Plain Language Summary
Rock glaciers are debris-ice landforms and indicators of the status of
perennially frozen ground, as known as permafrost, which is warming and
thawing under climate change. The West Kunlun is among the driest
mountain ranges in Asia where permafrost has been changing over the past
decades and the information of rock glaciers is completely lacking. In
this paper, we developed an effective workflow for mapping rock glaciers
in a semi-automated manner and characterized their geomorphology and
kinematics. The compiled dataset allows further investigation on rock
glaciers for multiple scientific motivations such as geohazard
management, water resource assessment, and permafrost change monitoring.
The documented geomorphic characteristics provide insights into the
genesis and evolution of rock glaciers in the arid mountains.
1 Introduction
Rock glaciers are debris-ice landforms widely distributed in areas of
mountain permafrost globally (Ballantyne 2018). Rock glaciers have drawn
a lot of research interest since their first identification at the
beginning of the 20th century (Capps 1910), because they serve as
visible indicators for alpine permafrost which is defined by its
underground temperature and has been warming and undergoing degradation
(Barsch 1996; Biskaborn et al. 2019). Inventorying rock glaciers is
therefore motivated by producing baseline knowledge for addressing
various scientific questions associated with alpine permafrost, such as
indicating permafrost occurrence through the rock glacier distribution,
characterizing permafrost changes in the warming climate, and assessing
the future hydrological significance of rock glaciers. Several studies
have revealed that multi-annual acceleration of rock glaciers is
synchronous with the rise of air and ground temperatures (Haeberli et
al. 2006; Delaloye et al. 2010; Delaloye et al. 2013; Sorg et al. 2015;
Marcer et al. 2021), and their short-term velocity variations are
sensitive to the pore pressure in the shear horizon which is adjusted by
the precipitation and snow melt conditions (Ikeda et al. 2008; Müller et
al. 2016; Wirz et al. 2016; Cicoira et al. 2019a; Cicoira et al. 2019b;
Kenner et al. 2019). Hence rock glacier inventories are valuable
databases for studying how climatic factors cause permafrost changes
manifesting in landform kinematics which can be quantified continuously
and remotely. Moreover, rock glaciers can contain massive amounts of
ground ice and contribute significantly to hydrological systems in some
catchments, such as the Andes, Himalayas, and Sierra Nevada (Azócar and
Brenning 2010; Millar et al. 2013; Geiger et al. 2014; Jones et al.
2018; Schaffer et al. 2019; Jones et al. 2021). A comprehensive
inventory of rock glaciers lays the foundation for estimating the
potential water storage and evaluating their future role in maintaining
water supplies.
Numerous efforts have been put into inventorying rock glaciers in
various mountain ranges worldwide in the past several decades, such as
in Central Europe (Chueca 1992; Roer and Nyenhuis 2007; Scotti et al.
2013; Onaca et al. 2017), South America (Brenning 2005; Falaschi et al.
2014; Rangecroft et al. 2014; Villarroel et al. 2018), and North America
(Ellis and Calkin 1979; Janke 2007; Millar and Westfall 2008; Liu et al.
2013). Rock glaciers are abundant in mountainous western China where a
vast area of alpine permafrost is underlying and undergoing accelerated
degradation in response to the warming climate (Yang et al. 2010; Cheng
et al. 2019; Yang et al. 2019; Yao et al. 2019; Zhao and Sheng 2019; Ni
et al. 2020; Zhao et al. 2020; IPCC 2021). However, few regional-scale
inventories of rock glaciers have been compiled until recently (Schmid
et al. 2015; Wang et al. 2017; Ran and Liu 2018), which hinders rock
glaciers functioning as a permafrost indicator. Such lack of knowledge
is attributed to the following reasons: (1) rock glaciers in western
China are mostly situated in remote and harsh environment where early in
situ investigations are scarce and limited to case studies or small
catchment-scale research (e.g., Cui 1985; Cui and Zhu 1988; Zhu et al.
1996; Harris et al. 1998); (2) mapping rock glaciers conventionally
relies on manually detecting and outlining the landforms from optical
images (Schmid et al. 2015), which is labor-intensive to apply to large
permafrost region (e.g., West Kunlun Mountains) following an exhaustive
strategy; (3) contentious opinions of identifying rock glaciers exist
due to the complexity of the landforms (Harris et al. 1998; Berthling
2011; Hu et al. 2021), which obscures the definition of rock glaciers
and makes it challenging to recognize the landforms.
To address these problems, recent research progress in compiling rock
glacier inventories includes (1) integrating InSAR techniques to
facilitate active rock glacier identification and kinematics
quantification (e.g., Liu et al. 2013; Barboux et al. 2014; Wang et al.
2017; Cai et al. 2021; Reinosch et al. 2021; Zhang et al. 2021); (2)
implementing Convolutional Neural Networks (CNN) to demonstrate the
feasibility of automating rock glacier delineation (Robson et al. 2020)
or to improve the consistency of existing rock glacier inventories
(Erharter et al. 2022); and (3) establishing widely accepted
inventorying guidelines by the international rock glacier research
community (RGIK, 2021).
Here we combine the InSAR technique and a state-of-the-art deep learning
network, namely DeepLabv3+ (Chen et al. 2018), to map rock glaciers
across the West Kunlun Mountains of China where widespread permafrost is
warming (Li 1986; Cheng et al. 2019), and knowledge of rock glaciers is
completely lacking. Manual delineation of rock glaciers based on InSAR
and high-resolution optical imagery in this study is guided by the
baseline concepts proposed by the International Permafrost Association
(IPA) Action Group on rock glaciers to ensure a standard high-quality
dataset utilized to train the deep learning network, and thus, the final
mapping results (RGIK, 2021). We adopted the deep learning method to
improve the mapping efficiency by automating the identification and
delineation tasks, and more importantly, to generate a more
comprehensive geodatabase by overcoming the limitations of InSAR-based
method (Cai et al., 2021).
This study aims to develop an automated approach to map rock glaciers on
a regional scale in western China, i.e., the West Kunlun Mountains. By
producing the first automatically mapped inventory at the mountain-range
scale, we demonstrate the effectiveness of using a deep-learning-based
method to delineate rock glaciers in a consistent manner across the vast
study area. We provide essential attributes to the mapped landforms
according to the inventorying guidelines. We also conduct statistical
analyses to summarize the spatial distribution and geomorphologic
characteristics of the mapped rock glaciers. The compiled inventory will
provide baseline knowledge for conducting long-term studies of rock
glaciers and permafrost in a changing climate.
2 Study area
The West Kunlun is a major mountain range situated in the northwest of
Tibetan Plateau, extending ~800 km from the eastern
margin of Pamir Plateau to the Keriya Pass of Kunlun Mountains, with a
total study area of ~124,000 km2(74–81.5°E, 35–39.5°N) (Figure 1). The elevation of the study region
ranges between 3,000 m and 7,500 m.
Across the vast study area, a cold desert climate (Köppen climate
classification BWk) is dominant (Peel et al. 2007). Climatic conditions
of the western part are revealed by the record of the nearest
meteorological station in Tashikurgan (75.23°E, 37.77°N; 3090 m a.s.l.)
during 1957–2017: the mean annual air temperature (MAAT) and mean
annual accumulated precipitation are 4.2°C and 51 mm, respectively (data
source: China Meteorological Administration,http://data.cma.cn/ ). The study area has been warming at a rate
of ~0.033°C/yr during the past six decades, similar to
the average warming rate (0.031°C/yr) across the entire plateau (Zhang
et al. 2020). In the eastern part, the MAAT is -6 °C and the annual
precipitation is 103.3 mm, as reported by the Tianshuihai meteorological
station (79.55°E, 35.36°N; 4844 m a.s.l) from 2015 to 2018 (Zhao et al.
2021).
The easternmost part of the study region is overlapped with the West
Kunlun permafrost survey area (78.8–81.4°E, 34.5–36.0°N; 4,200–6,100
m a.s.l.) established by the Cryosphere Research Station (CRS) on the
Qinghai-Tibet Plateau, Chinese Academy of Sciences, where in situ
observations are available to represent the state of permafrost in the
West Kunlun. Ice-rich permafrost is widely distributed in the survey
area (Zhao and Sheng, 2019). The mean annual ground temperature (MAGT)
is higher than -2.7°C as revealed by borehole measurements and
permafrost was warming at an average rate of 0.11°C/10 yr from 2010 to
2017 (Cheng et al. 2019; Zhao and Sheng, 2019). The lowest altitudinal
limit of permafrost occurrence is between 4,650 m and 4,800 m depending
on different slope aspects according to previous field surveys focusing
on a subregion of the West Kunlun (Li et al. 2012).