Inventory and evolution of glacial lakes since the Little Ice Age: Lessons from the case of Switzerland

Retreating glaciers give way to new landscapes with lakes as an important element. In this study, we combined available data on lake outlines with historical orthoimagery and glacier outlines for six time periods since the end of the Little Ice Age (LIA; ~1850). We generated a glacial lake inventory for modern times (2016) and traced the evolution of glacial lakes that formed in the deglaciated area since the LIA. In this deglaciated area, a total of 1192 lakes formed over the period of almost 170 years, 987 of them still in existence in 2016. Their total water surface in 2016 was 6.22 ± 0.25 km2. The largest lakes are > 0.4 km2 (40 ha) in size, while the majority (> 90%) are smaller than 0.01 km2. Annual increase rates in area and number peaked in 1946–1973, decreased towards the end of the 20th century, and reached a new high in the latest period 2006–2016. For a period of 43 years (1973–2016), we compared modelled overdeepenings from previous studies to actual lake genesis. For a better prioritization of formation probability, we included glacier‐morphological criteria such as glacier width and visible crevassing. About 40% of the modelled overdeepened area actually got covered by lakes. The inclusion of morphological aspects clearly aided in defining a lake formation probability to be linked to each modelled overdeepening. Additional morphological variables, namely dam material and type, surface runoff, and freeboard, were compiled for a subset of larger and ice‐contact lakes in 2016, constituting a basis for future hazard assessment.


| INTRODUCTION
The current period of atmospheric temperature rise has led to intense glacier retreat in most mountain regions worldwide since the mid-19th century (e.g., Zemp et al., 2015) and the formation of numerous (peri-)glacial lakes [e.g., Aggarwal et al., 2017;Buckel et al., 2018;Emmer et al., 2016;Federal Office for the Environment (FOEN), 2021]. Glacial lakes can form in the overdeepenings of exposed glacier beds, behind natural dams of moraines or landslides, but also behind glacier ice and even rockglacier dams (Carrivick & Tweed, 2013;Buckel et al., 2018;Costa & Schuster, 1988). A growing number of such new glacial lakes formed since the end of the Little Ice Age (LIA) around 1850 as a consequence of climatic change. In times of accelerating anthropogenic climate impacts, they represent an essential element in a newly emerging alpine landscape.
New glacial lakes can have far-reaching effects on environmental systems as well as on infrastructure and populations (Carrivick & Tweed, 2013;Schwanghart et al., 2016;Tufnell, 1984). They are often dammed by unstable or unconsolidated material and can therefore pose a hazard to downstream human activities, infrastructure and lives (Clague & Evans, 2000;Costa & Schuster, 1988;Richardson & Reynolds, 2000;Walder & Costa, 1996). Glacial lake hazards are likely to increase in most glaciated regions worldwide due to the growing number of such lakes, the progression of glacier retreat towards steeper terrain, and due to (increasing) destabilization of rock walls, rock glaciers, and glaciers (Deline et al., 2015;Haeberli et al., 2017).
Glacial lake outburst floods (GLOFs) can strongly shape their downstream environment through erosion and sediment deposition (e.g., Carling, 2009;Cook et al., 2018;Russell et al., 2006) while existing lakes interrupt the sediment transport to downstream lakes and rivers. In contrast, glacial lakes also provide opportunities and gain importance in terms of economic value for hydropower production or as a tourist attraction site (Haeberli et al., 2016;Farinotti et al., 2016).
A complete and consistent inventory of glacial lakes can hence support regional development and decision making in view of environmental, societal and economic concerns.
Many studies on new glacial lakes focus on the Himalaya-Karakoram-Hindu-Kush (e.g., Zhang et al., 2015) and the Andes (e.g., Emmer et al., 2016), whereas in the European Alps only one region (Austria) has been systematically inventoried on a national scale (Buckel et al., 2018;Emmer et al., 2015). Switzerland has a rich history of glacier monitoring activities, including some of the world's longest time series of glacier length changes, and an extensive mass balance monitoring programme since the 1960s [World Glacier Monitoring Service (WGMS)]. In contrast to the retreating glaciers, little research has dealt with the appearing landscape including glacial lakes (Haeberli et al., 2016). Some individual cases linked to GLOFs in Switzerland in modern times were investigated, for example at Lac de Mauvoisin in 1818 (Tufnell, 1984;Woodward, 2014), and the Sirwolte lakes in 1993 or the Weingarten lake in 2001 (Huggel et al., 2002(Huggel et al., , 2004. In this study, we reconstruct more than 150 years of glacial lake formation history based on countrywide glacier boundaries from several reference periods since the mid-19th century, and historic orthomosaics acquired by the Swiss Federal Office of Topography (Swisstopo). This systematic inventory of glacial lakes and their evolution are related to causal environmental conditions, and compared to previously modelled glacial overdeepenings for the period 1973-2016 (Linsbauer et al., 2012) in view of further prediction strategies.
Based on the described necessities, the objectives of this study are to (1) generate a comprehensive glacial lake inventory of Switzerland for the year 2016, (2) analyse lake evolution since the LIA and link it to glacier retreat and climatic conditions, and (3) assess the adequacy of modelled glacier bed overdeepenings to anticipate locations with possible future lake formation.

| STUDY SCOPE
This study focuses on the perimeter in the Swiss Alps that was deglaciated between $1850 (end of LIA) and 2016, which encompasses 743 km 2 along an elevation range from 930 m to almost 4000 m (Figure 1). Between these dates, the glacier area decreased from 1783 km 2 ($4.3% of the country surface; Maisch et al., 2000;Paul, 2003) to 1040 km 2 ($2.5% of the country surface; Swisstopo, 2020a). At the end of the LIA, many glaciers reached a peak extent, often leaving behind more or less distinct terminal moraines from which they have almost continuously retreated until today, with the exceptions of intermittent favourable climatic phases that partly caused glacier readvances in the 1910s/1920s and in the 1970s/1980s (Bauder et al., 2007;Huss et al., 2008).
Within the extent of the glacierized area in 1850 we analysed peri-glacial lakes, that is lakes that were located either in the proglacial area (maximum distance roughly 3 km to today's glacier tongue), or at or close to the glacier margin above the terminus. Supra-glacial lakes were disregarded due to their temporary nature and small size. Lakes still in touch with the glacier at a specific date are called ice-contact lakes at the respective time; their formation is potentially incomplete at the date of observation. After separation from the glacier, lakes become either stable in area or shrink due to sedimentary infill or sudden release (GLOF).

| DATA AND METHODS
Our analyses are based on lake and glacier boundaries from previous studies and national cartographic records by Swisstopo, which we complemented with lake and glacier boundaries derived from Swisstopo's national orthophoto mosaics (Swisstopo, 2020b; Table 1).
These datasets were combined to generate a time series of lake boundaries from 1900 to 2016 within the deglaciated area, similar to the study by Viani et al. (2017) for north-western Italy. Lake boundaries were either manually mapped or inferred by combining glacier and lake boundaries of present and subsequent points in time, respectively, as explained in Section 3.2. To establish a basis for future hazard assessment, several additional lake variables were retrieved using the orthophoto mosaic from 2014 to 2016 (hereafter referred to as '2016'; spatial resolution 0.25 m; Swisstopo, 2020a) and the corresponding national digital elevation model (DEM; SwissAlti3d, spatial resolution and vertical accuracy 2 m; Swisstopo, 2020b).

| Datasets previously available
Lake boundaries in vector format (minimum lake size 200 m 2 , i.e., 0.0002 km 2 ) are available from Swisstopo based on aerial imagery for the years 2006 and 2016 (Swisstopo, 2020c). We updated about one-third of these lake boundaries to fit the orthophotos from 2014 to 2016 and added a number of missing, smaller lakes to the 2006 dataset.
Glacier boundaries for the end of the LIA ($1850) were generated by Paul (2003) and Maisch et al. (2000) and are available in vector format at the Global Land Ice Measurement from Space (GLIMS) data browser (GLIMS, 2020;Raup et al., 2007). Freudiger et al. (2018 extracted and provided glacier boundaries from the georeferenced Siegfried maps, dating from roughly around 1900 (with up to several years earlier or later; for details, see Freudiger et al., 2018). Additionally, we used the vectorized (Maisch et al., 2000) glacier boundaries for the 1973 Swiss Glacier Inventory (Müller et al., 1976), also available from GLIMS.
To interpret the results with regard to climatic trends, we used

| Datasets produced in this study
We manually mapped complementary lake and glacier boundaries using Swisstopo's monochromatic orthophoto mosaics for 1946Swisstopo's monochromatic orthophoto mosaics for and 1979Swisstopo's monochromatic orthophoto mosaics for -1985Swisstopo, 2020d;Swisstopo, 2020e). These orthophotos have a spatial resolution of 1 m (1946) and 0.5 m (1980s) and were generated using the same DEM and set of ground control points. Thus, their geolocation is in high agreement and geolocation errors on flat and small surfaces (such as lakes) are minimal. The mapping was restricted to locations of subsequent lake formation.
We inferred lake extents for dates or regions where direct lake mapping was not possible (1900, 1973, snow-covered lakes in 1946 and 1980s), by using available or mapped glacier boundaries and the lake boundary of the next available date (Figure 2). This means that lake boundaries for a specific date t1 were either (1) taken from existing datasets, (2) mapped directly from orthophotos, or (3) inferred from the subsequent date t2 based on whether they were located completely or partly inside or outside the glacier polygon of date t1 ( Figure 2).  3.3 | Lake morphology and hazard assessment variables We define several morphological variables to describe the glacial lakes, their evolution, and their hazard potential. For each date and individual lake identifier (ID), the respective centre location and size were retrieved directly from the polygons. Additional information that was attached to each lake include the host Canton, the host mountain zone, a flag for ice-contact lakes in 2016, the lake level (lowest polygon elevation as extracted from the DEM), and the maximum formation time (starting at the end of the observation period before the lake appeared and ending in the last observation period with an increase in lake area).
Variables, which are considered relevant for hazard assessment include dam material and type, lake outflow, freeboard, and surrounding steepness. This information was derived from the 2016 orthophotos and for a subset of 225 lakes of relevant size (> 0.5 ha) or smaller ice-contact lakes, which have the potential to grow larger than the threshold size.
Both dam material and type are important for hazard assessment; for example, a debris dam is potentially prone to breaching, whereas a rock dam is stable and can only be overtopped. Dam material is also relevant for the assessment of ecological and hydropower potentials.
Dam material was defined visually from the 2016 orthophotos. The dam material categories included 'debris', 'rock', and 'ice', but also all possible mixed combinations since an accurate mapping from orthophotos is not always feasible. Dam type ('moraine', 'embedded') was assessed for all debris dams and defines, whether a lake is dammed by a moraine or located inside the subglacial till. 'Embedded' applied when the topography downstream of the lake was relatively flat (< 15 average for 300 m below the lake; cf. Aggarwal et al., 2017;Buckel et al., 2018;Petrov et al., 2017).
Lake surface outflow (categories yes/no) was also defined visually from the 2016 orthophotos. When a channelized riverbed was clearly visible, a surface outflow was assigned, even if a stream itself was not discernible in the image.
A freeboard of 0 m was assigned to all lakes with an outflow. In case of no outflow, the freeboard was defined as the elevation difference between the lake surface and the lowest point along the crest of the dam, independent of the crest-to-lake distance. This distance can reach high values of up to 200 m but was mostly smaller than 100 m. The freeboard was defined manually using the DEM and the respective hillshade. Freeboard was not defined for artificially dammed (regulated) and ice-dammed (en-/sub-glacial drainage) lakes.

| Mapping uncertainties
Manually mapped lake extents from multi-spectral remote sensing data are generally subject to relatively low uncertainty (e.g., Petrov et al., 2017). The uncertainty in lake area is linked to the image pixel size, which defines the smallest lake area that can be discerned based on a potential mapping accuracy. The vector data used in this study is based on very high-resolution aerial imagery, expert analyses at Swisstopo Based on these contrasting outcomes, we estimate AE1.5 m and AE2 m maximum uncertainty for the 1980s and 1973 lake outlines, respectively, which are based on monochromatic, high-resolution images. For the 1900 lake outlines, we estimate AE5 m uncertainty.

| Anticipation of lake formation
The location of possible future glacial lakes can be estimated by (1) calculations of glacier thickness (Farinotti et al., 2017;Linsbauer et al., 2012) or (2)  the location and volume of potential overdeepenings. However, the calculation of ice thickness is associated with an uncertainty of $30% (Farinotti et al., 2017), which strongly affects estimates on the potential volume of an overdeepening, whereas its location is more robust (Viani et al., 2020).
The visual method: Starting from a conceptual idea by Frey et al. (2010), recent studies analysed glacial lake formation by combining low surface slopes (< 10 ) with visual aspects of glacier morphology, that is a reduction of glacier width, a step in slope, and a visible rock step or surface crevasses below low gradient areas (Colonia et al., 2017;Magnin et al., 2020;Viani et al., 2020). Starting from the GlabTop overdeepenings, Magnin et al. (2020) attributed each overdeepening polygon with a probability category, which resulted from merging the varying intensities of the morphological criteria. They then compared the overdeepenings with and without the inclusion of morphological criteria to lakes that formed within a 10-year period in the Mont Blanc massive. They found that more lakes had formed in areas with higher probability values based on the visual method, but also that lakes had formed in areas where no overdeepenings were predicted by GlabTop.
In this study we used the approach by Magnin et al. (2020) and compared the lakes that formed during a period of $43 years (1973-

2016) over the Swiss Alps to the predicted overdeepenings from
GlabTop in order to assess GlabTop's accuracy and potential to estimate future lake formation. The GlabTop overdeepenings > 0.5 ha were weighted according to the visual glacier-morphological aspects (Table 2), which we extracted from the 1980s orthophotos and the Swisstopo DEM 25 Level 1 that had mainly been generated from the same data (Swisstopo, 2020f). For a detailed discussion of the chosen criteria please refer to Magnin et al. (2020).
For each lake, the values of the different criteria were summed up and classified according to the share that was reached from the potential maximum: 10 to < 30% = 1.
These categories zero to five can be seen as a representation of lake formation probability (Magnin et al., 2020).

| Lake inventory 2016
In total, we identified 1192 lakes that formed in the area that has become deglaciated since the LIA (Figure 3). Of these, 987 still existed in 2016. Twenty-four additional, purely artificial, lakes (reservoirs) with an area of around 8.5 km 2 were disregarded in our further analysis apart from the generation of hazard parameters. Three lakes (Theodul T A B L E 2 Type and attributed values of morphological criteria, following the approach by Magnin et al. (2020) Value attributed F I G U R E 3 Lake size and formation timing (first appearance) over the whole study area [Color figure can be viewed at wileyonlinelibrary.com] glacial lake; Bortelseewji; Griessee, all in Valais) that formed naturally were artificially dammed and enlarged for hydropower production.
The lakes are well distributed over the whole glaciated area of Switzerland ( Figure 3). Most lake area formed in the mountain zones with the most extensive LIA glaciation: most lake area formed in the southwest (1.92 AE 0.08 km 2 ), followed by the central northern Alps (1.33 AE 0.04 km 2 ) and south-eastern Switzerland (1.18 AE 0.05 km 2 ).
In the northwest, almost 1.2% of the deglaciated area was transformed into lake area, while the central Northern Alps showed with

| Lake changes and climate
The increase in lake area and number with reference to our assessment periods is relatively constant (Figure 6a). However, increase in average annual lake area and number over the last 10 years are several times higher than in previous decades, with a smaller peak in lake formation occurring in the third quarter of the 20th century ( Figure 6b).
Average lake size of all lakes increased from 0.0045 km 2 in 1900 to 0.0063 km 2 in 1973 and stayed roughly stable since then. In contrast, the median lake size was almost stable over the whole time series (roughly 0.001 km 2 ).
The somewhat over time, whereas in the period 1973-1982 the majority of the lakes were stable (Figure 9).
From the resulting > 300 lakes that were found to shrink (by more than the associated uncertainty) or disappear at some point in time, we could not retrieve a clear signal hinting to the lifetime of a lake. This was true also for smaller subsets when only considering larger lakes (i.e., > 0.001 km 2 and > 0.01 km 2 ). One hundred and eighty-seven lakes were found to disappear or shrink below the threshold size of 0.0002 km 2 during the study period, but the temporal resolution hampers a clear detection of the major responsible process(es). The time series shows, however, that a time frame of decades rather than centuries is sufficient for many lakes to strongly decrease in size or even disappear. Only one lake (small neighbouring lake of Fäldbachsee in Valais) was found to have re-grown after a period of shrinking, which might be due to a seasonal nature.

| Comparing mapped lakes to calculated overdeepenings
One hundred and sixty-six lakes larger than 1 ha with a total area of 9.02 km 2 were the maximum to be expected according to the calcu- About one-third of the overdeepened area without water in 2016 was found around lakes that did form, but generally not up to their maximum extent according to the calculated overdeepening ( Figure 10a). In other locations, no lakes formed at all (about twothirds). However, most of these areas were located on low-sloping surfaces, often with braided rivers and small, seasonal ponds. Conversely, the largest 'unpredicted' lake that formed is the partly supraglacial lake on Glacier de la Plaine Morte. This lake is at least partially ice-dammed and potentially not located in an overdeepening. More large, unpredicted lake area was found at Chüebodengletscher We then performed an additional analysis using the morphological criteria from Table 2 to assess lake formation probability. The evaluation of the overdeepened areas > 0.5 ha from GlabTop using supporting the findings by Magnin et al. (2020). All of the considered morphological aspects were equally important, that is no criterion seemed to be a single major driver leading to lake formation, which once more justifies the application of the visual approach.   in which glacial lake area has strongly increased globally (Shugar et al., 2020). Interestingly, Shugar et al. (2020) suggest that much of the increase in Switzerland took place prior to 2004, with lake volumes (estimated from areas) remaining constant or even slightly decreasing since then. This is clearly contradicted by the results presented here, and demonstrates the importance of high quality, manually corrected lake inventories for validating automated approaches, and for assessing temporal trends. A similarly sudden increase in lake number and area over the most recent decade was also reported for Austria by Buckel et al. (2018), while the increase in the Cordillera Blanca in the Peruvian Andes was almost linear since the mid-20th century (Emmer et al., 2020). Overall, the average increase in total lake area of around 0.11 km 2 /yr measured from 1948 to 2017 in the Cordillera Blanca (Emmer et al., 2020) is almost double the rate of around 0.069 km 2 /yr measured in Switzerland since 1946.
The period of strong increase in glacial lake number and area from the 1990s onwards has typically been linked with extensive glacier retreat (Chen et al., 2021;Shugar et al., 2020;Zhang et al., 2019).
In contrast, periods of little glacier change also led to decelerated lake formation or even lake area decrease (Gardelle et al., 2011). In Switzerland, the stable situation in glacier extent in the period 1973-1980s led to a comparatively small increase in lake number (Figure 6), and most of the new lake area emerged at existing lakes. Thus, the area increase per lake was with 595 m 2 /yr larger by a factor of 1.5 to 5 compared to all other periods. Only 29 new lakes appeared in that decade, which hosted less than 30% of the total area increase, whereas the majority of the new lake area can be assigned to 61 growing lakes that existed before 1973. Previous observations showed that ice-contact lakes negatively affect glacier mass balance (mechanical calving and increased melt at the lake-glacier interface; e.g., Clague & Evans, 2000;King et al., 2018), suggesting that glaciers in contact with pro-glacial lakes preferentially continued to retreat, even during this climatically favourable period with less negative mass balance. From the evolution of glaciers and climate during the 20th century a similar behaviour can be expected throughout the Alps; indeed, the disappearance of lakes due to advancing glaciers was observed in western Italy by Viani et al. (2017). A comparatively small lake area increase was also reported for Austria (Buckel et al., 2018), although the observed period   The elevation of lake formation did not increase over time, unlike expected from the intense glacier retreat at lower elevations, in contrast to lakes in the Aosta region (Viani et al., 2020). However, in the latest period substantially more lake area formed at higher elevations  During the mapping process, we noticed that several lakes also formed outside of the mapping perimeter, for example dammed by rock glaciers that developed out of terminal moraines, and at the distal side of lateral moraines. On the one hand, this highlights the advantage of an extended investigation perimeter as done in other studies (e.g., Gardelle et al., 2011;Mal et al., 2020). On the other hand, a larger perimeter shows the necessity to apply accurate and robust automatic mapping methods.
In our study area we did not find lakes that formed as a consequence of gravitational processes such as landslides or debris cones (that might be conditioned by glaciation, e.g., availability of debris).
Possibly, such situations exist in connection to former glaciation stages (Last Glacial Maximum), but mainly outside of our study area. This was mentioned also in Buckel et al. (2018), who found these types of lakes only in the lowest ranges of their study region (around 1700 m).
During the study period, many lakes strongly decreased in size and even disappeared. Some of these observations can clearly be attributed to reported cases of GLOFs (e.g., Lakes Weingarten and Sirwolte, glacial lake at Grubengletscher; Haeberli et al., 2001;Huggel et al., 2002Huggel et al., , 2004. These cases highlight the value of a highresolution database for intensely used regions such as the Alps, because even small lakes can lead to events with a strong impact. For Austria, a comparable area became exposed due to glacier retreat over the same time period ( Alps compared to Austria led to overall less eroded material to fill up the deglaciated overdeepenings. Viani et al. (2017) did not provide the deglaciated area but found a total lake area of 1.39 km 2 in the active glacier forefields of western Italy, with a considerably smaller glacier coverage in this region (Smiraglia et al., 2015). Overall, existing numbers on lake formation in different Alpine regions rank on a similar level.
In other mountain ranges, for example the Himalaya, central Andes, or Southern Alps of New Zealand, many lakes are larger than the largest Alpine lakes by an order of magnitude (e.g., Aggarwal et al., 2017;Emmer et al., 2016;Robertson et al., 2012). The reasons may be found in the generally larger size of glaciers and glacier valleys and the higher rates of erosion due to larger elevation differences, higher uplift rates, and greater precipitation amounts (Fitzsimons & Veit, 2001;Nikonov, 1989), which has overall led to the formation of large terminal moraines and associated lakes. Also, most of the lakes in the Himalaya and Cordillera Blanca, Peru, are moraine-dammed, whereas in the Alps this is only the case for on average 11-23% (this study, Aggarwal et al., 2017;Buckel et al., 2018;Viani et al., 2017;Viani et al., 2020).

| Predicting glacial lakes
Due to the considerable uncertainties in ice thickness estimates, our comparison of modelled and observed lakes was restricted to lakes larger than 0.05 km 2 for the period 1973 to 2016. The comparison suggested that most of the large lakes could be expected with high certainty based on the overdeepening calculations. On the one hand, concerning individual lakes, the overdeepening calculations using GlabTop seem to overestimate potential lake scenarios both in number and area, which was confirmed in previous studies (Magnin et al., 2020;Viani et al., 2020). On the other hand, expected and actual lake location match well in most cases of larger lakes [in line with findings by Magnin et al. (2020) and Viani et al. (2020)], whereas smaller lakes often were either not expected or did not form. The inclusion of morphological criteria would add a formation probability to each modelled overdeepening (as found also by Magnin et al., 2020) that could strongly support prioritizing cases for further investigation (e.g., ground penetrating radar measurements for better estimating the overdeepened area). When starting from the GlabTop calculations, we ignore lake formation probability in other locations.
We observed lake formation of mainly small lakes in the forefield of small cirque glaciers, behind moraines of neighbouring glaciers, or above small rock steps in larger flat forefields. Magnin et al. (2020) also observed lake formation outside of GlabTop overdeepenings but noted that these locations had a low formation probability also applying the visual method.
In their comparison of modelled overdeepenings to actual lakes, Magnin et al. (2020) and Viani et al. (2020) found an accordance of $60% and $45%, respectively. These numbers are to some extent dependent on the size threshold (1 ha in Magnin et al., 2020;0.5 ha in Viani et al., 2020) but are in a similar range as found in our study (34%, 0.5 ha).
In our study we did not consider the volume of the overdeepenings. The reasons are the uncertainties associated with ice thickness calculations (Farinotti et al., 2017), making it difficult to estimate lake volume (Magnin et al., 2020;Viani et al., 2020). For a better estimate in the case of large modelled overdeepenings, it is therefore advised to additionally perform in situ measurements, for example using ground penetrating radar (Viani et al., 2020). However, even with this information, the possibility of a narrow drainage channel in the glacier bed and the infill of sediments into the basin during glacier retreat might lead to much smaller water volumes and introduce uncertainties that persist until the formation of a lake.

| Lake sedimentation
The long time period covered in this study with six intervals revealed that in some periods many lakes decreased in size ( Figure 9). As soon as a lake starts to form, sediments are transported into the lake basin and contribute to its volume reduction (Leonard & Reasoner, 1999;Loso et al., 2004). Sediment fluxes are supposed to be highest immediately after deglaciation due to an oversteepened environment and paraglacial slope readjustment (Ballantyne, 2002;Hallet et al., 1996;Leonard & Reasoner, 1999;Loso et al., 2004). In shallow lakes, these processes can relatively quickly lead to a reduction in lake area.
Examples from our study suggest a relatively short lifetime for some of the lakes. For example, the glacial lake at Huefifirn in the Canton of Uri decreased by about 20% in area between 1985 (0.067 AE 0.001 km 2 ) and 2016 (0.053 AE 0.001 km 2 ; Figure 12). In the case of multiple lakes within a glacier forefield, the upper lake(s) act as a retention basin, substantially reducing the sediment influx for the lower lake. For an improved understanding of sedimentation into alpine lakes it is necessary to investigate lake bathymetry multitemporally as well as the sedimentation rate itself over longer time periods. Projected changes in rainfall, as well as altered rates of paraglacial slope adjustment and permafrost degradation under a warmer climate also imply that rates of sedimentation in the future may differ from what has been observed in the past.

| Mapping challenges
The mapping was based on end-of-summer images (August, September), during the most favourable snow and ice conditions of the year. Nevertheless, mapping in high alpine areas is challenged by seasonal snow and ice cover. Consequently, it cannot be excluded that (small) frozen lakes have been wrongly classified as glacier or seasonal snow and were missed in the mapping process.
The main uncertainty results from the mapping itself and depends on the resolution of the imagery. Using the uncertainty of 213 (mostly small) of 373 lakes were partly or fully snow-covered (images acquired mostly in July and August). Based on the assumption that a potential basin gets water-filled once it becomes exposed, we copied the lake boundaries from 1973 and adjusted them with the glacier boundaries from 1946, analogue to 1900 and 1973, and assumed a higher mapping uncertainty of AE4 m to account for these difficulties.
Possible variations in growth and contraction of lakes between dates with orthophotos and glacier outlines could not be captured. In these periods, also potential changes at the outflow of existing lakes, such as the increase or lowering of outflow elevation or the change of outflow position, had to be disregarded.
With the advent of several high-resolution satellite sensors over the last few years, some studies continued earlier efforts to investigate the possibility to automatically map glacial lakes based on such satellite imagery (e.g., Zhang et al., 2020;Wangchuk & Bolch, 2020;Qayyum et al., 2020). In this context, the high accuracy of our 2016 outlines makes them a useful reference dataset for any such attempts, which face major challenges such as high elevated lakes with seasonal snow cover, lakes of varying turbidity or cast shadow.

| CONCLUSIONS AND PERSPECTIVES
Our study represents one of the most complete inventories of glacial lakes in space and time since the LIA. Generating such a database was only possible due to the availability of high-quality data over longer time periods over entire Switzerland, which is not the standard in many other, especially larger and developing, countries.
Whereas most contemporary studies have been limited to the highresolution satellite era (c. 1990 onwards), the analysis of such a comprehensive dataset allowed us to better understand dynamics and magnitude of lake formation, growth, and decline. Such time series of lake evolution are especially important to disentangle the relation between climate, glacier change, and lake formation. Additionally, this study lays the groundwork for a national assessment of both, potentially hazardous lakes and such that provide economic opportunities. Although there are individual examples of technical interventions at lake sites in Switzerland (Haeberli et al., 2001), a national risk mitigation strategy is still missing (Faulkner, 2001;Haeberli et al., 2016). Using aerial imagery with spatial resolutions between 0.25 and 1 m as basic data enables maximum accuracy and completeness, both critical given that even small discharge peaks can potentially lead to damage and fatalities (Haeberli, 1983;Petrov et al., 2017;Byers et al., 2019). It also represents an outstanding reference dataset for the validation of global scale glacial lake mapping using multispectral satellites such as Landsat-8 or Sentinel-2 (e.g., Pekel et al., 2016;Shugar et al., 2020).
In view of lake and water resource management and hazard prevention, one important question to know in advance is where glacial lakes might appear and the water volume they might contain (e.g., Haeberli et al., 2016). In a second step it would be beneficial to reliably estimate the lifetime of these lakes (e.g., Emmer et al., 2020).
Our study provides a baseline for such investigations as it yields an overview of new glacial lakes as well as insights into their formation timing and parts of their life cycle. Our results showed a dynamic variability of many lakes in the order of decades, since 15-20% were found to disappear during the study period since the end of the LIA.
For some lakes, a larger set of images exists that could be exploited for in-depth studies of their areal development andby combining image analysis, DEM generation, and modelling approachesto analyse aspects such as sediment transfer into the lake basin. Periodic measurements of lake bathymetry constitute one possibility to quantitatively estimate sediment inputs and potential lake lifetimes.
F I G U R E 1 2 Sedimentation over a period of 31 years for a large glacial lake at Huefifirn [Color figure can be viewed at wileyonlinelibrary.com] The large number of lakes and their location in an often unstable and steep topographic context shows the need for a deeper analysis of their hazard potential (Magnin et al., 2020). Such an assessment can then constitute a starting point for a comprehensive adaptation strategy, for example by pointing out specific lakes for regular change monitoring. A regular monitoring approach could assist in the early detection of glacial-lake related hazards.