Glacial-dominated areas pose unique challenges to downstream communities in
adapting to recent and continuing global climate change, including increased
threats of glacial lake outburst floods (GLOFs) that can increase risk due
to flooding of downstream communities and cause substantial impacts on
regional social, environmental and economic systems. The Imja glacial lake
(or Imja Tsho) in Nepal, which has the potential to generate a GLOF, was studied
using a two-dimensional debris-flow inundation model in order to evaluate
the effectiveness of proposed measures to reduce possible flooding impacts
to downstream communities by lowering the lake level. The results indicate
that only minor flood impact reduction is achieved in the downstream
community of Dingboche with modest (
Recent worldwide retreat of glaciers (WGMS, 2013) has been very evident in the Mt. Everest region of Nepal where glacial lakes continue to form and grow, significantly increasing the risk of glacial lake outburst floods (GLOFs) (Bajracharya et al., 2007a; ICIMOD, 2011; Ives et al., 2010; Shrestha and Aryal, 2011). Many of these lakes are considered potentially dangerous (Bajracharya et al., 2007a; Bolch et al., 2008; Watanabe et al., 2009; ICIMOD, 2011); therefore, risk and vulnerability assessments of communities and assets located downstream of glacial lakes in this region have become necessary. Remedial actions have been taken to reduce the risk of GLOFs, in one case at Tsho Rolpa lake (Rana et al., 2000) and another under design at Imja Tsho (or Imja Lake) (UNDP, 2013). In the region near Imja Tsho (Fig. 1), there have been 2 GLOFs in recent decades (Nare in 1977 and Dig Tsho in 1985) resulting in significant damage to farms, villages, trails and some loss of life (Buchroithner et al., 1982; Ives, 1986; Vuichard and Zimmermann, 1986).
Location of Imja Tsho in the Khumbu region of Nepal.
Imja Tsho, located in the Khumbu region (27.9
The characterization of the risk of Imja Tsho is somewhat controversial, with some researchers declaring it to be relatively dangerous (Hammond, 1988; Kattelmann, 2003; Ives et al., 2010), and others concluding that it may be stable (Fujita et al., 2009; Watanabe et al., 2009; ICIMOD, 2011). ICIMOD (2011) identified Imja Tsho as one of six high-priority glacial lakes in Nepal that require detailed investigation, while other studies have stated that Imja Tsho is safe (Fujita et al., 2013) or very low risk (Hambrey et al., 2008). These conflicting classifications are confusing and can be misleading to the general public and communities downstream, who are the stakeholders these studies are meant to assist. Imja Tsho is among six glacial lakes identified in the Nepal National Adaptation Plan of Action (NAPA) as having at the most immediate risk of bursting (MoE, 2010). The United Nations Development Program (UNDP) is implementing the Community Based Flood and Glacial Lake Outburst Risk Reduction Project in an effort to reduce the possible risk to downstream communities posed by the lake. According to the UNDP project strategy (UNDP, 2013), the “GLOF risks arising from Imja Tsho will be significantly reduced by reducing the lake volume through an artificial controlled drainage system combined with a community-based early warning system.” They recommend lowering the lake level by at least 3 m to achieve this risk reduction.
The village of Dingboche on the Imja Khola showing the river (right) and the relative height from the river course to the arable land and houses (about 22 m).
Dingboche is probably the most risk prone area from a potential Imja GLOF.
The villages of Chukkung (
Bridge across the Dudh Koshi at the Star Lion Resort near the village of Phakding showing the bridge and the relative height from the river course to the arable land and houses (about 10 to 13 m).
In this paper we present a new, two-dimensional debris-flow model for predicting the potential GLOF hazard from Imja Tsho in terms of inundation depth in downstream communities and present a measure of uncertainty in the GLOF inundation predictions. We analyze four scenarios: current lake conditions, and three risk mitigation scenarios with the lake water level lowered 3, 10 or 20 m below the current level. Finally, we discuss possible methods for lowering the lake water level to reduce the GLOF hazard. To the authors' knowledge, this is the first attempt to quantify the impact that various flood control alternatives would have on potential GLOF damage in downstream villages.
To model the propagation of a GLOF from Imja Tsho to the downstream
community of Dingboche, we used two digital elevation models (DEMs) for the
Imja Tsho GLOF model: (1) a 5 m
Roughness coefficients for eight categories of land cover in the basin were assigned using land cover maps derived from 2006 ASTER imagery for the Sagarmatha National park (Bajracharya and Uddin, 2010). These values agreed well with those that Cenderelli and Wohl (2001) calculated for the Imja Khola (0.15 and 0.30 for the riverbed and floodplain, respectively), and values that the FLO-2D manual (FLO-2D, 2012) recommends for the types of land cover found in the basin.
Somos-Valenzuela et al. (2014) conducted a bathymetric survey of Imja Tsho
in 2012 and estimated the lake volume was 61.7
In order for a GLOF to occur from Imja Tsho, a triggering event is necessary. Such triggers may include slow melting of the ice core within the damming moraine, seepage and piping through the dam and earthquakes (Kattelmann and Watanabe, 1998; Somos-Valenzuela et al., 2013). Other factors that may trigger a GLOF from Imja Tsho include excessive rain and potential blockage of the outlet that may be produced by the ice that has calved at the glacier terminus.
We observed seepage from the base of the southern portion of the damming
moraine during five visits to the lake between 2011 and 2014. In September
2013 the seepage was measured using a tape measure and porTable velocity
meter (Global Water Flow Probe FP111 turboprop positive displacement sensor
with a range of 0.1–6.1 m s
To model a potential moraine breach initiating a GLOF from Imja Tsho, we use a combination of moraine breach analysis tools. First, the shape, final size and failure time of the breach are estimated from empirical equations. Failure time is the time needed for complete development of the breach from the initial breakthrough to the end of lateral enlargement (Froehlich, 2008). Second, these parameters are used in a Hydrologic Engineering Center – River Analysis System (HEC-RAS) dam breach model (USACE, 2010) to simulate the breach hydrograph, which is then used as input to a 2-D downstream inundation model.
There are a number of empirical dam breach equations in the literature
(Wahl, 2010; Westoby et al., 2014). However, the equations developed by
Froehlich (1995) were selected for use here because Wahl (2004) found these
equations to have the lowest uncertainty among a large number of equations
studied. Froehlich's equations (Froehlich, 1995) were used to predict breach
width (
Moraine parameter values used in the breaching equations.
Froehlich's equations provide estimates of the breaching parameters, but to
simulate the downstream inundation, the full hydrograph of the breaching
event is needed. To obtain full breach hydrographs (lower, predicted and
upper), we use the HEC-RAS dam break model (USACE, 2010) with the breach
width (
FLO-2D is used to calculate the flooding downstream of Imja Tsho due to a
potential GLOF with the breaching hydrograph discussed in the previous
section. The model is suitable to simulate the propagation of the debris
flow (FLO-2D, 2012), since the effects of sediments and debris have been
shown to be very important factors in GLOF events (Osti and Egashira, 2009).
Although the geometry of the grid within FLO-2D is two dimensional, the flow
is modeled in eight directions and the model solves the one-dimensional
Saint Venant equation independently in each direction. The continuity and
momentum equations are solved with a central, finite difference method using
an explicit time-stepping scheme. The total friction slope can be expressed
as (O'Brien et al., 1993; Julien, 2010; FLO-2D, 2012)
The Imja GLOF model described above was used to model four scenarios of a GLOF occurring from Imja Tsho: current lake conditions with the water surface at 5010 m, and three flood mitigation scenarios with the lake water level lowered 3, 10 and 20 m. The flood mitigation scenarios represent possible lake lowering efforts starting with the current UNDP project to lower the lake at least 3 m. In case this scenario does not provide any significant flood reduction downstream at Dingboche, the other scenarios can provide some guidance as to how much more the lake might need to be lowered to achieve reduced risk. It is important to note that no other studies have analyzed the potential benefits of lowering Imja Tsho, and the selection of a preferred lake lowering alternative needs to be based on such an analysis as that presented here.
Prediction error and uncertainty bands for the Froehlich breaching equations.
Breach parameter expected values and uncertainty bands for four scenarios.
Table 3 shows the results of using Eqs. (
Inundation model expected values and uncertainty bands at Imja Tsho, Dingboche and Phakding for current lake conditions.
Discharge hydrographs for potential moraine breaches at Imja Tsho were
computed using the HEC-RAS dam break module and the lower bound, predicted
and upper bound breach parameters under the current conditions scenario (see
Table 4). The HEC-RAS hydrograph peak discharges were matched to the peak
discharge values in Table 3 by adjusting the failure time within the range
shown in the table. The predicted peak discharge is 8394 m
The FLO-2D inundation model was used to compute the results of the four potential Imja Tsho GLOF scenarios. The first scenario considers the lake in its current condition with the lake level at 5010 m above mean sea level. Then alternatives with lake levels 3, 10, and 20 m lower than this were considered.
Upper bound, expected and lower bound GLOF hydrograph at Dingboche (cross section shown in Fig. 4) under current lake conditions.
Upper bound, expected and lower bound GLOF flood stage at Dingboche (cross section shown in Fig. 4) under current conditions.
The results (lower bound, expected value, upper bound) of modeling a
potential GLOF from Imja Tsho under current conditions are shown in Table 4
and Figs. 4–6 at Dingboche (at the cross section indicated in Fig. 4).
Table 4 also shows the flood arrival time, peak time, peak stage and flood
peak discharge just downstream of Imja Tsho and at Dingboche. Fig. 4 shows
the expected GLOF discharge hydrograph and bounds at Dingboche. The flood
arrives at Dingboche 1 h after the breaching begins (range 0.6–1.9 h),
peaks at 1.3 h (range 0.8–2.8 h) and is over after about 7 h.
Figure 5 shows the flood stage at Dingboche (upper bound, expected value,
lower bound). The highest expected flood stage is 22.4 m (range 18.4–26.4 m)
and the peak flow is 7544 m
Inundation at Dingboche under current lake conditions:
The inundated area at Dingboche was mapped in Geographic Information System (GIS) and shows that, under the expected value simulation, most of the inundation is in the farming terrace areas and not the main lodges and other infrastructure along the primary trekking trail through the village (see Fig. 6a). With no lake lowering, about 9.4 ha of farmland will be inundated and 29 structures impacted (see Table 5, 0 m lowering scenario).
Inundation model results at Dingboche and Phakding for lake lowering scenarios under current conditions.
* Relative to the peak value for 0 m lake lowering.
Further downstream at Phakding, the flooding also has an impact on potential
flooding. At Phakding, the flood arrives 3.1 h (range 2.4–4.4 h) after
the breaching begins and peaks at 3.2 h (range 2.6–4.7 h) with a peak
discharge of 3412 m
Upper bound, expected and lower bound GLOF hydrograph under current conditions at Phakding.
A proposal to reduce the risk of a GLOF from Imja Tsho that is currently (2014) under implementation is to lower the water level of the lake at least 3 m (UNDP, 2013). The Imja GLOF model was used to assess the potential flood reduction at Dingboche if such a plan were to be implemented. To this end, the model was run with lake levels 3, 10 and 20 m lower than the current conditions scenario level (5010 m). The results for these scenarios are shown in Table 5 and Figs. 8–10. Figures 8 and 9 show the hydrographs and flood stage, respectively, at Dingboche for the 0, 3, 10 and 20 m lake lowering scenarios. Figure 10 maps the inundation depth at Dingboche for the different lake lowering scenarios. Lowering the lake 3 m (Fig. 10a) results in a 2.7 % reduction in the peak flood depth at Dingboche (compared to the 0 m lowering scenario) with the peak flood height lowering 0.6 m (from 22.4 to 21.8 m). This flood height still leads to significant inundation of homes and farmlands. With the lake lowered by 3 m about 8.6 ha of farmland and 25 structures are impacted by the flooding. In contrast, lowering the lake 10 m (Fig. 10b) or 20 m (Fig. 10c) results in a 14 and 36 % flood height reduction, respectively, at Dingboche, with respective peak flood heights of 19.2 and 14.4 m. These scenarios lead to considerable reduction in inundated area, especially the 20 m lowering scenario where the flood stays mostly in the historic flood plain of the river, inundates little farmland and floods no structures. When the lake is lowered 10 m about 4 ha of farmland will be inundated and 18 structures impacted and at 20 m lowering about 1 ha of farmland will be inundated and 0 structures impacted. Additionally, the peak discharge is reduced by 6.5, 40.6 and 73.8 % as a result of lowering the lake by 3, 10 or 20 m, respectively (Table 5).
GLOF hydrographs at Dingboche under current lake conditions for 0, 3, 10 and 20 m lake lowering scenarios.
GLOF flood stage at Dingboche under current lake conditions for 0 m (expected), 3, 10 and 20 m lake lowering scenarios.
Further downstream at Phakding, the lake lowering scenarios have an impact on potential flooding as well (see Fig. 11). Lowering the lake 3 m results in a 13.9 % reduction of peak flow at Phakding. In contrast, lowering the lake 10 m results in 49.3 % reduction at Phakding, and lowering 20 m results in an 81.6 % reduction at Phakding (see Table 5).
Inundation depth at Dingboche under current lake conditions:
HEC-RAS has been used to simulate GLOFs in the Nepal Himalaya by several other researchers. Cenderelli and Wohl (2001) used one-dimensional, steady-flow HEC-RAS modeling to estimate peak discharges of the 1977 Nare and 1985 Dig Tsho GLOFs. Osti and Egashira (2009) modeled the 1998 GLOF at Tam Pokhari using HEC-RAS to perform one-dimensional, unsteady-flow calculations; however, the model could not be used to model the debris flow. They note that the GLOF was very strong and damaging, and the peak discharge was much higher than the results of the water-only computations. Other models have been used to model GLOFs. Dwivedi (2007) modeled the 1998 Tam Pokhari GLOF using the SOBEK flood model (Alkema and Middelkoop, 2005) and a 40 m resolution DEM. Several breaching scenarios were simulated, and eroded sediments were not considered in the model. Shrestha et al. (2013) modeled a potential debris-flow GLOF originating at Tsho Rolpa in the Rolwaling Valley of Nepal including the moraine breaching process due to an assumed seepage failure. Laboratory experiments were conducted to verify the model and good agreement with model results were obtained. Worni et al. (2012) who used the dynamic, erosion-based dam break model BASic EnvironMENT for simulation of environmental flow and natural hazard simulation (BASEMENT) (Faeh et al., 2011) to model a debris-flow GLOF in the Argentinian Patagonia. BASEMENT is a tool for the analysis of breaching processes of non-cohesive earthen dam structures and water-sediment flows (Volz et al., 2010). Schneider et al. (2014) simulated the cascade of mass movement processes of an avalanche triggered GLOF from Lake 513 in the Cordillera Blanca of Peru by coupling different physically based numerical models. Glacial lake dam overtopping hydrographs and water volumes were used as input for downstream debris-flow modeling with RAMMS (Christen et al., 2010).
GLOF hydrographs at Phakding under current conditions for 0, 3, 10 and 20 m lake lowering scenarios.
Bajracharya et al. (2007b) developed a water-flow GLOF model for Imja Tsho
that simulated flood flow in the Imja Khola and Dudh Kosi from the lake to
about 45 km downstream of the outlet near Phakding. That one-dimensional
model used a DEM derived from satellite data (30 m
We can compare the results of ICIMOD (2011) with those reported here, since
this is being asked of the consultants working on the UNDP Imja Tsho risk
reduction project (UNDP, 2013). For the breaching process, ICIMOD reported a
breaching time of 2.9 h with a peak discharge of 5817 m
To date there is no agreed upon set of hazard indicators for Imja Tsho, or other potentially dangerous glacial lakes for that matter (Somos-Valenzuela et al., 2013); however, the definition of GLOF hazard from Imja Tsho was discussed in consultations with community members in Dingboche in September 2012 and subsequently. The hazard of an Imja Tsho GLOF did not exist 30 years ago (see Somos-Valenzuela et al. (2014) and Watanabe et al. (2009) for images showing the evolution of the lake), and the community members' vulnerabilities stem from the location of their homes and farms relative to the flood plain. For them, hazard is having their farms or homes flooded or washed away, a prospect that they want reduced and preferably eliminated (Somos-Valenzuela et al., 2013).
There are many impacts that might occur from a rapid uncontrolled GLOF from Imja Tsho that will be felt downstream, in particular there are several stretches of the main trekking trail from Namche Bazar to Lukla that run quite close to the river and would be washed away, as they were in the 1985 Dig Tsho GLOF (Ives, 1986; Vuichard and Zimmermann, 1986). In addition, some parts of villages along the trail have fields and houses near the river, and they may also be impacted. The feasibility of possible remedial actions at Imja Tsho to reduce the hazard to downstream communities was evaluated. One scenario that appears to have significant risk reduction possibility is the slow, controlled lowering of the lake by 20 m. Lowering the lake by any amount will reduce the probability of GLOF occurrence for at least two reasons: the hydrostatic pressure on the west to east width of the moraine at the water level would be increased and more difficult to breach, and reducing the GLOF's occurrence probability would decrease the hazard level downstream.
Of the methods to reduce glacial lake risk, e.g., relocation of people and assets from the flood path, strengthening the lake outlet (Kattelmann and Watanabe, 1998), the one that has been employed the most is lowering the lake level. This has been used at nearly 40 dangerous glacial lakes in Peru since the 1950s (Portocarrero, 2014). Typically, the lake is lowered to a safe level by siphoning or draining and then excavating the damming moraine and installing a drainage channel at the desired elevation. Often, a reinforced earthen dam is then constructed to replace the original unconsolidated moraine dam, such that if a surge wave overtops the dam it will contain much of the excess water and not fail from erosion (Somos-Valenzuela et al., 2013). In order to lower glacial lakes, siphons are often used, e.g., at Hualcán Lake (Lake 513) in Peru (Portocarrero, 2014). Lowering glacial lakes more than about 5 m is infeasible at altitudes of 5000 m, but greater lowering can be achieved using an incremental method as discussed below (Somos-Valenzuela et al., 2013).
The UNDP Imja Tsho project suggests that lowering the lake at least 3 m will achieve significant risk reduction downstream (UNDP, 2013), but there is no requirement to estimate the remaining risk when the lake is lowered to different levels (3, 10, 20 m). This work has attempted to analyze this question. Siphons may be used at Imja Tsho to progressively lower the level in 3–5 m increments followed by excavation of the lake outlet.
Previous studies have suggested deepening and strengthening the outlet of
Imja Tsho (Maskey, 2012). However, there are many difficulties in
implementing this method. First, the natural flow of the outlet must be
interrupted somehow in order to perform any excavation of the channel. In
the siphoning method discussed above, the lake is lowered and then the
outlet can be excavated to that level without needing to divert the outlet
flow. One method that has been proposed is to build a coffer dam and divert
water to flow over another part of the damming moraine and then excavate the
existing outlet channel of the lake to increase its depth and discharge
(Maskey, 2012). The difficulties of employing this method include (1) possibly
encountering ice during the excavation, significantly weakening the
moraine and possibly inducing a GLOF; (2) diversion of the outlet flow might
cause excessive erosion that could weaken the moraine and potentially lead
to a GLOF; and (3) the existence of small ponds in the outlet complex that
are separated with shallow necks (with as little as 1.5 m depth) through
which the lake water flows might prevent the draining of the lake unless
they were also excavated (Somos-Valenzuela et al., 2013). The difficulty of
encountering buried ice exists in any method employing excavation of the
damming moraine and the moraine must be examined in detail with geophysical
methods before this can be done safely. The example of Tsho Rolpa is being
used as a model lake lowering system for Imja Tsho (UNDP, 2013). An outlet
channel was constructed at Tsho Rolpa and 3 m lowering was achieved;
however, the design called for lowering the lake by 20 m which was never
attempted because of funding limitations (Rana et al., 2000; Mool et al.,
2001). Our results show that lowering Imja Tsho 3 m would not lead to a
significant inundation reduction downstream. The lake should be lowered at
least 10 m and probably 20 m to achieve significant hazard reduction.
Lowering Imja Tsho will require draining: (1) the normal inflow to the lake,
(2) the volume of the lake expansion during the time of drainage and (3) the
volume of the lake necessary to achieve 3 m lowering per drainage cycle
(Somos-Valenzuela et al., 2013). Lake discharge was measured in May 2012 by
a group from Kathmandu University using a tracer dilution method (Maskey,
2012) and by the authors using a timed float method (Somos-Valenzuela et
al., 2013). In both cases the flow was found to be approximately 1 m
The authors again measured the flow with a flow meter in September 2013 and
found the discharge to be approximately 2 m
Methods for reducing the downstream inundation hazard from a GLOF originating at Imja Tsho in Nepal were explored. A two-dimensional debris-flow model was developed to assess the downstream inundation. Inundation reducing scenarios were analyzed and an alternative under design, lowering the lake at least 3 m, was found not to have significant flood reduction benefits. The results indicate that the lake needs to be lowered about 20 m in order to completely reduce the impacts that a GLOF could have at Dingboche and further downstream. The results show that a GLOF occurring under the current lake conditions would result in inundation of much of the farming areas (about 9.4 ha and 29 structures impacted) at Dingboche but not the main lodges and other infrastructure along the primary trekking trail through the village. Lowering lake 3 m does not change this result much, but 10 m lowering reduces the impact substantially with about 4 ha of farmland and 18 structures impacted, and at 20 m lowering almost all impact at Dingboche is prevented. All cases involving lowering the lake would require a coordinated sequence of siphoning to lower the water level in 3 m increments, followed by outlet excavation to maintain the new level. The process would be repeated as needed to reach the desired lake level.
The authors acknowledge the support of the USAID Climate Change Resilient Development (CCRD) project and the Fulbright Foundation for the support of Somos-Valenzuela. The support of the software developers of FLO-2D made much of the work reported here possible. The comments of T. Watanabe, S. Bajracharya and an anonymous reviewer are greatly appreciated. Edited by: M. Mikos