Located in the Russian Far East, the Kamchatka Peninsula occupies ∼ 260 000 km2. It extends approximately 1250 km from the Koryak Highlands in the north to Cape Lopatka in the south and separates the North Pacific Ocean to the east from the Sea of Okhotsk to the west (Fig. 1). The peninsula has been shaped by both volcanic and glacial forces (Braitseva et al., 1995, 1997; Ponomareva et al., 2007, 2013; Barr and Clark, 2012a, b) and is dominated by three mountain ranges: the Sredinny (Central) Range, the Eastern Volcanic Plateau and the Vostocny (Eastern) Range (Fig. 1). Between these ranges lies the Central Kamchatka Depression, with relatively flat terrain punctuated by high volcanic peaks (> 4000 m above sea level; a.s.l.). The three principal mountain ranges are orientated north-east to south-west, reflecting the position and eastward movement of the Kuril–Kamchatka subduction zone, which forms an offshore trench running almost parallel to the peninsula's eastern coastline (Bulin, 1978). This subduction zone is responsible for the 30 active and ∼ 300 inactive volcanoes on the peninsula (Solomina et al., 2007).

Kamchatka lies between 50 and 61∘ N, and 155 and 163∘ E, and its proximity to the Pacific Ocean and Sea of Okhotsk results in a significantly milder climate than much of adjacent Siberia. Average summer temperatures range from 10 to 15 ∘C (Ivanov, 2002), whilst average winter temperatures range from -8 to -10 ∘C in the south-east, and from -26 to -28 ∘C in the interior and north-west, where cold temperatures are amplified by the effects of continentality and higher altitudes. A similar gradient is visible in the precipitation regime, with average annual values of ∼ 1500–2000 mm in the south-east and 400–600 mm in the north-west. This gradient is partly a consequence of the mountain chains acting as orographic barriers, impeding and disrupting air flow from the Pacific (Barr and Clark, 2011). During winter, cold, dry air is drawn from the Siberian High over central Siberia (Barr and Solomina, 2014). However, this is tempered by the development of a low pressure system over the Sea of Okhotsk, which sustains precipitation and elevates temperatures during winter (Velichko and Spasskaya, 2002). During summer, the Pacific High, to the south-east, brings warm–moist air masses across the peninsula. Precipitation peaks during autumn, aided by the delayed onset of the East Asian monsoon (Velichko and Spasskaya, 2002).

At present, glaciers on the peninsula are found within the three principal mountain ranges and on the high volcanic peaks of the Central Kamchatka Depression (see Fig. 1). The first attempts to estimate the extent of the region's glaciers were published as part of the “Catalogue of Glaciers in the USSR, volume 20” (Vinogradov, 1968; Khromova et al., 2014). This was based on observations from topographic maps and aerial imagery captured in the early 1950s (various years), supplemented by field studies (Kotlyakov, 1980). In this survey, 405 glaciers were documented, with a total area of ∼ 874 km2 (Vinogradov, 1968). This was used as a basis for the USSR Glacier Inventory (UGI), and the dataset was subsequently incorporated within both the World Glacier Inventory (WGI) (WGMS, 1989) and Randolph Glacier Inventory (RGI) version 2 (Arendt et al., 2012). Subsequent estimates of the total number and surface area of glaciers on the peninsula tend to differ. For example, Muraviev (1999) reports 448 glaciers, with a total area of ∼ 906 km2, while Solomina et al. (2007) report 446 glaciers, with a total area of ∼ 900 km2. Most recently, Earl and Gardner (2016) conducted extensive mapping of glaciers in northern Asia using automated remote sensing techniques (i.e. NDSI), with a size threshold of 0.02 km2 and documented 984 glaciers on Kamchatka, covering a total area of 770.3 km2. These data were incorporated into the most recent version of the RGI (version 5, released in July 2015) (Arendt et al., 2015). However, it is of note that the Earl and Gardner (2016) dataset is based on a mosaic of satellite images from multiple years (2000, 2002, 2009, 2011, 2013), meaning their estimate does not reflect glacier extent at a single point in time (or during a single year).

In addition to these peninsula-wide inventories, investigations into recent glacier fluctuations in specific regions of Kamchatka have been conducted (locations identified in Fig. 1). For example, based on the analysis of satellite imagery, data from the UGI, and aerial photographs, Muraviev and Nosenko (2013) identified a 16.6 % decline in glacier surface area in the northern sector of the Sredinny Range between 1950 and 2013. Muraviev (2014), using these same techniques, identified a 19.2 % decrease in glacier surface area in the Alney–Chashakondzha region between 1950 and 2013 and a 22.9 % decrease on the Kronotsky Peninsula between 1950 and 2010. On the basis of fieldwork (2007–2010) and the analysis of aerial photographs (from 1974), Manevich et al. (2015) documented fluctuations of 27 glaciers on the slopes of the Avachinsky volcanic group between 1974 and 2010 and found that, during this period, seven glaciers advanced, two retreated and eight remained largely stationary. Other, smaller-scale studies include investigations of glaciers on the slopes of Ichinsky Volcano (55.690∘ N, 157.726∘ E) (see Matoba et al., 2007), Koryak Volcano (53.321∘ N, 158.706∘ E) (see Manevich and Samoilenko, 2012), in the Klyuchevskaya volcanic group (∼ 56.069∘ N, 160.467∘ E) (see Shiraiwa et al., 2001), and at Koryto glacier (54.846∘ N, 161.758∘ E) on the Kronotsky Peninsula (see Yamaguchi et al., 1998, 2007, 2008). As a result of these investigations, some of the glaciers in the Central Kamchatka Depression have been documented as “surge-type” (Vinogradov et al., 1985). For example, Bilchenok glacier (56.100∘ N, 160.482∘ E) in the Klyuchevskaya volcanic group is known to have surged ∼ 2 km in 1959/1960 and again in 1982/84 (Muraviev et al., 2012). In this example, surging appears unrelated to climate and is likely driven by the strengthening of seismic activity at Ushkovsky Volcano (56.069∘ N, 160.467∘ E), upon which the glacier sits (Muraviev et al., 2012). Direct mass balance observations from Kamchatka's glaciers are limited, with the longest continuous record obtained for Kozelsky glacier (53.245∘ N, 158.846∘ E), spanning the 1973–1997 period. Other mass balance measurements, from Koryto, Mutnovsky SW (52.448∘ N, 158.181∘ E), Mutnovsky NE (52.460∘ N, 158.220∘ E), and Kropotkina (54.321∘ N, 160.034∘ E) glaciers, are far shorter and often discontinuous (see Barr and Solomina, 2014).

Changes in glacier extent on the Kamchatka Peninsula between 2000 and 2014 were determined through visual analysis of multi-spectral Landsat 7 Enhanced Thematic Mapper Plus (ETM+), Landsat 8 Operational Land Imager (OLI), and Terra Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) satellite images (see Table 1). Orthorectified scenes were obtained from the USGS Earth Explorer website (http://earthexplorer.usgs.gov/) and projected using WGS 1984 Universal Transverse Mercator (UTM) zone 57N, then filtered to find those with minimal cloud cover and snow cover. To achieve this, images were restricted to those captured during the ablation season (for Kamchatka this occurs from late July to early September). Satellite images captured prior to 2000 (e.g. Landsat MSS and CORONA) were obtained and analysed; however images captured during the ablation season, with minimal snow cover and cloud cover were limited, and full coverage of the peninsula's glaciers was lacking. As a result, the present study only focuses on 2000 and 2014 (i.e. those years with full satellite coverage of the region's glaciers). In total, 14 Landsat 7 (ETM+) scenes and 1 ASTER image were used for 2000, with 14 Landsat 8 (OLI) scenes used for 2014 (see Table 1). The Landsat scenes have a spatial resolution of 30 m for all bands except panchromatic band 8 (15 m) and Thermal Infrared Sensor (TIRS) band 6 (60 m), whilst the ASTER image (bands 1–3) has a spatial resolution of 15 m. To analyse glacier topography, SRTM 1 Arc-Second Global elevation data were obtained from the USGS Earth Explorer website. This reflects the surface topography of the glaciers in February 2000, with a horizontal resolution of ∼ 30 m and with vertical errors < 16 m (at the 90 % confidence level) (Farr et al., 2007).

As highlighted by Paul et al. (2013), the identification and mapping of glaciers from satellite images can be conducted using semi-automated techniques (e.g. Paul and Andreassen, 2009; Kamp and Pan, 2015; Smith et al., 2015; Earl and Gardner, 2016) or through manual digitising (e.g. DeBeer and Sharp, 2007; Stokes et al., 2013). In the present study, in order to establish the best method for mapping glaciers on Kamchatka, both semi-automated techniques and manual digitisation were trialled. The semi-automated techniques (specifically a RED/SWIR ratio and Normalised Difference Snow Index (NDSI)) produced rapid outputs and were easily applied across large regions but had difficulty differentiating ice from the surrounding environment (Fig. 2). The RED/SWIR band ratio had particular problems in distinguishing shaded areas and dirty ice, while the NDSI sometimes failed to distinguish ice from clouds and water bodies. Due to these difficulties, combined with the extensive manual editing and post-processing required to produce precise glacier margins, the semi-automated techniques were rejected in favour of manual mapping. Although time-consuming and sometimes subjective (Bhambri et al., 2011; Paul et al., 2013; Pfeffer et al., 2014), manual mapping remains the best method to extract detailed information on glacier dimensions (Kääb, 2005) whilst additionally allowing snow patches and areas covered by shadow to be identified (see Fig. 2).

To help differentiate glaciers from their surroundings and aid in the visual detection of glacier margins, both true- and false-colour composites of the satellite images were generated (Malinverni et al., 2008), with the high-resolution band 8 (15 m) used to sharpen the image. The true-colour composites were used for the identification of glaciers, while the false-colour composites helped differentiate ice and cloud, due to the strong absorption of ice in the SWIR band when compared to clouds (Nuimura et al., 2015). Glaciers were mapped in ArcGIS 10.2.2 (by a single operator) as polygon shapefiles, and features < 0.02 km2 in 2000 were excluded from the analysis on the grounds that they are likely to be snow or ice patches, rather than glaciers (Paul and Andreassen, 2009; Paul et al., 2009; Bajracharya and Shresta, 2011; Frey et al., 2012; Racoviteanu et al., 2015). To help verify that each of the digitised polygons was a glacier rather than a transient snow patch, scenes from intervening years, including 2015, were also obtained and analysed (Paul et al., 2009). However, the glacier inventory produced here does not contain detailed information about glacier dimensions for these intervening years since complete peninsula-wide imagery without extensive snow cover or cloud cover was unavailable for these periods, meaning that not all glaciers were mapped. For each of the glaciers mapped in 2000 and 2014, two-dimensional surface area was measured directly from the shapefiles, and area differences between 2000 and 2014 were calculated. Where glaciers fragmented into several distinct ice masses over this period, the net change in area was calculated based on the sum of the total area of each of the glacier “fragments” (following DeBeer and Sharp, 2007). For each glacier identified in the inventory in 2000, maximum, minimum and median altitude as well as mean surface slope were calculated from the SRTM digital elevation model (DEM), and generalised glacier aspect was estimated from a line connecting the glacier's maximum and minimum altitudes. The mean annual receipt of solar radiation at the surface of each glacier was calculated using the “Solar Radiation” tool in ArcGIS (algorithms developed by Fu and Rich, 2002), and glacier length was estimated along inferred flow lines.

Potential sources of error in this study arise through the digitisation process and with difficulties in correctly identifying areas of glacial ice. The accuracy of digitisation depends partly on the spatial resolution of the satellite imagery, snow conditions and the contrast between glacier ice and the surrounding environment (DeBeer and Sharp, 2007; Stokes et al., 2013). In Kamchatka, it is notable that the maritime conditions create particular difficulty with locating cloud-free imagery, and persistent snow cover often hinders the clear identification of glacier margins (Paul and Andreassen, 2009). Where glacier margins were partially obscured by shadow, debris or cloud, careful analysis using multiple true- and false-colour composite band images, as well as thermal imagery, was undertaken. Error was calculated following the method described in Bajracharya et al. (2014): RMSE=∑i=1n(ai-a^i)2n, where ai denotes glacier area and a^i is the glacier area calculated on the pixel base (i.e. the total number of pixels within a polygon, multiplied by the highest image resolution (15 m) used for mapping), and n is the number of polygons digitised. Error was found to be ∼ 3.6 % in 2000 and ∼ 3.4 % in 2014. These values are comparable to those reported in other studies (Bolch et al., 2010; Bhambri et al., 2013; Bajracharya et al., 2014).

NCEP/NCAR reanalysis data (see Kalnay et al., 1996) were used to analyse recent fluctuations in climate across Kamchatka. The NCEP reanalysis data integrate available land surface data from climate stations with the recovery of data from radiosondes, aircraft, ships and satellites to assimilate a quality-controlled gridded (1.88∘ latitude × 1.88∘ longitude) record of surface climate variables for the period 1948–present (Kalnay et al., 1996). The precipitation data reflect surface conditions, and temperature data reflect conditions at a height of 2 m a.s.l. Reanalysis data were used in preference to analysing individual weather stations, since the latter are often situated in coastal locations, distant from Kamchatka's glaciers. Additionally, many of the available station records include periods of missing data.

In total, 673 glaciers, with a combined surface area of 775.7 ± 27.9 km2, were identified and mapped on the peninsula in 2000 (see Fig. 1). Summary statistics for these glaciers are presented in Table 2, revealing a mean surface area of 1.15 km2 (ranging from 0.02 to 48.82 km2), with ∼ 77 % < 1 km2, and less than 4 % > 5 km2, though the latter represent just over one-third of the total glacier area. Glaciers range in length from 0.17 km to 20.96 km, while ranging in altitude from 273 to 4407 m a.s.l., with a mean altitudinal range of 465 m, and a mean surface slope of 17∘. The mean annual receipt of solar radiation at the surface of each glacier ranges from 489.8 to 1012.4 (kWh m-2). Across the peninsula, 52 glaciers have tongues covered with debris (rock and ash), the majority of which exist on volcanic slopes in the Central Kamchatka Depression (see Yamaguchi et al., 2007).

In 2014, 738 glaciers, covering an area of 592.9 ± 20.4 km2, were identified and mapped on the peninsula. This represents an additional 65 glaciers, but an overall area loss of 182.9 ± 6.6 km2 (∼ 24 %), relative to 2000. The increase in glacier number occurred despite the loss of 46 glaciers during this period (these glaciers were all < 0.5 km2 in 2000 and either completely disappeared or fell below the size threshold of 0.02 km2, by 2014) (see Fig. 3) and primarily reflects the fragmentation of larger glaciers. Of the peninsula's 673 glaciers identified in 2000, 654 (∼ 97 %) experienced a reduction in surface area by 2014, and this decline is seen across the peninsula (Fig. 4). Of the 17 glaciers that increased in area during this period, the majority experienced minor growth. One glacier increased by ∼ 140 % (see Fig. 4b) but was very close to the size threshold of 0.02 km2 in 2000, and the area gain over the period of observation only equates to 0.03 km2. In addition to areal decline, signs of stationary thinning (downwasting) are apparent in many areas, reflected by the exposure of many rock outcrops and the fragmentation of glaciers (see Fig. 5). The exposure of bedrock in this way further accelerates glacier wastage as local albedo is reduced (Paul et al., 2007).

When analysed across the peninsula, the NCEP data appear to indicate a reduction in precipitation between the 1970s and 1990s but a precipitation increase thereafter (see Fig. 6a). This recent increase in precipitation is particularly notable during autumn (Fig. 6a), though more importantly for glacier mass balance, winter precipitation also increases during this period, likely contributing directly to increased snow accumulation. The temperature data (Fig. 6b) appear to show a warming trend from the 1950s to the late 1990s, which has continued (with some fluctuations) to the present day. Most significantly, there has been a sharp increase in average summer temperatures (June, July and August) since the early 21st century (see Fig. 6b).

Glacier size (area, perimeter, length and altitudinal range), altitude (maximum, median and minimum), surface slope and aspect (based on the glaciers mapped in 2000) were compared to absolute and relative changes in glacier surface area across the peninsula. These comparisons reveal a strong negative and statistically significant (p< 0.001) correlation between the 2000 to 2014 change in glacier surface area (in km2) and glacier area (r=-0.75), length (r=-0.65) and altitudinal range (r=-0.48) in 2000 (Fig. 7a–c). When change in area is expressed as a percentage of the original glacier size (rather than total area), these correlations become positive (r=0.24, 0.34 and 0.34 for surface area, length and altitudinal range, respectively) and are weakened (though they remain statistically significant) (Fig. 7d–f). There is also a negative and statistically significant correlation between glacier surface area change (in km2) and the maximum (r=-0.37) and median (r=-0.25) glacier altitude. Again, when area change is expressed as a percentage, these correlations become positive (r=0.23 and 0.17 for maximum and median, respectively) and are weakened (though they remain statistically significant). When glacier minimum altitude or surface slope are considered, there is no statistically significant relationship with glacier area loss (when expressed as total area, or as percentage change). Glaciers are predominantly found with an aspect bias towards northerly and western directions (54.98 %), and glacier aspect shows a weak (r=0.15) but statistically significant relationship with changes in glacier surface area, though only when expressed in km2, rather than as a percentage. Similarly, analysis of insolation patterns reveals a weak but statistically significant correlation (r=0.18) with change in glacier surface area, but only when expressed in km2, rather than as a percentage.

Based on mapping in 2000, ∼ 8 % (n=52) of the peninsula's glaciers are classed as debris-covered. When the glacier population is split into “debris-covered” and “non-debris-covered” samples in this way, the former lose on average ∼ 11 % (0.62 km2) of their surface area between 2000 and 2014, while the latter lose ∼ 44 % (0.24 km2). Thus, over the 2000–2014 period, debris-covered glaciers lose less surface area than non-debris-covered examples, though this relationship is only statistically significant when values are expressed as a percentage, rather than in km2.

Based on the data presented in the Catalogue of Glaciers in the USSR (see Sect. 2.3) and the findings of the present study, there was a ∼ 11.2 % decline in glacier surface area on the Kamchatka Peninsula between the 1950s (exact date unspecified) and 2000 and a further ∼ 23.6 % loss between 2000 and 2014. Assuming linear trends, this indicates an area-loss rate of ∼ 0.24–0.29 % a-1 between 1950 and 2000 and a notable acceleration to ∼ 1.76 % a-1 since 2000. The reduction in glacier area during the late 20th century coincides with negative trends in glacier mass balance on the peninsula (see Barr and Solomina, 2014). Unfortunately, since 2000, mass balance data have not been collected for Kamchatka's glaciers, and it is not possible to assess whether accelerated mass loss over the early 21st century has coincided with the accelerated glacier shrinkage identified here.

Similar rates of retreat have been found for glaciers in other Asian mountain ranges. For example, in the Altai Mountains, Narozhniy and Zemtsov (2011) document a 10.2 % decrease in glacier area between 1956 and 2008, while Kamp and Pan (2015) document a 13 % decrease between 1998/2001 and 2010/2011. In the Tien Shan, Farinotti et al. (2015) document an 18 ± 6 % decrease in glacier surface area between 1961 and 2012. In the Kodar Mountains, Stokes et al. (2013) document a 44 % decrease in the area of exposed glacial ice between 1963 and 2010, with a 40 % loss since 1995, and in the Caucasus Mountains, Tielidze (2016) documents a 36.9±2.2 % decline in glacier surface area between 1960 and 2014, though regional differences are noted.

In terms of glacier surface area, it is notable that our estimate from 2014 (592.9 ± 20.4 km2) differs significantly from Earl and Gardner's (2016) estimate (770.3 km2). This we attribute to their semi-automated approach to mapping (i.e. NDSI), which can lead to an overestimation of glacier area as a result of snow patches being erroneously classified as glaciers (Man et al., 2014) (see Fig. 8), combined with the fact that their inventory was generated from a composite of satellite images, meaning their mapping does not reflect glacier extent during a single specified year (as noted in Sect. 2.3).

In the small number of studies to consider the issue (noted in Sect. 2.3), the retreat of Kamchatka's glaciers over recent decades has typically been attributed to variations in climate. For example, Yamaguchi et al. (2008) consider the retreat of Koryto glacier between 1711 and 2000 to be a result of decreased precipitation over this period. Similarly, in the northern part of the Sredinny Range, Muraviev and Nosenko (2013) note that from 1950 to 2002, average summer temperatures increased, while solid precipitation (snowfall) decreased, suggesting that this caused the retreat of the region's glaciers.

Based on climatic trends identified from the NCEP data, it might be argued that the loss of glacier surface area across Kamchatka between the 1950s and 2000 reflects the combined influence of rising temperatures and declining precipitation. However, since 2000, despite an increase in precipitation, glacier area loss has continued and appears to have accelerated. This might reflect a delayed response to earlier drier conditions, but is also likely driven by a notable increase in temperatures since the mid-1990s. In particular, the pronounced rise in summer temperature (see Fig. 6c) is likely to have increased the intensity of melt, whilst rising autumn temperatures (Fig. 6b) may have lengthened the ablation season (simultaneously shortening the accumulation season). Therefore, it would appear that temperature has been the primary control on early 21st century glacier fluctuations in Kamchatka, with rising temperatures driving a notable decline in glacier surface area, despite a corresponding rise in precipitation (even during winter).

Despite evidence for climatic control over glacier fluctuations in Kamchatka (outlined in Sect. 5.2), the individual response of these glaciers is likely to have been modulated by other local non-climatic factors (Evans, 2006; Tennant et al., 2012; Stokes et al., 2013). These factors potentially include glacier size (area, perimeter length and altitudinal range), altitude surface slope, and aspect, as well as volcanic controls and debris cover (Huss, 2012; Fischer et al., 2015).