Previous geodetic estimates of mass changes in the Karakoram
revealed balanced budgets or a possible slight mass gain since
Overview map of the study region.
Glacier meltwater is of high importance for the run-off of the Indus River
(Immerzeel et al., 2010) but the exact glacier share is not known. This is
partly due to the lack of knowledge about precipitation, snow cover and snow
water equivalent, but also about the mass balance, characteristics and
responses of glaciers to climate change. Karakoram glaciers, which occupy a
large portion of the glacierized area of the Indus basin, have recently shown
unusual behaviour: on average no significant area changes but frequent
advances and surge activities have been observed during the last decades
(Bhambri et al., 2013; Bolch et al., 2012; Copland et al., 2011; Hewitt,
2011; Rankl et al., 2014). Geodetic mass estimates revealed balanced glacier
mass budgets or even slight mass gain since
Declassified stereo satellite images from the 1960s and 1970s, such as Corona
KH-4 and Hexagon KH-9, have been proven to be suitable for generating digital
terrain models (DTMs) and assessing glacier mass changes since the 1960s (Bolch
et al., 2008; Pieczonka et al., 2013; Maurer et al., 2016). Hence, the aim of
this study is to revisit existing information and extend the time series back
to some of the earliest available satellite imagery. We focus on the Hunza
catchment in the central Karakoram (Fig. 1) where high heterogeneity of
glacier behaviour was found in previous studies (e.g. Bolch et al., 2012;
Quincey and Luckman, 2014). Moreover, suitable Hexagon KH-9 data from the
1970s and recent stereo data from
Glacier mass budget for different periods, selected glaciers and glacier types.
The SRTM DTM, with a spatial resolution of 1 arcsec (
Two high-resolution Cartosat-1 stereo scenes captured on 11 July 2010
(Table S1) were used to compare and investigate the consistency of the
results obtained with the lower-resolution ASTER DTM. Cartosat-1 (IRS-P5) was
launched by Indian Space Research Organisation (ISRO) in May 2005. The
satellite has two high-resolution (2.5 m) panchromatic sensors recording
stereo images along the track (Titarov, 2008). The major advantage of this
data set, besides the high spatial resolution, is the 12-bit radiometric
resolution. Unfortunately, the spatial coverage is relatively small (
Declassified Hexagon KH-9 imagery which has a ground resolution of about 8 m
and a coverage of about
The ICIMOD glacier inventory (Bajracharya and Shrestha, 2011), also available
through the GLIMS database (
All KH-9 DTMs were generated with Erdas Imagine 2014 Photogrammetry Suite
using the frame camera model with a focal length of 30.5 cm. Image
preprocessing includes the elimination of internal distortions based on the
regularly distributed réseau crosses (originally included to correct film
distortion effects) and their removal thereafter, following Pieczonka et
al. (2013). GCPs were collected from Landsat 7 ETM
Fiducials were measured manually considering the principal point in the image
centre. All stereo images have been processed with a root mean squared error
(RMSE) of
The Cartosat-1 stereo pairs have been processed using PCI OrthoEngine 2014 with 32 and 35 GCPs. To improve the quality of the DTMs, image enhancement techniques were applied prior to DTM generation in order to overcome low image contrast and temporal differences in image acquisition. The RSME varied between 0.3 and 3.9 pixels (Table S2). The spatial resolution of all DTMs was chosen as 30 m.
In order to obtain reliable results on glacier surface elevation changes, the
DTMs must be properly co-registered (Nuth and Kääb, 2011). As we
observed tilts when differencing the original DTMs, we first minimized
elevation differences between the different DTMs with respect to the SRTM1
master DTM by applying a first-order trend correction. We considered only
elevation differences (
To calculate surface elevation changes, we subtract each older DTM from a more recent DTM and mosaicked the difference grids to facilitate processing. Where the ASTER DTMs from different time periods overlapped, we calculate a weighted mean elevation change based on the time of acquisition and glacier coverage (Fig. S1, Table S3).
Data voids and mismatches that result in incorrect elevation values can occur
in areas with low image contrast such as cast shadows and bright snow.
Mismatches due to snow in the accumulation regions led to unrealistic low
elevation values using KH-9 data that would subsequently lead to unrealistic
surface-lowering values in parts of the accumulation region (Pieczonka and
Bolch, 2015). However, thickness change distributions for glaciers with
negative mass budgets typically have a minimum lowering at the glacier head
with increasing values towards the glacier front following a non-linear trend
(Huss et al., 2010). This pattern is different for surging glaciers that
often exhibit high positive
As both debris-covered glaciers and surge-type glaciers are common in the
investigated region we could not apply a general threshold to remove
The penetration of the radar beams into firn and snow has to be considered in
case of the comparison of DTMs generated from microwave data such as the
SRTM1. However, the value can only be estimated as it depends on several
unknown parameters (e.g. snow depth and characteristics) and is therefore one
major source of uncertainty (Kääb et al., 2015; Gardelle et al.,
2013). We applied a penetration correction of 2.4
Elevation difference between the ASTER and Hexagon KH-9 DTMs (above), the Hexagon KH-9 and the SRTM1 DTMs (below, left) and the SRTM1 and ASTER DTMs (below, right). Black dots indicate surge-type glaciers, the numbers indicate the larger glaciers: 1 is Batura, 2 is Pasu, 3 is Barpu, 4 is Hispar, 5 is Yazghil, 6 is Khurdopin, 7 is Vijerab.
There is no best method to estimate the uncertainty (
The overall uncertainty of the DTM difference is the average of
We did not apply a seasonality correction as most of the images were acquired close to the end of the ablation period but assume the effect is well within the considered uncertainty.
Results for the period 1999 to
Longitudinal profiles of surface elevation changes for selected glaciers for the entire period (1973–2009) and the subperiods 1973–1999 nd 1999–2009.
Our extended time series indicates on average balanced budgets of the
glaciers in Hunza valley for the
period 1973 to 1999 and slight, but insignificant mass loss for 1999 to
Most glaciers experienced similar mass budgets for both investigation periods. However, it seems that some glaciers had more negative budgets post-2000. This is especially true for the debris-covered Batura Glacier, the tongue of which showed significant lowering during 1999–2009. Over the entire study period there is no significant difference in the mass budgets of surge-type and non-surge-type glaciers, a result also found by Gardelle et al. (2013), and surge-type glaciers also showed more negative values in the recent period. Khurdopin Glacier, for example, experienced a significant thickening near the snout and a significant lowering around the ELA, both of which combined to produce a near-balanced mass budget between 1973 and 1999. Significant lowering on the lower part of the tongue resulted in a net mass loss between 1999 and 2009 (Figs. 2, 3).
For the recent period, the westernmost tributary of Hispar Glacier shows a
clear sign of surging with significant elevation gain of more than 100 m at
the confluence with the main glacier. This tributary also clearly thickened
in the middle reaches during the earlier period (Fig. 2). These elevation
change characteristics are likely due to the fact that the ice builds up in
the reservoir area during the quiescent phase and the active surge event
transfers ice mass to lower elevations where it is then more prone to
melting. The very high surface-lowering rates observed in the middle part of
the tongue of Hispar Glacier also hints to a past surge event. Covering a short period after the active
surge only would therefore probably overestimate mass loss. It is, hence,
recommended to cover the entire surge cycle of surge-type glaciers in order
to be better able to relate their mass budgets to climate forcing. This is
the case for most of the glaciers in our study region as we cover a period of
almost 40 years, and the surge periodicity in the Karakoram is rather short
with averages between
Declassified KH-9 Hexagon data have been proven to be valuable for assessing
geodetic glacier mass budgets (Pieczonka and Bolch, 2015; Pieczonka et al.,
2013; Maurer et al., 2016). The main challenges of obtaining accurate results
are miscorrelations in the accumulation regions of glaciers and significant
tilts and shifts of elevation trends making careful co-registration and
postprocessing necessary. This leads to higher uncertainties compared to more
recent data with a similar spatial resolution. The glacier volume changes
calculated based on the automatically derived ASTER DTMs (AST14DEM) and the
SRTM DTM were similar to those using better quality higher-resolution SPOT5
DTMs for a similar region (Gardelle et al., 2013). Off-glacier DTM
differences show good agreement in general, but so do regions with higher
deviations where the quality of the ASTER DTMs was lower (e.g. the western
part of the study areas; Fig. S1). We found no significant difference between
the mass budget results of Khurdopin Glacier calculated using Cartosat-1 data
and the values calculated using the ASTER data (
A further source of uncertainty is the data voids in the original SRTM data
and voids due to the outlier filtering. About 20 % of the total
glacierized area for all analysed periods were identified as outliers and
filled afterwards by kriging
interpolation. The voids are almost entirely located in the accumulation
regions where surface elevation
changes are relatively small (e.g. Schwitter and Raymond, 1993) and where we
restricted the maximum possible deviation. To assess the influence of the
filling on the result we calculated the elevation change value for the whole
basin and the period 1973–2009 (a) without void filling (result: mean
elevation difference
One of the major sources of uncertainty is the penetration of the radar beam
into snow and ice when using the SRTM DTM. Gardelle et al. (2013) estimated
a mean penetration of 3.4 m with values up to more than 9 m in the
accumulation region. This value is higher than the 2.4 m we applied here
following Kääb et al. (2012). However, applying the higher
penetration value would only lead to a slight difference in mass change of
Based on 1973 Hexagon, SRTM and
The utilized data and resulting DEM differencing grids are available from the
first author upon request. Certain licence restrictions may apply to
satellite imagery. The glacier outlines from the ICIMOD inventory can be
downloaded through
Tobias Bolch designed the study, performed all analyses, generated the figures and wrote the draft of the manuscript. Kriti Mukherjee generated the raw Hexagon and Cartosat-1 DTMs and co-registered the data. Tobias Bolch and Tino Pieczonka supported the generation of the DTMs and the co-registration. Joseph Shea contributed to the study design, and all authors contributed to the final form of the article.
The authors declare that they have no conflict of interest.
This study was performed within the Cryosphere Initiative of the
International Center for Integrated Mountain Development (ICIMOD), with the
targeted support of the UK Department for International Development (DFID).
ICIMOD is funded in part by the governments of Afghanistan, Bangladesh,
Bhutan, China, India, Myanmar, Nepal and Pakistan, and the Cryosphere
Initiative is funded by the Norwegian Ministry of Foreign affairs. The views
expressed are those of the authors and do not necessarily reflect their
organizations or funding institutions. T. Bolch acknowledges funding through
the ESA project Glaciers_cci (4000109873/14/I-NB), Deutsche
Forschungsgemeinschaft (DFG) and The University of Colorado CHARIS project
(Contribution to High Asia Runoff from Ice and Snow – (