow regions reduced the manual editing workload, but most of the corrections remained for manual editing.

As a final step we corrected the terminus position of some debris-covered glaciers with the multitemporal outlines from the classified glacier lakes (Section 3.2). If a lake did not touch the glacier boundary at time t ₁, even though at time t ₂the lake grows into the glacier polygon of t ₁, then the gap between the lake and glacier terminus at time t ₁ is added to the glacier extent, suggesting the glacier was debris-covered and not properly identified at t ₁. Additionally, areas that have been classified as a lake at t ₁ but as ice at t ₂ are reclassified as a lake at t ₂. In most cases the ‘wrong’ classification was due to frozen and/or still snow-covered lake surfaces.

3.2. Classification of lakes

Lake outlines in the vicinity of glaciers are derived from the year 2011 scene by combining a Normalized Difference Snow Index (NDSI) from Landsat TM bands 2 and 5 with a threshold of –0.37 and an additional threshold for band 4 (digital number <39). In a second step, shadow areas at the time of image acquisition are generated from the DEM and subtracted from the lake map, with all lakes smaller than 0.001 km² being removed. Due to inaccuracies and artefacts of the DEM as well as clearly different spectral signatures of the lakes (e.g. due to variable turbidity), a complete automatic classification was not possible and manual corrections based on visual comparison with contrast-enhanced false-colour composites are applied to >50% of all lakes. For the purpose of determining changes of the lakes, we restrict the sample to a subset of all lakes by applying a 500 m distance threshold to a glacier outline (i.e. all lakes farther away are not considered).

The areas covered by lakes for 2000 and 1985 are derived by digitally subtracting the corrected glacier extents of 1985 and 2000 from the 2011 lake extents. This works best for proglacial lakes that usually form in overdeepenings of the glacier bed (e.g. Reference Linsbauer, Paul and HaeberliLinsbauer and others, 2012) and become progressively exposed during glacier retreat. While for most lakes the simple subtraction of glacier extents already provided good results, some lakes still had to be manually edited. This could be due to a change of the lake extent and/or its position, or because a small lake followed the retreating terminus.

3.3. Drainage divides and topographic parameters

To split the contiguous ice masses (or glacier complexes) represented by the glacier polygons of the initial raster-to-vector conversion into individual glacier entities, digital drainage divides (or watersheds) are required (e.g. Reference PaulPaul and others, 2002). Several semi-automatic approaches are available for this purpose (Reference Kienholz, Hock and ArendtKienholz and others, 2013) which all produce valid results but differ in the amount of manual editing required. Apart from that, a key issue here is that differences in the elevations of the DEMs used for orthorectification of the satellite scenes on the one hand, and the drainage divides on the other hand, can result in a poor match of the divides with the glacier outlines along mountain crests. Hence, after digital intersection, a large number of so-called sliver polygons resulted that have been manually corrected (e.g. Reference Frey, Paul and StrozziFrey and others, 2012; Reference Kienholz, Hock and ArendtKienholz and others, 2013).

Here we use the semi-automatic method of Reference Bolch, Menounos and WheateBolch and others (2010) which is based on watershed analysis with standard functions implemented in a GIS. As we used a different DEM (ASTER GDEM2) for the divides than was used for orthorectification of the Landsat scenes (L1T product) by USGS (the GLS2000 DEM that is largely based on a void-filled version of the SRTM DEM in this region), a large number of sliver polygons resulted that were manually corrected to the extent possible (i.e. overlaps smaller than about two pixels were not corrected). Though this implies that some individual glacier areas are now too small or too large, this has little impact on change assessment as all glaciers have the same divides in each year. The correction of the divides was based on the 1985 glacier outlines, and basin IDs were also assigned for the 1985 extents. Tracking of glacier area changes through time is also based on the 1985 basins, regardless of a potential glacier split.

After glacier complexes were separated, zonal statistics were calculated using the glacier polygons as a zone and the DEM, slope and aspect grids as the values for the calculation of statistics (minimum, mean, median and maximum for elevation, mean for slope and aspect) following Reference Paul and AndreassenPaul and others (2009). Furthermore, glacier hypsography (area–elevation distribution) was derived for the entire region.

3.4. Accuracy determination

We have not determined the accuracy of the derived outlines quantitatively as the required reference data have not been available. However, several factors can be considered in a qualitative sense for a rough estimation of accuracy. In this region the main factors influencing the accuracy of the glacier outlines are: (1) seasonal snow and (2) manual editing of glaciers under debris cover or located in shadow. Errors in the position of drainage divides will influence the area of individual glaciers (e.g. size class assignment), but not the derived changes in area as the same divides are used for all three years. The influence of (1) is likely the largest uncertainty but can only be estimated until a scene with better snow conditions becomes available. As mentioned above, seasonal snow influences mostly small glaciers situated at high elevations, but in a systematic way (overestimation of the real area). As snow conditions were slightly worse in 1985 than in 2000, the result for this period is an overestimation of the area loss while it is an underestimation for the second period. We estimate this bias to be up to 5% for the first and ~10–20% for the second period. This has to be considered when interpreting the area changes of small glaciers.

For the accuracy determination related to manual editing, (2), we refer to Reference PaulPaul and others (2013) who derived an accuracy of ~3–5% (in the mean) from a multiple digitizing experiment of several glaciers with different sizes. Given that only a few glaciers are debris-covered in our study region but shadow caused problems in accurately identifying where glacier boundaries are located, the accuracy of the area determination for shadow-free glaciers is likely better than 5%, but areas can be in error by ~10% (or even more in individual cases) for glaciers in shadow. However, as over-and underestimation can occur, the overall uncertainty due to shadow or debris is small in the mean for the entire sample. Based on these considerations, we assume an overall uncertainty of 5% for the derived total glacier area in each year.

4.1. Glacier inventory

The number and total area of glaciers per size class is plotted in Figure 4 and shown by values in Table 1. The histogram of the number count has a similar shape to that for other regions of the world (see Reference PfefferPfeffer and others, 2014), but also has some differences. On a global scale, most glaciers are found in the classes 0.1–1 km², with a near-exponential drop of their number towards smaller and larger glaciers. In our study region, most of the 1664 glaciers analysed (813 in 1985) are found in the class 0.1–0.5 km², while the classes 0.05–0.1 km², 0.5–1 km² and 1–5 km² each contain about one-third of that number. Only 2.2% of all glaciers are larger than 5 km², and glaciers in the 1–5 km² class cover most of the area (39%), which is more than all glaciers smaller than 1 km² (33%) and larger than 5 km² (28%). Hence, most glaciers in the study region (69%) are small (<0.5 km²) and their numbers have declined substantially (~1/4) since 1985, while glaciers larger than 1 km² contributed most to the total area loss (73%) (Section 4.2). With an assumed overall uncertainty of 5% (could be higher for individual glaciers and 2011), the total area in each year is 1192.5 ± 59.6 km² for 1985, 951.0 ± 47.6 km² for 2000, and 899.2 ± 45.0 km² for 2011.

Plotting mean elevation vs glacier aspect (Fig. 5a) reveals ~100–200 m lower elevations for glaciers facing southeast and south than glaciers facing west. Interestingly, glaciers facing north have lower mean elevation values than those facing west, northwest or northeast. This points to a reduced importance of solar radiation and topographic shading for the location of glaciers. Furthermore, there is a strong dependence on geographic latitude and longitude, resulting in mean elevation differences of ~500 m for both (Fig. 5b). As the spatial pattern of mean elevation values in Figure 6 clearly reveals, they strongly increase with distance from the coast (decrease in precipitation) but also towards the north (decrease in precipitation and increase in temperature). Though a general upward shift of the mean elevation values through time can be observed (Fig. 5b) the dependence itself does not change much over the observation period.

Fig. 4.

Number of glaciers (bars) and area covered (line graph) per size class for the three investigated years.

Fig. 5.

(a) Mean elevation vs glacier aspect for the 1985 dataset (dots) and sector averages for the three investigated years (lines). (b) Averages of mean elevation vs longitude (solid) and latitude (dashed) for all three years.

Fig. 6.

Spatial distribution of mean elevation for all glaciers larger than 0.05 km². The increasing trends from south to north and west to east are well recognizable.

Plotting mean size vs minimum and maximum elevation (Fig. 7a) reveals the large spread of values for small glaciers. For glaciers larger than ~2 km² there is virtually no dependence of size on maximum elevation but a clearer dependence on minimum elevation. This implies that a larger accumulation area results in a glacier that can reach farther down. The size dependence of the mean slope of a glacier (Fig. 7b) is much less clear than in other regions, pointing to the different topographic characteristics of this region (e.g. larger glaciers with flat accumulation and steep ablation areas are as frequent as vice versa).

Fig. 7.

(a) Maximum (red dots) and minimum (black dots) elevation vs glacier size. (b) Mean slope vs glacier size. Both plots refer to the1985 glacier sample.

4.2. Glacier changes

In Figure 8 we show two example regions with glacier outlines from three points in time and the lake extents in 2011 to visualize the partly extreme shrinkage. The strong retreat of Glaciar Inexplorado II and III seen in the upper right part of Figure 8a (marked as No. 1) has already been shown in Reference Rivera, Bown, Carrión and ZentenoRivera and others (2012). When expanding the view to the surrounding glaciers, the high spatial variability of the changes and the fact that extreme area losses also occurred elsewhere can be seen (e.g. No. 2 in Fig. 8a and examples in Fig. 8b). While most of the smaller mountain glaciers (located at high elevations) show little shrinkage, it seems that in particular the largest valley glaciers with extended tongues at low elevations have lost much of their area, mostly by thinning. This impression is confirmed in the example shown in Figure 8b where two larger glaciers (Nos 3 and 4) on the valley floor disappeared. Several other large glaciers in this subregion lost their tongues or large parts of them. In most cases, tributary glaciers were separated from the main tongues at steeper slopes, resulting in a loss of contact with the accumulation area and a simple down-wasting of the remaining dead ice afterwards.

When plotting the relative change in area vs glacier size (Fig. 9), the large spread of change values towards smaller glaciers can be seen. The dots at –100% are indicative of the 374 (mostly very small) glaciers that completely disappeared between 1985 and 2011. Somewhat smaller relative area change values can be seen towards larger glaciers, but eight glaciers that were larger than 5 km² in 1985 also lost 28–48% of their area (most are depicted in Fig. 8). Mean relative area loss values (1985–2011) range from –80% to –41% for the two smallest classes (corresponding to a reduction in their number of 1/4) to between –18% and –25% for the other classes (Table 1), confirming the limited dependence of relative area loss on glacier size and the strong shrinkage of the larger glaciers (those larger than 1 km² contributed 73% of the total area loss). This is the first region we are aware of where the relative loss of area is highest (–37%) for the largest size class.

Table 1.

Number and area of glaciers per size class in the three investigated years, and the absolute and relative area changes

To further elucidate where the area changes took place, we determined the relative area changes also for distinct elevation intervals (Fig. 10). While the hypsometry for the three years already indicates a substantial area loss at low elevations, the bar graphs showing the relative changes per 100 m elevation bin confirm this finding. Below 1000 m, between 50% and nearly 100% of the area is lost in the respective elevation intervals. Moreover, the bars have about the same lengths in the two periods, indicating that the abundant seasonal snow in the 2011 scene has only a minor impact at these elevations. In absolute terms the strongest changes took place where most of the ice is found (~1700 m). Above 1700 m there is little change in the hypsometry from 2000 to 2011, confirming the impact of the seasonal snow at these elevations.

No correlation of area loss with location (latitude or longitude) is found, even when restricting the sample to better comparable glaciers smaller than 1 and larger than 0.1 km². However, we found clear trends in the average values of minimum and mean elevation (Table 2) that are largely due to the strong shrinkage of ablation areas. The increase of minimum elevation by 46 m from 1985 to 2011 is even more pronounced (95 m) when only looking at the subsample of 275 (209 in 2011) glaciers larger than 1 km². Further excluding glaciers that are in contact with a lake (i.e. have no change in minimum elevation) decreases the sample to 187 glaciers with a mean increase in minimum elevation of 105 m. The related changes in mean elevation are 44, 66 and 61 m (for the classes all, >1 km², and no contact to lakes, respectively).

Table 2.

Average values of minimum, mean and maximum glacier elevation for the three investigated years and different subsamples (all glaciers, glaciers larger than 1 km² and glaciers not in contact with a lake)

4.3. Lakes and their changes

Together with fast glacier retreat comes a rapid development of proglacial lakes. In principle, this development takes place in the emerging overdeepenings of large and flat tongues of valley glaciers and former confluence regions of ice flow (e.g. Reference Linsbauer, Paul and HaeberliLinsbauer and others 2012), but it is also observed for cirque glaciers that have either melted away completely or were reduced to some remnants bordering neighbouring rock walls (A, B and C in Fig. 8).

Between 1985 and 2011, >100 new lakes had appeared and 65 others had been growing due to glacier retreat, leading to a total area increase of 11.6 km² or 59% (Table 3). Assuming that all boundary pixels of the changed lake areas are wrong, this increase has an uncertainty (one standard deviation) of 15% or ± 1.74 km². The rate of growth was 0.5 km² a^–1 in both periods. Lakes were appearing and growing everywhere between ~600 and 1900 m, as they are closely linked to the minimum elevation of the glacier tongues which have a large spread (Fig. 6). Due to DEM artefacts over the comparably dark and low-contrast lake surfaces, their elevations could not be determined.

Table 3.

Count and area of lakes and their changes in time

5.1. Glacier inventory and changes

The peak of the glacier number in the 0.1–0.5 km² size class is in line with other regions of the world (Reference PfefferPfeffer and others, 2014), but the steep drop in the number towards the next larger size class has not been observed in previous studies. It might be a result of the steep topography and the maritime climatic conditions resulting in a high number of perennial snow and ice patches along mountain crests (Fig. 2). This number is also influenced by the assigned drainage divides that are to some extent arbitrary without being wrong as their number depends on the rules applied for separating the ice masses (minimum size, consideration of tributaries that are just in contact with the main glacier, use of topographic structures for subdivision, etc.). Therefore the number of glaciers in a specific size class should not be overinterpreted.

While in the first period 99% of all glaciers decreased in size, only 51% decreased in the second period. This is also reflected in the steep decline in the total number of glaciers from 1985 to 2000 but near-unchanged values until 2011 for the two smallest size classes. This can likely be attributed to the larger amounts of seasonal snow at high elevations in 1985 as well as in 2011. On the one hand it might lead to a stronger decrease than in reality until 2000; on the other hand there will be an underestimation of area loss for the second period, as the snow in 2011 could only be removed down to the 2000 extents. While real glacier shrinkage between 2000 and 2011 in these two size classes might have gone unnoticed due to the snow, it is also possible that these glaciers have not changed too much as they are mostly steep and can quickly adjust to climatic change. Furthermore, the large amounts of snow available at the higher elevations in this region (Fig. 2) may have helped to protect these glaciers from further area loss. The next available image with better snow conditions will tell the full story. The other larger size classes have lost area in all years, which might in part be due to the lower elevations they reach (Fig. 7a).

The strong shrinkage of larger glaciers is also reflected in the mean relative area changes per size class. In contrast to all other regions of the world investigated so far, the relative area loss increases towards larger glaciers and the smallest loss rates are observed for the 0.5–1 km² size class (Table 1). However, with the area change for the smallest glaciers possibly underestimated, their relative loss rates might actually be higher than those for the largest glaciers. Nevertheless, the high loss rates found in this region for the largest glaciers are remarkable.

Glacier shrinkage at low elevations is high, reaching 50–100% below 1000 m in both periods (Fig. 10). This dramatic decline is also obvious from the area changes shown in Figure 8 and highlights the importance of analysing area changes also in this regard. To some extent it seems that the glaciers rapidly losing their tongues (glaciers A–D in Fig. 8) are the shape of things to come in other regions of the world, once the ice becomes sufficiently thin (e.g. Reference Jouvet, Huss, Funk and BlatterJouvet and others, 2011). The rapid disintegration of glaciers at this stage is then caused by the comparably flat bed topography of large valley glaciers, resulting in a slowly increasing bed elevation over large parts of the tongue (e.g. Reference Linsbauer, Paul and HaeberliLinsbauer and others, 2012). When also becoming disconnected from the accumulation area in this situation, the glacier tongue rapidly turns to dead ice and will inevitably disappear.

Fig. 9.

Relative change in glacier area from 1985 to 2011 vs glacier size in 1985. Disappeared glaciers can be seen on the –100% line.

Fig. 10.

Area–elevation distribution for the three years (lines) and relative area changes (dA) for the two periods (bars) in 100 m elevation bins. Below 1000 m the area loss in both periods is similar.

The resulting changes in glacier length are also extreme. The case of Glaciar Inexplorado II (No. 1 in Fig. 8a) has already been highlighted by Reference Rivera, Bown, Carrión and ZentenoRivera and others (2012), revealing current retreat rates of >300 m a^–1. In a cumulative sense, this glacier retreated >4 km in 25 years. Similar values between 1 and 4 km can also be found for other glaciers in the region (see Fig. 8). To give an order of magnitude for comparison with other regions in the world, the kilometre scale is the typical retreat distance of glaciers in the European Alps since the end of the Little Ice Age, i.e. over a period of ~150 years (e.g. Reference Hoelzle, Haeberli, Dischl and PeschkeHoelzle and others, 2003). So far it has been assumed that (on a global scale) glacier retreat in the Alps was strong.

Fig. 8.

Glacier changes from 1985 (black lines) to 2000 (red) to 2011 (yellow), and lake outlines (white) for two test sites (see Fig. 1 for location). Glaciers numbered 1–4 lost large parts of their ablation area (or disappeared) over the investigated period, and the annotated lakes represent typical locations of lake emergence: A (at confluences), B (below large tongues) and C (cirques).

The variability of mean elevation with mean aspect sector (~200 m in altitude) is in line with other regions. However, the lowest elevations are found in the southeast sector (rather than the south), while the mean is highest in the western sector (rather than the north). This points to a smaller influence of radiation exposure on mean elevation as found in other regions (e.g. Reference Evans and CoxEvans and Cox, 2005). The maritime climatic setting might also be a reason for this difference. The strong dependence of mean glacier elevation on distance to the coast (i.e. longitude in this region) shown in Figures 5b and 6 (a nearly 500 m decrease) confirms the strong dependency of precipitation amounts on mean elevation that has also been found in other regions of the world (e.g. Reference Le Bris, Paul, Frey and BolchLe Bris and others, 2011; Reference Paul, Andreassen and WinsvoldPaul and others, 2011; Reference Rastner, Bolch, Mölg, Machguth, Le Bris and PaulRastner and others, 2012). On top of the longitude dependence is a latitude dependence that is likely related to both decreasing precipitation amounts towards the north and increasing mean temperatures (Reference Condom, Coudrain, Sicart and ThéryCondom and others, 2007). The dependence on latitude has also been found by Reference Davies and GlasserDavies and Glasser (2012), though over a much larger latitude range.

5.2. Accuracy

Overall, our derived area changes for individual size classes (–18 to –37%) are higher than the uncertainty of the area determination (~5% in the mean over all size classes) and are thus significant. However, for glaciers smaller than 1.0 km² the uncertainty is in the same order of magnitude as the observed relative area change and can thus be considered insignificant. When the two periods are analysed separately, this only applies to the second period (2000–11), with very small changes (Table 1) due to the impact of seasonal snow in 2011. Hence, it is not possible to say with certainty how glaciers in this period have changed, and forthcoming satellite data are needed to reveal the trends. For the other glaciers we assumed a 5% uncertainty of the total area values for each investigated year (can be higher for glaciers in shadow), resulting in the uncertainties provided above.

5.3. Lakes

Glaciers and their changes, especially the global recession trends, are an important topic in environmental science. However, the focus is usually on the issue of glacier melt, its causes and consequences, i.e. on the part of the landscape that is getting lost. On the other hand, there is also ‘new landscape’ appearing to replace what has been lost. In principle, we have hardly any idea what this new landscape will look like prior to its exposure. This is also due to our limited understanding of the processes occurring at the ice/bed interface (Reference Cook and SwiftCook and Swift, 2012). A most striking landscape change is certainly the development of new lakes in regions where a glacier had once been.

The size and number of the overdeepenings that have filled with water in the study region are already comparable to other largely deglaciated mountain ranges (e.g. the Pyrenees, southwest Europe, or the Wind River Range, Wyoming, USA) as can be seen on satellite imagery. During the study period, all lakes were either growing or stable; none were shrinking outside a one-pixel threshold along the perimeter. This trend will likely continue as many more flat glacier tongues surrounded by steep rock walls exist. In other regions of the world, such lakes pose a natural hazard as they can break out in disastrous floods (e.g. Reference Clague and EvansClague and Evans, 2000; Reference Frey, Haeberli, Linsbauer, Huggel and PaulFrey and others, 2010; Reference Bolch, Peters, Yegorov, Pradhan, Buchroithner and BlagoveshchenskyBolch and others, 2011; Reference IturrizagaIturrizaga, 2011). This problem might be smaller in this sparsely populated region of Chile, but it should still be taken into account when planning new infrastructure (e.g. in the context of hydropower projects in the region).

5.4. Climatic interpretation

In the present study, only area changes have been analysed, but length changes are obvious as well and the satellite images also show a substantial thinning of extant glaciers (e.g. by the emergence of rock outcrops). All this provides indirect evidence that strong mass losses have occurred due to mainly negative mass balances. Apart from volcanic activity, which might be the reason for changes of specific glaciers due to thermal melt, lubrication and ash layers (Reference Rivera, Bown, Carrión and ZentenoRivera and others, 2012), this can be the consequence of increased temperature, decreased precipitation or both. Climatic measurements in this region are sparse in time and space, and existing scientific information is partly contradictory. There is agreement on a slight cooling trend along the coast (Reference Bown and RiveraBown and Rivera, 2007; Reference Falvey and GarreaudFalvey and Garreaud, 2009) since the mid-20th century; in contrast, temperature trends at higher altitudes seem to be less clear, with conflicting results from different researchers. However, these have low credibility in any case, due to the complete absence of station data above 200 m altitude. The precipitation trend is clearer, with ~–560 mm between 1960–2000 for Puerto Montt, clearly negative (Reference Bown and RiveraBown and Rivera, 2007), but in the end not more than ~20% of the total amount.

The high sensitivity of glaciers in a maritime climate (Reference Oerlemans and FortuinOerlemans and Fortuin, 1992; Reference Raper and BraithwaiteRaper and Braithwaite, 2006; Reference Rupper and RoeRupper and Roe, 2008) together with the decrease in precipitation and possibly some slight warming at higher elevations might be the best explanation for the fast glacier retreat. However, in other regions with high precipitation, mass balance is more sensitive to temperature due to its influence on the rain/snow threshold (e.g. Reference OerlemansOerlemans, 1997; Reference AndersonAnderson and others, 2010). Attempting to explain the massive downwasting and collapse of many larger glaciers in this region from climatic evidence alone is thus unsatisfactory. We assume that the strong response of these glaciers to small changes in climate might be amplified by their special hypsometric setting (e.g. the separation from accumulation areas at high elevations).