Approximately
25 % of the glacierized area in the Everest region is covered by debris,
yet the surface mass balance of debris-covered portions of these glaciers has
not been measured directly. In this study, ground-based measurements of
surface elevation and ice depth are combined with terrestrial photogrammetry,
unmanned aerial vehicle (UAV) and satellite elevation models to derive the
surface mass balance of the debris-covered tongue of Changri Nup Glacier,
located in the Everest region. Over the debris-covered tongue, the mean
elevation change between 2011 and 2015 is
Predicting the future of the Himalayan cryosphere and water resources depends on understanding the impact of climate change on glaciers (Lutz et al., 2014). About 14–18 % of the total glacierized area in the Himalayas is debris covered (Kääb et al., 2012). This ratio increases to between 25 and 36 % in the Everest region of Nepal (Nuimura et al., 2012; Shea et al., 2015; Thakuri et al., 2014). However, the role played by debris on the surface mass balance of glaciers and, in turn, on the glacier response to climate change remains unclear (Kääb et al., 2012). Indeed, this debris layer insulates the glacier surface from the atmosphere when it reaches a sufficient thickness and complicates the response to climate change compared to clean-ice glaciers (Jouvet et al., 2011; Kirkbride and Deline, 2013; Østrem, 1959; Pellicciotti et al., 2015).
In comparison with debris-free (clean) ice, melt is enhanced when the surface is covered by a very thin layer of debris (1–2 cm) as a result of increased absorption of solar radiation and related heat transfer. On the other hand, debris layers thicker than a few centimetres reduce ice melt rates as less surface heat will be conducted through the debris layer and transferred to the ice (Østrem, 1959; Nakawo and Young, 1981; Mattson, 1993; Kayastha et al., 2000; Mihalcea et al., 2006; Nicholson and Benn, 2006; Reid and Brock, 2010; Lambrecht et al., 2011; Lejeune et al., 2013; Brock et al., 2010). However, several studies based on remote sensing data have shown comparable rates of elevation change on debris-covered and clean-ice glaciers at similar altitudes in the Himalayas and Karakoram (Gardelle et al., 2013; Kääb et al., 2012). Some studies hypothesized that this “debris-cover anomaly” could be explained by increased ice cliff ablation and englacial melt on debris-covered glaciers (Buri et al., 2015; Immerzeel et al., 2014; Inoue and Yoshida, 1980; Miles et al., 2016). Yet Ragettli et al. (2015) observed different thinning rates on clean and debris-covered glaciers at similar elevations in Langtang Valley (Nepal) using remote sensing techniques. To date, the debris-cover anomaly hypothesis has not been tested with field-based observation.
To add complexity, the surface area of debris-covered tongues has increased
in recent decades due to glacier surface lowering and unstable adjacent
slopes, processes that are likely associated with climate change
(Bhambri et al., 2011; Bolch et
al., 2008; Schmidt and Nüsser, 2009; Shukla et al., 2009). Between 1962
and 2011, the proportion of Everest region glaciers covered by rock debris
increased by 17.6
For these reasons, it is urgent to determine the mass balance sensitivity of debris-covered glaciers to climate change. Unfortunately, there are very few surface mass balance measurements which have been carried out on debris-covered glaciers (Mihalcea et al., 2006). First, the surface mass balance field measurements from ablation stakes are sparse. Second, these measurements cannot be expected to be representative given that the ice ablation exhibits a strong spatial variability depending on the debris thickness or type (Azam et al., 2014; Berthier and Vincent, 2012; Hagg et al., 2008; Inoue and Yoshida, 1980; Mihalcea et al., 2006), and measurements can only be made at locations where the ice surface can be reached. Furthermore, geodetic measurements based on the difference between digital elevation models (DEMs) derived from satellite or aerial imagery only determine surface height change and glacier-wide mass balance and are typically unable to resolve the spatial pattern of surface mass balance (Immerzeel et al., 2014).
In this paper, we assess the surface mass balance of the entire debris-covered tongue of a Himalayan glacier (Changri Nup Glacier) using the ice flux method (Berthier and Vincent, 2012; Nuimura et al., 2011; Nuth et al., 2012). DEMs constructed from (i) terrestrial photogrammetry surveys in 2011 and 2014, (ii) an unmanned aerial vehicle (UAV) survey in 2015 and (iii) satellite stereo pair imagery acquired in 2009 and 2014 are used to estimate changes in glacier thickness. The surface mass balance of the debris-covered area is inferred from the difference between the ice flux measured through a cross section at the upper limit of the debris-covered area and the observed elevation changes. Finally, we compare our field-based estimate of the debris-covered glacier mass balance against surface mass balances observed at nearby debris-free glaciers and quantify the overall reduction in ablation due to debris cover.
Debris-covered Changri Nup Glacier (27.987
Study area overview showing the general location (inset map) and delineation of debris-free and debris-covered Changri Nup glaciers. Background is from ESRI basemap imagery.
A smaller debris-free glacier known locally as White Changri Nup Glacier
(Fig. 1; 27.97
Additional mass balance data used in this study are taken from nearby
Pokalde (27.9
A suite of field-based and remote sensing methods were used to calculate the mass balance of clean and debris-covered Changri Nup glaciers. These included photogrammetric surveys, field-based differential global position system (DGPS) and ground-penetrating radar (GPR) surveys, unmanned aerial vehicle (UAV) surveys, point surface mass balance (SMB) measurements and satellite-derived height changes.
For all ground control points (GCPs) and ablation stake measurements, we
used a Topcon DGPS unit. Occupation times were typically 1 min with
1 s sampling, and the number of visible satellites (GPS and GLONAS) was
greater than 6. GCP and ablation stake locations have an intrinsic accuracy
of
Terrestrial photogrammetric surveys were carried out in the last week of October 2011 and in the last week of November 2014. The photographs were made using a Canon EOS5D Mark II digital reflex camera with a Canon 50 mm f/2.8 AF fixed focus lens. The 21.1 million pixel images were captured in a raw uncompressed format.
Oblique terrestrial photographs that covered most of the debris-covered
tongue were collected from three bases and under similar conditions in
October 2011 and November 2014. Camera positions were between 1100 m and
2000 m from the glacier, which results in a ground-scaled pixel size of 0.14
to 0.25 m. Camera locations were 280, 264 and 253 m apart and the base
formed by the camera locations was roughly perpendicular to the sightings.
The base-to-distance ratio (
Photogrammetric measurements were used to build DEMs over the glacier tongue
downstream of cross section M (Fig. 2). To geometrically correct the
images, GCPs that included 28 large (2
Map of debris-covered Changri Nup Glacier showing the glacierized area (light blue), DGPS cross sections (blue), delineated debris-covered tongue (dashed black line) and UAV imagery extent (black line). TP is terrestrial photogrammetry and background is from ESRI basemap imagery. TP control markers are painted crosses, and TP control features are characteristic boulders.
GPR measurements were performed on 25 October 2011 to measure ice thickness
on the transverse cross section M, located upstream of the debris-covered
area at ca. 5525 m a.s.l. (Fig. 2). We used a pulse radar system (Icefield
Instruments, Canada) based on the Narod transmitter (Narod and
Clarke, 1994) with separate transmitter and receiver, a frequency centred
near 4.2 MHz and an antenna length of 10 m. Transmitter and receiver were
towed in snow sledges along the transverse profile, separated by a fixed
distance of 20 m, and measurements were made every 10 m. The positions of
the receiver and the transmitter were recorded with static DGPS
measurements and have an accuracy of
To estimate the ice depth, the speed of electromagnetic wave propagation in
ice was assumed to be 167 m
DGPS measurements were collected on six transverse profiles located on the tongue of the glacier in the last weeks of October 2011, November 2014 and November 2015 (Fig. 2). To measure ice flow velocities between 2011 and 2015 along profile M, we made repeat DGPS measurements of (i) ablation stakes and (ii) six painted rocks. Seven bamboo stakes were installed up to a depth of 6 m in 2011 along profile M (Fig. 2). Six of these stakes were replaced in 2014 at their original locations, and all were resurveyed in 2015. To delineate the active part of the glacier from stagnant ice, velocities were also obtained with repeat DGPS measurements performed on more than 75 painted or recognizable rocks in 2011, 2012, 2014 and 2015 (Fig. 2). Some measurements performed on painted stones were discarded when the stones slipped on ice or rolled down on steep slopes.
A detailed survey of the glacier surface was conducted on 22–24 November 2015 using the senseFly eBee UAV. Over the course of five survey flights, a total of 582 photos were collected with the onboard Canon Ixus from an average altitude of 325 m above the glacier surface (Fig. 2). Prior to the survey flights, we collected DGPS measurements of 34 ground control points that consisted of (i) red fabrics with painted white squares and (ii) white crosses used also for the photogrammetry (Fig. 2). Twenty-four GCPs were used to process the imagery and create a DEM with Agisoft, and 10 GCPs were reserved as independent checks on the accuracy of the DEM. The original resolutions of the orthomosaic and DEM are 10 and 20 cm respectively.
The images from the survey were processed using the Structure for Motion (SfM) algorithm, which is implemented in the software package Agisoft Photoscan Professional version 1.2.0 (Agisoft, 2014). First, a feature recognition and matching algorithm was applied on a set of overlapping pictures resulting in a set of points in 3-D space derived from the matching features and camera positions. This positioning of the sparse point cloud was then corrected using the DGPS control point measurements. Multiview stereo techniques were then used to generate a dense point cloud of the glacier surface. This dense point cloud was used to construct the DEM, which was then used to generate a geometrically corrected mosaic of all input images. A detailed description of the processing steps can be found in Kraaijenbrink et al. (2016).
Based on the 10 independent GCPs, the average error in the UAV-derived DEM
is
Map of measured glacier surface velocities (m a
To calculate surface height changes over a larger area and longer time period, we also used DEMs derived from two satellite stereo acquisitions. The 2014 DEM was derived from two SPOT7 images acquired on 28 October 2014. The ground resolution of each image is 1.5 m and the base to height ratio between the two images is 0.24. The images are slightly covered by snow above approximately 4800 m a.s.l. The 2014 DEM was derived without GCPs using the commercial software PCI Geomatica 2015. The 2009 DEM was derived from two SPOT5 images acquired on 28 October and 4 November 2009. The ground resolution of each image is 2.5 m and the base to height ratio is 0.45. The 2009 DEM was derived using 23 GCPs extracted from the 2014 SPOT7 DEM and the corresponding 1.5 m orthoimage. Output resolution of both DEMs was set to 6 m.
The two DEMs were horizontally shifted to minimize the standard deviation of
elevation differences on stable terrain (Berthier et al.,
2007). Glaciers were masked out using the inventory from Gardelle et al. (2013). We excluded off-glacier pixels for which the elevation difference
was larger than 3 times the normalized median absolute deviation. The
vertical shift between the two DEMs was calculated as the median elevation
difference on flat and stable zones near the glaciers (1.67 km
The uncertainty of the elevation difference between the two DEMs is assessed
from the statistical distribution of the elevation differences over stable
terrain (Magnússon et al., 2016;
Rolstad et al., 2009). The standard deviation of elevation differences on
stable ground (
Point SMB measurements, with uncertainties of
We estimate the ice flux
The demarcation between active glacier flow and stagnant glacier ice downstream of cross section M is crucial for our SMB assessment (Eq. 2). However, the strongly heterogeneous debris layer covering this tongue may mask the true glacier margin, and buried ice may not be connected to the active part of the glacier. Indeed, this glacier has been in retreat over the last decades and many stagnant ice areas are no longer connected. From remote sensing optical images, it is very challenging to delineate the margins of debris-covered glaciers (Paul et al., 2013). For instance, several previous studies (Quincey et al., 2009; Rowan et al., 2015) have indicated that Changri Nup Glacier was connected to the Khumbu Glacier, a distance of nearly 3.5 km from the terminus delineated in this study. Similarly, the inventories most commonly used in this region connect the debris-covered Changri Nup, the debris-free Changri Nup and the Changri Shar glaciers Bolch et al., 2011; Gardelle et al., 2013; Nuimura et al., 2012, 2015).
For Changri Nup Glacier, zones of active glacier flow were delineated using
horizontal velocities derived from repeat DGPS measurements. Velocities
derived from freely available optical imagery (e.g. Landsat) cannot resolve
velocities less than 5–10 m a
Despite the presence of stagnant ice far downstream of the terminus, the
delineation of the terminus is clear in most places (dashed line in Fig. 3). For example, a proglacial stream flows on a thick layer of sand in a
flat area immediately below the delineated snout. However, at some locations
the boundary between active and stagnant ice is unclear, and here we
spatially interpolated measured ice flow velocities using a Kriging
interpolation method (Fig. 3). With this approach and obvious features in
the field (slope change, visible ice), the debris-covered ablation area was
estimated to be 1.494 km
Mean elevation changes (m a
The ice flux at cross section M (Fig. 2) was obtained by multiplying the
surface area of this cross section with the mean cross-sectional ice flow
velocity. From the GPR measurements (Fig. 4a), the maximum observed ice
thickness is 150 m, and the cross-sectional area was estimated to be 79 300 m
A mean cross-sectional ice velocity can be calculated from surface
velocities and assumptions about the relation between mean surface velocity
and depth-averaged velocity. Here, two approaches are used to estimate the
mean surface velocity. The first uses all surface velocities observed along
the flux gate between 2011 and 2015, and the mean surface velocity is
calculated by fitting a second-order polynomial function (Fig. 4b).
Unfortunately, surface velocities were not measured near the glacier
margins. We thus assume that ice flow velocity decreases linearly to zero at
the margin of the glacier (Fig. 4b), and obtain a mean surface velocity of
9.7 m a
The next step is the conversion from mean surface velocity to depth-averaged
velocity. Without basal sliding, theoretical calculations suggest that the
depth-averaged velocity is 80 % of the mean surface velocity (for
Elevation changes are directly measured along DGPS profiles and calculated by differencing DEMs from terrestrial photogrammetry, UAV surveys and satellite stereo pair imagery.
For the mostly debris-free region between profiles M and N, where
photogrammetric measurements are not available, we calculated a mean
elevation change from repeat DGPS measurements along profiles M and N. In
general, elevation changes in clean-ice areas are expected to have low
cross-glacier variability (Berthier and
Vincent, 2012; Fischer et al., 2005; Vincent et al., 2009). At profile M,
this is confirmed by the similarity in elevation profiles between years, and
the mean rate of elevation change is
Surface elevation profiles (m a.s.l.) for 2011 (black), 2014 (blue) and 2015 (red) from DGPS measurements (dots), terrestrial photogrammetry (black and blue lines) and UAV survey (red lines). Note that the right (left) bank is on the left (right) of each profile.
Downstream of profile N, elevation changes were calculated for two periods
(2011–2014 and 2011–2015) by differencing DEMs obtained from terrestrial
photogrammetric measurements and the UAV survey. Due to terrain obstruction,
thickness changes can only be calculated for 60 % of the ablation area
downstream of profile N when photogrammetry data is used. Our results show a
highly heterogeneous down-wasting pattern of the tongue of Changri Nup
Glacier (Figs. 5 and 6). Overall, a negative change in surface elevation is
observed over the monitored area. Mean elevation changes of
Elevation changes obtained from photogrammetry and UAV data have been validated using DGPS measurements and high-resolution satellite stereo pair imagery. First, we directly compare point elevation data from photogrammetric and UAV DEMs with DGPS elevations observed at independent GCPs (i.e. those not used in DEM generation). Differences between DGPS and photogrammetric elevations for 25 independent GCPs near the terminus and profile R have a root mean square error (RMSE) of 0.63 m. A similar comparison between DGPS spot heights and UAV-derived elevations at 10 independent points gives a RMSE of 0.25 m. A direct comparison between photogrammetric and DGPS elevations on cross-glacier profiles (Fig. 5) shows that differences are generally less than 1 m.
We also compare surface elevation changes obtained from photogrammetric DEM
differencing and repeat DGPS measurements (Table 1). As photogrammetric
measurements are incomplete along the transverse profiles due to terrain
obstruction, we only consider the sections of the profiles where both DGPS
and photogrammetric elevation data are available. At profiles R, P and
Z the differences in rates of surface elevation change measured with the two
approaches are approximately 0.1 m a
Elevation changes (m a
From repeat DGPS profiles, we found a mean elevation change (2011–2014) of
As a final test, elevation changes downvalley of the delineated glacier
terminus were calculated from photogrammetry and UAV data. In this small
(0.014 km
Finally, photogrammetric and UAV-derived elevation changes (2011–2015) can
be compared to elevation changes measured from the stereo-pair DEMs
(2009–2014), though the periods of measurement are slightly different. From
2009 to 2014, we find a mean elevation change of
Assuming that the thickness changes described above are representative of
the total area below the flux gate (below profile M), we calculate an
area-weighted elevation change equal to
The total uncertainty in our estimated SMB is related to uncertainties in
(i) the delineation of the surface area of the tongue, (ii) the elevation
changes of the tongue, (iii) the thickness of cross section M and (iv) the
mean cross-sectional velocity at cross section M. Total uncertainty was
assessed following the calculation of the area-averaged surface mass balance
(
Using Eq. (3), the overall squared error (
The uncertainty relative to the cross-sectional area of profile M has been
assessed using an ice thickness uncertainty of 10 m (Bauder et al., 2003).
Uncertainty relative to the mean cross-sectional velocity is assumed to be
10 % of the calculated velocity (Huss et al., 2007).
Following Eq. (4), the overall error
High-resolution surface elevation changes derived in this study from photogrammetry, UAV surveys and satellite stereo pairs highlight the fact that elevation changes over debris-covered glaciers are highly spatially variable (Fig. 6). This is already well known over debris-covered glaciers where elevation changes depend on both the variability in debris thickness and the spatial distribution of ponds or cliffs (Immerzeel et al., 2014; Nuimura et al., 2012). However, this study shows that neither repeat DGPS measurements obtained from transverse profiles nor an ablation stake network are sufficient to obtain a representative mean elevation change or surface mass balance over debris-covered glaciers. The spatial variability in height changes (Fig. 6) also precludes comparisons between direct (glaciological) observations of SMB on clean and debris-covered glaciers.
The overall surface lowering rates and mass balances of debris-covered glaciers remains controversial. Several recent studies suggested that elevation changes on debris-covered and debris-free glaciers are similar in the Himalayas and Karakoram (Gardelle et al., 2013; Kääb et al., 2012; Pellicciotti et al., 2015). Conversely, Nuimura et al. (2012) showed that the debris-covered areas are subject to higher rates of lowering than debris-free areas in Khumbu region, though the 400 m difference in mean elevation between the debris-covered and debris-free areas (5102 and 5521 m a.s.l. respectively) may account for this conclusion.
Comparisons between the mass balances of debris-covered and debris-free glaciers (as opposed to comparisons of surface elevation change only) are hindered by methodological deficiencies and uncertainties. First, geodetic studies typically provide only glacier- or region-wide mass balances based on elevation changes (Bolch et al., 2008, 2011; Nuimura et al., 2012). Geodetic methods are unable to determine a separate surface mass balance for debris-covered areas, because they do not account for the emergence velocity. Moreover, the size, altitude and dynamic behaviour of clean and debris-covered glaciers are different and the comparison between glacier-wide mass balances cannot distinguish ablation rates between debris-covered and debris-free areas. In addition, most of these studies in Nepal have been carried out on catchments with a predominance of debris-covered glaciers (Bolch et al., 2011) and cannot be compared with catchments dominated by debris-free glaciers. Second, the uncertainties related to these remote sensing methods (e.g. the delineation of the glaciers, elevation bias due to the radar penetration into the ice, elevation change assessment and snow density) are large (Pellicciotti et al., 2015). Finally, the regional average mass balances obtained from geodetic methods mask strong differences among glaciers and cannot be used to compare ablation rates between debris-covered and debris-free ice.
In contrast with full-glacier geodetic results, our method based on ice flux
calculations and surface lowering observations from photogrammetric and UAV
DEMs enables the calculation of an average SMB (
As our estimate of SMB incorporates the spatial variability in surface lowering, we compare the area-averaged SMB obtained for Changri Nup Glacier with direct SMB measurements from debris-free ice and glaciers in the region (Fig. 7). These include point SMB measurements from profile M (Fig. 2), White Changri Nup Glacier (5390 to 5600 m a.s.l.), Pokalde Glacier (5505 to 5636 m a.s.l.), and Mera and Naulek glaciers (5112 to 5415 m a.s.l.). Also displayed on Fig. 7 are the 2014–2015 point SMB measurements from the stake farm located in the debris-covered area of Changri Nup Glacier (Fig. 2).
Surface mass balance as a function of elevation for Changri Nup, Mera and Pokalde glaciers over the period 2011–2015. The upper cross corresponds to the mean surface mass balance obtained on the debris-covered tongue. The blue dashed line represents the mean vertical gradient of mass balance observed at White Changri Nup glaciers and is extrapolated from the mean of SMB measurements at profile M. The lower large cross corresponds to the surface mass balance of a hypothetical clean-ice glacier. Note that surface mass balances of the stake farm on Changri Nup Glacier were measured in 2014–2015 only. The heights of each cross correspond to the uncertainty on inferred SMB.
The average SMB assessed over the debris-covered Changri Nup Glacier tongue
(
To estimate the effect debris cover has on the SMB, we estimate the average
SMB for Changri Nup with the vertical gradient of SMB
(
Several studies have suggested that supraglacial ponds and ice cliffs considerably enhance glacier ablation for debris-covered glaciers (Benn et al., 2012; Brun et al., 2016; Buri et al., 2015; Miles et al., 2016; Sakai et al., 2000; Zhang et al., 2011; Juen et al., 2014). Although supraglacial ponds and ice cliffs are present on the debris-covered tongue of the Changri Nup Glacier, the overall mass loss is still considerably reduced due to the debris cover and we conclude that the insulating effect dominates at this site.
This conclusion seems to contradict the results of several studies (Gardelle et al., 2013; Kääb et al., 2012) which revealed comparable rates of elevation changes on debris-covered and clean-ice glaciers. However, these previous results came from geodetic measurements and do not account for the effect of ice dynamics (i.e. difference in emergence velocities between debris-covered and clean-ice glaciers). To overcome this issue, Kääb et al., 2012) compared elevation changes between debris-covered and clean ice using neighbouring ICESat footprints (separated by approximately 1 km), in an attempt to minimize differences in emergence velocity. Still, the geodetic method does not permit direct comparisons of ablation rates, and only the ice flux method employed here allows for reliable estimates of average glacier mass balance over the terminus and comparisons with other glaciers.
The calculated surface mass balance of the debris-covered area of Changri Nup
Glacier has been obtained from (i) the ice flux at a cross section close to
the boundary between debris-free area and debris-covered area and (ii) elevation changes of the tongue. From the calculated ice flux we estimate an
average emergence velocity for the debris-covered tongue of
A vertical mass balance gradient derived from nearby debris-free glaciers in
the studied region suggests that the average SMB would be
Our method to obtain the surface mass balance of the debris-covered area is reliable. However, the application of the method requires accurate and extensive field data and is hard to transpose to numerous or larger glaciers. A precise delineation of the debris-covered glacier tongue is required. For this purpose, ice flow velocities derived from DGPS field measurements are needed given that ice flow velocities are very low in the debris-covered areas in the vicinity of the margins. In addition, GPR measurements performed on a transverse cross section are also mandatory.
Our results have important implications for studies modelling the future evolution of debris-covered glaciers (Rowan et al., 2015; Shea et al., 2015). An empirical model of debris-covered glacier melt that takes into consideration the relevant processes (surface melt, englacial/subglacial melt and ice cliff and surface pond migration and density) will be an important development.
The DEM data can be accessed upon request by contacting Christian Vincent (christian.vincent@univ-grenoble-alpes.fr).
This work has been supported by the French Service d'Observation GLACIOCLIM, the French National Research Agency (ANR) through ANR-09-CEP-005-01-PAPRIKA and ANR-13-SENV-0005-04-PRESHINE and has been supported by a grant from Labex OSUG@2020 (Investissements d'avenir – ANR10 LABX56). This study was carried out within the framework of the Ev-K2-CNR Project in collaboration with the Nepal Academy of Science and Technology as foreseen by the Memorandum of Understanding between Nepal and Italy, and thanks to contributions from the Italian National Research Council, the Italian Ministry of Education, University and Research and the Italian Ministry of Foreign Affairs. Funding for the UAV survey was generously provided by the United Kingdom Department for International Development (DFID) and by the Ministry of Foreign Affairs, Government of Norway. This project has received funding from the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation programme (grant agreement No. 676819). EB acknowledges support from the French Space Agency (CNES) through the TOSCA Top Glaciers project. SPOT5 HRG images were obtained thanks to ISIS-CNES project #510. This work was supported by public funds received in the framework of GEOSUD, a project (ANR-10-EQPX-20) of the program “Investissements d'Avenir” managed by the French National Research Agency. The International Centre for Integrated Mountain Development is funded in part by the governments of Afghanistan, Bangladesh, Bhutan, China, India, Myanmar, Nepal and Pakistan. The views expressed are those of the authors and do not necessarily reflect their organizations or funding institutions. We thank the French private company SINTEGRA for its useful support. We thank J. O. Hagen Editor, J. Steiner, and an anonymous reviewer whose thorough comments improved the quality of the manuscript. Edited by: J. O. Hagen Reviewed by: J. Steiner and one anonymous referee