Page 2
Page 3
Page 4
Page 5
Page 6
Page 7
Page 8
Page 9
Page 10
Page 11
Page 12
Page 13
Page 14
Page 15
Page 16
References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.
The Cryosphere, 11, 2527–2542, 2017 https://doi.org/10.5194/tc-11-2527-2017 © Author(s) 2017. This work is distributed under the Creative Commons Attribution 3.0 License. 1,2 1 2 2 1 1 2 1 1 Defu Zou , Lin Zhao , Yu Sheng , Ji Chen , Guojie Hu , Tonghua Wu , Jichun Wu , Changwei Xie , Xiaodong Wu , 1 1 1 1 1 1 1 1 Qiangqiang Pang , Wu Wang , Erji Du , Wangping Li , Guangyue Liu , Jing Li , Yanhui Qin , Yongping Qiao , 1 1 2 Zhiwei Wang , Jianzong Shi , and Guodong Cheng Cryosphere Research Station on Qinghai–Xizang Plateau, State Key Laboratory of Cryospheric Science, Northwest Institute of Eco–Environment and Resources (NIEER), Chinese Academy of Sciences (CAS), Lanzhou, 730000, China State Key Laboratory of Frozen Soil Engineering, NIEER, CAS, Lanzhou, 730000, China Correspondence to: Lin Zhao (linzhao@lzb.ac.cn) Received: 30 July 2016 – Discussion started: 20 October 2016 Revised: 25 September 2017 – Accepted: 4 October 2017 – Published: 8 November 2017 Abstract. The Tibetan Plateau (TP) has the largest areas of permafrost distribution and basic data for use in future re- permafrost terrain in the mid- and low-latitude regions of the search on the Tibetan Plateau permafrost. world. Some permafrost distribution maps have been com- piled but, due to limited data sources, ambiguous criteria, inadequate validation, and deficiency of high-quality spa- 1 Introduction tial data sets, there is high uncertainty in the mapping of the permafrost distribution on the TP. We generated a new Permafrost is a major component of the cryosphere and is permafrost map based on freezing and thawing indices from sensitive to climate changes (Wu et al., 2002b; Haeberli modified Moderate Resolution Imaging Spectroradiometer and Hohmann, 2008; Li et al., 2008; Gruber, 2012). Due (MODIS) land surface temperatures (LSTs) and validated to its unique and extremely high altitude (mean elevation this map using various ground-based data sets. The soil ther- over 4000 m) and low mean annual air temperatures (gen- mal properties of five soil types across the TP were esti- erally lower than2 C with an intra-annual amplitude over mated according to an empirical equation and soil prop- 20 C in the permafrost region), the Tibetan Plateau (TP) pos- erties (moisture content and bulk density). The tempera- sesses the largest areas of permafrost in the mid- and low- ture at the top of permafrost (TTOP) model was applied latitude regions of the world (Zhou et al., 2000; Zhao et al., to simulate the permafrost distribution. Permafrost, season- 2004, 2010; Yang et al., 2010). Permafrost and its dynamics ally frozen ground, and unfrozen ground covered areas of 6 2 6 2 complicate the water and energy exchange between soil and 1.06 10 km (0.97–1.15 10 km , 90 % confidence in- 6 6 2 atmosphere and thereby introduce greater uncertainty into terval) (40 %), 1.46 10 (56 %), and 0.03 10 km (1 %), global climate models (GCMs) that predict climate change respectively, excluding glaciers and lakes. Ground-based ob- (Romanovsky et al., 2002; Smith and Riseborough, 2002; servations of the permafrost distribution across the five in- Cheng and Wu, 2007; Riseborough et al., 2008; Zhao et al., vestigated regions (IRs, located in the transition zones of the 2010). To generate better quantitative simulations, a more ac- permafrost and seasonally frozen ground) and three highway curate TP permafrost distribution is needed. An accurate con- transects (across the entire permafrost regions from north to temporary permafrost distribution map is also important as a south) were used to validate the model. Validation results baseline with which to estimate future permafrost degrada- showed that the kappa coefficient varied from 0.38 to 0.78 tion. with a mean of 0.57 for the five IRs and 0.62 to 0.74 with A significant amount of research has been conducted on a mean of 0.68 within the three transects. Compared with the permafrost distribution of the TP, and many permafrost earlier studies, the TTOP modelling results show greater ac- maps have been compiled to evaluate the distribution and curacy. The results provide more detailed information on the thermal states of permafrost (Shi and Mi, 1988; Li and Cheng, 1996; Brown et al., 1997, 1998; Qiu et al., 2000; Published by Copernicus Publications on behalf of the European Geosciences Union. 2528 D. Zou et al.: A new map of permafrost distribution on the Tibetan Plateau Wang et al., 2006). These maps have been used to study snow cover) calculated from measured radiance; Gillespie, the responses of permafrost to climate change (Ran et al., 2014) products derived from different satellite images have 2012). However, considerable differences in the permafrost been applied to global and regional permafrost distribution areas and boundaries occur in these maps, from 1.12 10 research (Kääb, 2008; Hachem et al., 2009; Nguyen et al., 6 2 to 1.50 10 km , due to different data collection periods, 2009; Langer et al., 2010; Westermann et al., 2012, 2015). data sets, and methods (Yang et al., 2010; Ran et al., 2012). These products are effective alternatives to GST, especially These maps represent different assessments of the permafrost for regions with limited in situ observations, such as the TP. distribution on the TP at different times. However, the LSTs observed by satellite sensors are instan- Permafrost boundaries in the earlier maps have often been taneous values at the passing times and must be transformed plotted on topographic maps by hand using conventional into mean daily temperature to serve as the thermal state of cartographic techniques (Tong and Li, 1983; Shi and Mi, each day before use. Wang et al. (2011) averaged the twice- 1988; Li and Cheng, 1996). The standard and most widely daily LST observations of the Moderate Resolution Imag- used map is the Map of Permafrost on the Qinghai-Tibetan ing Spectroradiometer (MODIS) sensors on board the Terra Plateau (Li and Cheng, 1996). In this map, the permafrost satellite to drive the TTOP model. Their results show a sys- boundaries were mainly determined using air temperature tematic bias with the ground observations because of differ- isotherms combined with field data, satellite images, and ent observation times (Wang et al., 2011). In addition, the many relevant maps. After 2000, GIS techniques have been limited availability of soil thermal property spatial data sets applied to permafrost mapping of the TP. Simple empirical creates another problem when modelling permafrost distri- models with minimal data requirements were established to bution. Most soil surveys have been carried out in seasonally consider the permafrost characteristics on the TP. These in- frozen ground or permafrost along the Qinghai–Tibet High- clude the elevation model (Li and Cheng, 1999) and mean an- way (Li et al., 2014, 2015a) rather than in permafrost regions nual ground temperature (MAGT) (Nan et al., 2002). Mean- in the plateau hinterland due to the harsh climate and incon- while, some models with simplified physical processes appli- venient access. Therefore, soil thermal properties have gen- cable to high-latitude permafrost were transferred to simulate erally been estimated via soil types generated from a limited the permafrost distribution on the TP. Examples are the frost number of plateau geologic classification studies (Wang et index (Nelson and Outcalt, 1983) and the temperature at the al., 2011; Li et al., 2015b). Overall, there are an insufficient top of permafrost (TTOP) (Smith and Riseborough, 1996; number of field investigations to use for modelling and vali- Wu et al., 2002a). These models link permafrost tempera- dating the accuracy of the maps. ture with surface temperature using seasonal surface transfer Field survey data sets have recently been obtained based functions and subsurface thermal properties, which can pro- on the project “Investigation of Permafrost and Its Environ- vide reasonable assessments of permafrost distribution when ment over The Qinghai–Xizang (Tibet) Plateau” conducted the permafrost upper boundary conditions and regional soil by the Cryosphere Research Station on Qinghai–Xizang thermal properties are satisfied. Recently, a global permafrost Plateau, Chinese Academy of Sciences (CAS). These data zonation index (PZI) was established based on the relation- could provide validation of permafrost distribution maps. In ships between the air temperature and the occurrence of per- addition, new information on remote sensing LST applica- mafrost (Gruber, 2012). The PZI can represent broad spatial tions and spatial soil characteristics on the TP have been patterns but does not provide the actual extent of the per- studied. An empirical model of daily mean LST was estab- mafrost. lished and performed well in continuous permafrost regions Most temperature fields that have previously been used of the central TP (Zou et al., 2014). Li et al. (2015b) stud- in these models were generated from spatially interpolated ied the relationships between environmental factors and soil air temperature (Pang et al., 2011) or coarser-resolution (e.g. types in the permafrost region of the TP, and they used a 0.125 0.125 ) atmospheric reanalysis data (Gruber, 2012; decision-making tree to spatialize the soil types. The results Qin et al., 2015). Although air temperature produces inac- exhibited good reliability and could be used to realize the curate and low-resolution estimates of ground surface tem- spatialization of soil thermal properties. perature (GST, defined as the surface or near-surface temper- This study aims to generate a new permafrost distribution ature of the ground (bedrock or surficial deposit) and mea- map on the TP using remote sensing LST products and in- sured in the uppermost centimetres of the ground), it was vestigated soil properties. A multiple linear regression model still widely used in practical applications because of limited based on MODIS LST was established, and ground-based GST observations. In these studies, the N factor has been the LST observations were employed to calibrate the results. Soil most effective way to transform the air temperature to the thermal conductivities of each soil type on the TP were calcu- GST (Lunardini, 1978; Klene et al., 2001). With the recent lated via in situ observed soil moisture content and bulk den- development of infrared remote sensing technology, an in- sity. The TTOP model was used to simulate the permafrost creasing number of land surface temperature (LST, defined distribution, and the results were validated by the observed as the average temperature of an element of the exact surface permafrost distributions of boreholes, five investigated re- of the Earth (e.g. surface of ground, vegetation canopy, or gions (IRs, located in the transition zones of the permafrost The Cryosphere, 11, 2527–2542, 2017 www.the-cryosphere.net/11/2527/2017/ D. Zou et al.: A new map of permafrost distribution on the Tibetan Plateau 2529 Table 1. Field survey sample statistics in five investigated regions. so the weighted mean moisture content by depth was used to denote the mean state of each pit. The soil moisture content Investigated region (IR) and bulk density were used to calculate the soil thermal con- ductivity. There was a total of 125 boreholes and 199 soil pits WQ B–Q AEJ GZ XKL Total in five IRs of the field survey (Table 1). Boreholes 21 40 13 23 28 125 Soil pits 74 55 – 19 51 199 2.1.2 Permafrost maps of five investigated regions The permafrost maps of the five IRs were used as the vali- dation data in this study. In a local region, the elevation and and seasonally frozen ground), and three transects (across the terrain factors have greater influence on permafrost occur- entire permafrost regions from north to south). The TTOP rence than longitude and latitude, especially on mountainous modelling result was also compared with two recent bench- permafrost (Riseborough et al., 2008). The lower limit of per- mark maps (made in 1996 and 2006). mafrost (LLP) was determined based on the linear regression relationship between MAGT and borehole elevation. The el- 2 Materials and methods evation where MAGT equals 0 C was regarded as the LLP (Li et al., 2012; Zhang et al., 2012; Chen et al., 2016). In view 2.1 Field survey data sets of the influence of aspects on LLPs, the boreholes were clas- sified into three types: north-facing, south-facing, and east– The investigation of permafrost and its environments on west facing and the LLP of each type was determined. The the TP was conducted from 2009 to 2014. We studied permafrost distribution was generated based on the LLPs of five IRs – WenQuan (WQ) (Zhang et al., 2011, 2012) and different aspects and the digital elevation model (DEM) data, Budongquan–Qingshuihe (B–Q) in the eastern TP, AErJin and a portion of the observed results of boreholes and geo- (AEJ) in the north-eastern TP, GaiZe (GZ) (Chen et al., 2016) physical methods (GPR and TEM) was reserved to validate in the southern TP, and XiKunLun (XKL) (Li et al., 2012) the maps (Li et al., 2012; Zhang et al., 2012). For example, in the western TP (Fig. 1), which are located in the transi- the results of GZ IR showed that the LLP was about 4950 m tion zones between permafrost and seasonally frozen ground for north-facing, 5000 m for east–west-facing, and 5100 m with different climatic and geographic conditions. Ground- for south-facing slopes (Chen et al., 2016). based observations, mechanical excavation, geophysical ex- ploration (ground penetrating radar, GPR; time-domain elec- 2.1.3 Permafrost distribution of three highway tromagnetic, TEM), and borehole drilling were used, and transects comprehensive surveys of the permafrost distribution bound- ary, soil, vegetation, climate, and landform were carried out Three highway transects were established as follows (Fig. 1): in all five IRs. The data sets of ground temperature profiles, National Highway 214 (Qinghai–Yunnan Highway, hereafter spatial distribution of vegetation (Wang et al., 2016), and soil G214) from the northern Ela Mountain to Qingshuihe town, types (Li et al., 2014, 2015a) were obtained and a long-term National Highway 109 (Qinghai–Xizang Highway, hereafter permafrost monitoring network was established, including G109) from Xidatan to Nagqu, and National Highway 219 automatic weather station and borehole records. (Xinjiang–Xizang Highway, hereafter G219) from Kudi to Shiquanhe town. The overall transect lengths of G214, G109, 2.1.1 Boreholes and soil pits and G219 were approximately 400, 750, and 900 km, re- spectively. Three transects across the entire permafrost re- Field survey data sets including ground temperature, soil gions from north to south in the eastern, central, and western moisture content, and bulk density were obtained. The TP were established. Many permafrost geological conditions ground temperature, measured by temperature probes at dif- were discovered in the process of the construction and reno- ferent depths (generally set at 0.5 m intervals from 0 to 5 m, vation of the three highways, and many permafrost roadbed 1 m from 5 to 20 m, 2 m from 20 to 40 m, 5 m from 40 to monitoring sections were subsequently set along the high- 60 m, and 10 m greater than 60 m) in boreholes were used to ways (Jin et al., 2008; Sheng et al., 2015). Based on these determine the existence of permafrost. The soil samples were background data and our investigation results (geophysical collected according to depth increments at each pit. The field and drilling exploration), the permafrost distribution limits bulk density (weight of the soil per unit volume) was mea- and geothermal features of the three transects were generated sured by the clod method. Samples for moisture determina- and used as the validation data sets. tion were stored in aluminum sampling boxes and carefully sealed to prevent moisture changes. The soil moisture con- tent was expressed by weight as the ratio of the mass of water present to the dry weight of the soil sample (Wu et al., 2016). The sampling period was concentrated from July to October, www.the-cryosphere.net/11/2527/2017/ The Cryosphere, 11, 2527–2542, 2017 2530 D. Zou et al.: A new map of permafrost distribution on the Tibetan Plateau Figure 1. Spatial distribution of the field survey regions of the Tibetan Plateau (based on the permafrost distribution map made in 1996). 2.2 Spatial data sets 2.2.2 MODIS LST products 2.2.1 Existing two benchmark permafrost maps The MODIS LST data used in this study were the 1 km grid- ded clear-sky MOD11A2 (Terra MODIS) and MYD11A2 One of the most widely used permafrost distribution bench- (Aqua MODIS) products (reprocessing version 5), which mark maps is the Map of Permafrost on the Qinghai–Tibetan span from 2003 to 2012. The results of the radiance-based Plateau, compiled by the Lanzhou Institute of Glaciology and temperature-based validation indicated that the accuracy and Geocryology, Chinese Academy of Sciences (hereafter of the global MODIS LST product is better than 1 C in most TP-1996) to support basic research on cryospheric dynam- cases, including lakes, homogeneous vegetation, and soils ics in China (Li and Cheng, 1996). TP-1996 synthesizes under clear-sky conditions (Wan et al., 2002, 2004; Coll et field data, literature, aerial photographs, satellite images, and al., 2005; Wan and Li, 2008; Wan, 2008). Langer et al. (2010) many relevant maps and shows that the area of permafrost is and Westermann et al. (2011) focused on weekly averages 6 2 1.41 10 km . The other is the Map of the Glaciers, Frozen and demonstrated an agreement generally better than 2 C Ground and Deserts in China, compiled by Cold and Arid for MODIS LST in the summer season, at permafrost sites in Regions Environmental and Engineering Research Institute, Siberia and Svalbard, Norway, respectively. Chinese Academy of Sciences (hereafter TP-2006) (Wang Both MOD11A2 and MYD11A2 provide two observa- et al., 2006). In this map, the permafrost distribution was tions (daytime and night-time), which means that there are generated using a 0.5 C MAGT isotherm as a threshold, four LST observations per day for the same pixel. The tem- 6 2 which shows that the area of permafrost is 1.12 10 km . poral resolution of MOD11A2/MYD11A2 was 8 days. The The MAGT was interpolated based on the relationship be- LST values represent the 8-day average LST values (the tween elevation/latitude and the MAGT observation from all missing values were ignored in the calculation) (Wan, 2009; 76 boreholes along the Qinghai–Xizang Highway (Nan et al., Wan and Dozier, 1996), and there are theoretically 46 groups 2002). of LST values per year. While the 8-day MODIS LST prod- ucts have more reliable data than the daily products, they still have numerous missing values when establishing the The Cryosphere, 11, 2527–2542, 2017 www.the-cryosphere.net/11/2527/2017/ D. Zou et al.: A new map of permafrost distribution on the Tibetan Plateau 2531 mean daily LST empirical models due to clouds or other fac- Model (SoLIM) in conjunction with soil type and environ- tors (Prince et al., 1998). In this study, the Harmonic Analy- ment factor data (Li et al., 2014, 2015b). According to the sis Time-Series (HANTS) algorithm, developed to deal with USDA soil taxonomy system, there are five soil orders on time series of irregularly spaced observations and to identify the TP as follows: Gelisols, Aridisols, Mollisols, Inceptisols, and fill in the missing values (Roerink et al., 2000), was ap- and Entisols. Considering the availability of soil sample pa- plied to smoothen and reconstructing MODIS LST series on rameters, the characteristics of sampling regions, and model a per-pixel basis for the entire study area. The parameters set applicability, the empirical model of soil thermal conductiv- for the HANTS analysis were described in detail by Xu et ity proposed by Kersten (1949) was adopted in this study. al. (2013). The equation of thawed soil thermal conductivity is The full coverage of the entire TP requires a total of 13 .0:6243 / k D 0:1442.0:7 log!C 0:4/ 10 : (2) swathes (h23v04, h23v05, h24v04, h24v05, h24v06, h25v04, h25v05, h25v06, h26v04, h26v05, h26v06, h27v05, and Furthermore, the equation of frozen soil thermal conductivity h27v06) of the MOD11A2/MYD11A2 products. The Terra is overpass time is around 10:30(local time) in its descend- .0:8116 / ing mode and 22:30 in its ascending mode. The Aqua over- d k D 0:01096 10 pass time is around 13:30 in its ascending mode and 01:30 .0:9115 / C 0:00461 10 !; (3) in its descending mode (https://modis.gsfc.nasa.gov/). The MODIS LSTs represent instantaneous observation values, 1 1 where k =k is the thermal conductivity (W m K / of t f and the overpass times of the satellites do not accurately cor- thawed/frozen soil, ! is the soil moisture content (%), and respond to standard meteorological observation times (Bei- is the soil bulk density (kg m /. Both ! and were d d jing time: 02:00, 08:00, 14:00, and 20:00) (China Meteoro- measured via soil samples collected in the field survey. The logical Administration, 2003). Therefore, an arithmetic mean soil samples were classified according to soil orders; mois- of the four LST observations with the same weights will pro- ture content and bulk density values were averaged within duce a large deviation from the mean daily LST (Wang et al., soil orders to eliminate abnormal values (Table 2, the val- 2011). We used a multiple linear regression to distribute dif- ues show the mean with standard deviation of soil thermal ferent weights to each MODIS LST observation to establish parameters of each type). the mean daily LST empirical model. The details of process- ing were described by Zou et al. (2014). The model valida- 2.2.4 Glacier and lake data tion at three permafrost sites showed that the determination, mean error, mean absolute error, and root mean squared er- The spatial distribution and data on areas of glaciers and ror of mean daily LST were 0.91 to 0.93, 0.21 to 1, 2.28 lakes on the TP were from the Second Glacier Inven- to 2.42, and 2.96 to 3.05 C, respectively. In this study, the tory Dataset of China (Guo et al., 2014) and the Chinese empirical formula is as follows: Cryosphere Information System (Li, 1998) provided by the Cold and Arid Regions Science Data Center (http://westdc. LST D 0:18 Terra C 0:269 Terra daily day night westgis.ac.cn). C 0:143 Aqua C 0:435 Aqua C 0:896; (1) day night 2.3 TTOP model where LST is the mean daily LST, Terra is the day- daily day time LST observation of MOD11A2, Terra is the night- night Considering the model’s usefulness and sophistication, spa- time LST observation of MOD11A2, Aqua is the daytime day tial scales, and available data sets (Riseborough et al., 2008), LST observation of MYD11A2, and Aqua is the night- night we selected the temperature at the top of permafrost (TTOP) time LST observation of MYD11A2. model (Smith and Riseborough, 1996) to simulate the per- The calculations of the thawing indices (thawing degree mafrost distribution on the TP. days, TDD) and freezing indices (freezing degree days, The TTOP model can be expressed as follows: FDD) were based on the 8-day average LST calculated from the previous processing. The procedures were realized us- k =k TDD FDD t f TTOPD ing the IDL programming language, and the FDD and TDD from 2003 to 2012 were obtained and averaged as the model .r n I /.n I / k t t f f D ; (4) inputs. where P is the annual period (365 days). TDD (n I / is 2.2.3 Soil thermal properties t t the ground surface thawing index, and FDD (n I / is the f f Soil thermal characteristics were modelled according to pa- ground surface freezing index. n and n are n factors of the t f rameters measured from soil types encountered in the field. thawing and freezing seasons, and I and I are the air tem- t f The classification of soil types was performed using the De- perature thawing and freezing indices. r D k =k is defined k t f cision Tree See 5.0 software and the Soil Land Inference as the ratio of the thermal conductivity coefficient when soil www.the-cryosphere.net/11/2527/2017/ The Cryosphere, 11, 2527–2542, 2017 2532 D. Zou et al.: A new map of permafrost distribution on the Tibetan Plateau Figure 2. Flow diagram of the modelling scheme. Table 2. Soil thermal parameters of each type on the Tibetan Plateau. Soil order Sample Moisture Bulk density Thawed soil Frozen soil number content ( %) (kg m / thermal conductivity thermal conductivity 1 1 1 1 (W m K / (W m K / Aridisols 43 7.76 (3.0) 1601.9 (173.2) 1.47 (0.42) 1.25 (0.63) Entisols 10 8.79 (6.64) 1447.7 (164.8) 1.23 (0.17) 1.01 (0.33) Gelisols 56 22.24 (13.79) 1277.6 (310.0) 1.22 (0.36) 1.62 (0.44) Inceptisols 94 16.22 (7.37) 1313.4 (221.7) 1.18 (0.34) 1.30 (0.53) Mollisols 14 20.00 (5.66) 1186.9 (141.3) 1.05 (0.23) 1.22 (0.48) is thawing and freezing. We used the modified MODIS LST approximately 03:00) was employed as input data for the de- data processed in Sect. 2.2.2 as the GST to derive the TDD termination of unfrozen ground area. The uncertainty anal- and FDD, and the r was calculated from the soil properties ysis of total permafrost area was conducted with R statis- derived from processes in Sect. 2.2.3. tical software (version 3.3.1, www.r-project.org) using the From Eq. (4), if the FDD is greater than percentile method, and we used a 90 % confidence interval TDDk /k .n I >r n I /, then TTOP will be to determine the range of the total permafrost area. The mod- t f f f k t t below 0 C and permafrost exists. This processing was real- elling scheme in this study is shown in Fig. 2. ized in the ArcGIS software programme with the following expression: 2.4 Accuracy evaluation 1; TTOP 0 permafrost DD (5) The permafrost distribution of the borehole locations, five 0; TTOP > 0 seasonally frozen ground IRs, and three transects were used to estimate the accura- The glacier and lake regions were excluded from the per- cies of the three maps (TP-1996, TP-2006, and TP-2016). mafrost distribution modelling of the TTOP model. In ad- The spatial distribution of borehole temperature data across dition to permafrost and seasonally frozen ground, unfrozen a permafrost domain or seasonally frozen ground area was ground was also identified in this study. The unfrozen ground used as the criterion of advantages and disadvantages of re- was defined as the region where the extreme minimum sults for the three time snapshots of 1996, 2006, and 2016. LST 0 C. The night Aqua MODIS LST (observation time The permafrost distribution across the five IRs and three tran- The Cryosphere, 11, 2527–2542, 2017 www.the-cryosphere.net/11/2527/2017/ D. Zou et al.: A new map of permafrost distribution on the Tibetan Plateau 2533 sects were selected as the real values with which to validate at permafrost or seasonally frozen ground in five IRs of three the three maps. maps. Different combinations were set up to analyse the dif- To evaluate the agreement of the simulated permafrost dis- ferences between the three results. Columns a, b, and c show tribution and the observed results, the kappa coefficient (K ) the results of TP-1996, TP-2006, and TP-2016, and rows 1, 2, (Cohen, 1960), which measures the degree of agreement, was 3, 4, and 5 show the results of XKL, GZ, AEJ, B–Q, and WQ selected for accuracy evaluation. IRs, respectively. The results show that TP-1996 is insensi- tive to the geographical boundaries across all five IRs, and s=n.a b Ca b /=n 1 1 0 0 there are many erroneous interpretations of both permafrost K D ; (6) and seasonally frozen ground. TP-2006 had higher sensitivity 1.a b Ca b /=n 1 1 0 0 to the boundaries, especially in WQ IR; however the recogni- where the total number of pixels is n, and s is the number tion of the other four IRs is inadequate and the areas of per- of pixels in which the simulation and investigated results mafrost distribution were overestimated. Compared to TP- agree. The number of investigated result pixels with per- 1996 and TP-2006, TP-2016 performed better at identifying mafrost is a , and those without are a , and the simulated 1 0 the geographic boundary of permafrost distribution, identi- map pixel numbers are b and b . Empirically and statisti- 1 0 fying almost all the boundaries of the five IRs correctly, es- cally arbitrary quality values for K have been proposed. Co- pecially for the seasonally frozen ground in the valley of the hen (1960) suggested that K 0:8 signifies excellent agree- north-western XKL IR (Fig. 4c1) and that around the lakes of ment, 0:6 K< 0:8 represents substantial agreement, 0:4 the eastern AEJ IR (Fig. 4c3). TP-2016 had some errors that K< 0:6 represents moderate agreement, 0:2 K< 0:4 repre- were mainly affected by local terrain factors. These included sents fair agreement, and a lack of agreement corresponds to the seasonally frozen ground distributed in valleys and a few K< 0:2. permafrost spots at the margin, such as the two seasonally frozen ground boreholes in the northern AEJ IR (Fig. 4c3) and three permafrost boreholes at the south-western limit of 3 Results GZ IR (Fig. 4c2). 3.1 Permafrost distribution modelling of TTOP 3.3 Validation with five investigated regions (IRs) Figure 3 shows the simulated permafrost distribution of The permafrost distributions of the five IRs were employed the TTOP model on the TP (TP-2016). The distribution as true values to validate the modelling results of the three areas of permafrost and seasonally frozen ground were 6 2 maps in order to analyse their performance in geographi- 1.06 10 km with a 90 % confidence interval of 0.97– 6 2 6 2 cal boundary recognition ability. TP-1996 was the worst at 1.15 10 km , and 1.46 10 km . This estimate excluded recognizing the boundaries of permafrost in the five IRs. It glaciers and lakes, which account for 40 and 56 % of the to- misidentified all boundaries, with a low kappa coefficient tal TP area, respectively. The result shows that the permafrost (K< 0:2), due to greater misjudgment or overestimation of distribution was centred in southern Qinghai and northern Ti- permafrost pixels. TP-2006 also performed poorly in the bet. The northern Qiangtang Plateau and the Kunlun Moun- XKL, GZ, and AEJ IRs (K< 0:2) but performed better in the tains were the regions with the most permafrost, which ex- B–Q and WQ IRs, with a kappa coefficient reaching 0.63 tends west and north-west to the Karakoram mountains. The and 0.77. TP-2016 had poor performance in the AEJ IR. permafrost continuity decreases gradually as the elevation The kappa coefficient was only 0.38, which is a slight im- decreases and the ground temperature increases with increas- provement over estimates of the former two. In addition, it ing distance from the central region. The geographic north- represents moderate agreement with the XKL and GZ IRs ern and southern boundaries of permafrost were Xidatan and and substantial agreement with the B–Q and WQ IRs, which Anduo from the mark sites of the Qinghai–Xizang Highway. have kappa coefficients of 0.54, 0.48, 0.68, and 0.78, respec- There were a few areas of permafrost in the high mountains tively. The average accuracies of TP-1996, TP-2006, and TP- from Anduo to the southern Tibet Valley. Due to the exis- 2016 were 0.06, 0.35, and 0.57. TP-2016 performed best in tence of the Bayan Har Mountains and Anemaqen Mountain, the validation with the investigated permafrost distribution the elevations of which are above 5000 m, there is permafrost from both the individual and mean accuracies of the five IRs occurrence in the eastern TP. Some unfrozen ground exists in (Table 3). The TP-2016 performed better at identifying the the south-eastern margin of the TP, and the size of this area 6 2 permafrost boundary in the regions with complex terrain be- is approximately 0.03 10 km (account for 1 % of the total cause of sharp changes in the LST within short distances, TP area). such as the WQ, B–Q, and XKL IRs. For GZ and AEJ IRs, where surface relief is much lower, the TP-2016 does not per- 3.2 Validation with borehole observations form as well as the other three IRs. The worst performance Boreholes can determine whether permafrost exists or not. in AEJ IR might also be due to no soil pits in the investiga- Figure 4 shows the spatial distribution of borehole locations tion and the soil thermal properties are inferred completely www.the-cryosphere.net/11/2527/2017/ The Cryosphere, 11, 2527–2542, 2017 2534 D. Zou et al.: A new map of permafrost distribution on the Tibetan Plateau Figure 3. Spatial distribution of permafrost with the derived TTOP on the Tibetan Plateau. Table 3. Kappa coefficient statistics in five investigated regions of The observed MAGT of the borehole closest to Ayakekumu three maps. Lake was 3 C, which indicates the existence of seasonally frozen ground there. TP-2016 accurately modelled this phe- Investigated TP-1996 TP-2006 TP-2016 nomenon. In the regions around the AEJ IR, TP-2016 sim- region ulated the seasonally frozen ground around Aqikekule Lake (area approximately 350 km / and its source river, and this WQ 0 0.77 0.78 was not found in the other two maps. Most lakes on the TP B–Q 0 0.63 0.68 are formed due to tectogenesis. The major axis basically re- AEJ 0 0 0.38 GZ 0.15 0.19 0.48 mains consistent with the main structure directions and the XKL 0.14 0.17 0.54 secondary level fracture direction, and there generally exists penetrative or non-penetrative taliks under and around tec- Average 0.06 0.35 0.57 tonic lakes (Zhou et al., 2000). TP-2016 also shows season- ally frozen ground in the mountainous region in proximity to the Pitileke River, while the other two maps did not identify from the relationship between the environmental factors and this area. TP-2016 performed better at identifying the sea- the soil samples of the other four IRs. sonally frozen ground formed by the surface water. The results of the AEJ IR and surrounding area were The permafrost distribution of TP-1996 and TP-2006 was selected to compare the differences among the three maps modelled according to the relationship between temperature (Fig. 5). In the AEJ IR, the investigated result shows that and 3-dimensional zonalities (longitude, latitude, and eleva- the seasonally frozen ground is mainly distributed at the tion) (Cheng, 1984). The higher weight given to elevation northern valley and the eastern Ayakekumu Lake surround- from the regression equation determined that it has a greater ing areas with permafrost. TP-2006 shows all judgements influence than that of longitude and latitude when interpolat- for permafrost in the AEJ IR, and the permafrost area ing temperature (air temperature or MAGT). The high con- is clearly overestimated. TP-1996 shows some seasonally tinuity and low variability of the elevation difference in per- frozen ground in the north-western AEJ IR, but the locations mafrost regions lead to results that appear more continuous. were misjudged. TP-2016 judged approximately 30 % sea- However, the temperature differences caused by local factors sonally frozen ground in the northern and eastern AEJ IR. (e.g. lakes or rivers) are largely masked and this results in an Although there were few correct pixels, the locations in the excessive occurrence of the lower extrapolated temperature. eastern part were at the geographic boundary of permafrost. The Cryosphere, 11, 2527–2542, 2017 www.the-cryosphere.net/11/2527/2017/ D. Zou et al.: A new map of permafrost distribution on the Tibetan Plateau 2535 Figure 4. Spatial distribution of boreholes in five IRs of three maps. This may explain the overestimated area of permafrost dis- Table 4. Kappa coefficient statistics in three transects of three maps. tribution in the previous TP-1996 and TP-2006. The use of remote sensing data can better reveal the spatial heterogene- Transect TP-1996 TP-2006 TP-2016 ity of LST. Relative to the two benchmark maps, the TP-2016 G214 0.32 0.41 0.62 result driven by the processed MODIS LST in this paper was G109 0.21 0.59 0.69 very sensitive to seasonally frozen ground formed by surface G219 0.47 0.49 0.74 water, and the results show that there are many seasonally Average 0.34 0.50 0.68 frozen ground areas surrounding lakes and major rivers that correspond to the previous studies (Lin et al., 2011; Niu et al., 2011). the highest accuracy. The kappa coefficients were 0.62, 0.69, 3.4 Validation with three transects and 0.74 for G214, G109, and G219, respectively, with an average of 0.68. TP-2016 performed best in the validation The permafrost distribution of the three transects (G214, with the investigated permafrost distribution from both the G109, and G219) of three maps were extracted for compar- individual and averaged accuracies of the three transects. In ison with the investigated results to comprehensively eval- the three transects across all permafrost regions from north to uate their performance on the mainly permafrost developed south in the eastern, central, and western TP, which include regions of the TP. The accuracy statistics of the three maps most permafrost distribution characteristics in TP, the vali- for the three transects are listed in Table 4. TP-1996 had the dation results should be a synthetic evaluation of the three worst accuracy of the three maps with an average kappa co- maps. efficient of 0.34. The accuracy of TP-2006 was higher than Figure 6 shows the distributions of permafrost and sea- that of TP-1996 with an average kappa coefficient of 0.50. sonally frozen ground along the G109 transect of the three It performed well, especially for transect G109. TP-2016 has maps and the investigated result. The elevation and mark sites www.the-cryosphere.net/11/2527/2017/ The Cryosphere, 11, 2527–2542, 2017 2536 D. Zou et al.: A new map of permafrost distribution on the Tibetan Plateau Figure 5. Comparison of the three maps in and around the AErJin investigated region (a TP-1996, b TP-2006, c TP-2016). Figure 6. Comparison of permafrost distribution of three maps along the G109 transect with investigated results (P: permafrost, SFG: sea- sonally frozen ground; XDT: Xidatan, KLSYK: Kunlun Mountain Peak, WDL: Wudaoliang, BLH: Beilu River, FHSYK: Fenghuo Mountain Peak, WL: Wuli, TTH: Tuotuo River, KXL: Kaixin Mountain Ridge, TTH’: Tongtian River, YSP: Yanshiping town, TGL: Tangula Mountain Peak, AD: Anduo town, LDH: Liangdaohe, NQ: Nagqu town, G109-IR: investigated results of permafrost distribution in the G109 transect). were also added for analysis. For convenient comparison, the to northern AD), one region of seasonally frozen ground only G109 transect was divided into five segments according to (from southern LDH to NQ) and two regions in which per- the investigated result as follows: two continuous permafrost mafrost and seasonally frozen ground coexist (from WL to regions (from XDT to southern FHSYK, and southern YSP YSP, and AD to LDH). The comparison shows that the three The Cryosphere, 11, 2527–2542, 2017 www.the-cryosphere.net/11/2527/2017/ D. Zou et al.: A new map of permafrost distribution on the Tibetan Plateau 2537 Table 5. Kappa coefficient statistics for three maps. ing geologic surveys of the Qinghai–Xizang Highway and Railway (Jin et al., 2008). From the kappa coefficients of the TP-1996 TP-2006 TP-2016 three maps and investigated result (Table 4) along the G109 transect, TP-2016 can better identify the seasonally frozen TP-1996 1 0.56 0.53 ground that is several kilometres wide and caused by local TP-2006 – 1 0.71 factors (surface water, geothermal, and permeation/radiation TP-2016 – – 1 effects). 3.5 Spatial difference among the three maps maps performed well in two continuous permafrost regions. The kappa coefficients of each pair among the three maps Almost all permafrost is identified correctly except for sev- were calculated (Table 5) to analyse the spatial difference. eral seasonally frozen ground areas in the CMEH and BLH TP-1996 had low consistency with both TP-2006 and TP- of TP-1996. In the region of seasonally frozen ground only, 2016. The kappa coefficients were 0.56 and 0.53, respec- TP-1996 judged permafrost from AD to NQ, which is differ- tively, which indicates a large difference. TP-2006 had sub- ent from the investigated result and overestimated the per- stantial agreement with TP-2016 and the kappa coefficient mafrost area in this region. TP-2006 and TP-2016 identi- reached 0.71. The spatial differences between each pair fied that only seasonally frozen ground exists in this region, among the three maps were compared (Fig. 7). Compared which is consistent with the investigated result. In two re- with TP-2006 and TP-2016, TP-1996 overestimated the per- gions where permafrost and seasonally frozen ground coex- mafrost area, which was mainly distributed in the south- ist, a large difference was seen between the three maps and eastern TP, south margin of continuous permafrost, and pre- the investigated result. TP-2006 shows that continuous per- dominantly continuous and island permafrost in the South- mafrost exists from XDT to northern AD, performed poorly ern TP. In addition, TP-1996 misidentified some seasonally in the recognition of the seasonally frozen ground, and over- frozen ground on the continuous permafrost edge and the in- estimated the area of permafrost in the G109 transect. TP- terior TP. The permafrost distribution area of TP-2006 was 1996 performed better than TP-2006 and recognized some of similar to that of TP-2016. Differences mainly existed in the seasonally frozen ground in TTH, TTH’, YSP, and AD. the interior TP regions, southern margin of continuous per- However, the widths and locations reveal bias from that of the mafrost, and the regions surrounding the Bayan Har Moun- investigated result. TP-2016 identified almost all locations of tains and eastern Nyainqêntanglha Mountains. seasonally frozen ground correctly with a smaller width dif- ference, and was more consistent with the investigated re- sult than the former two. Both TP-2006 and TP-2016 iden- tified the sporadic permafrost in LDH, which was generally 4 Discussion expected to be the southern limit of permafrost in previous studies. TTOP was formulated with the modified MODIS LST, rather In the G109 transect, seasonally frozen ground mainly than ground surface temperature (GST) in this study. It is exists due to the surface water effects, regional geologic well known that MODIS LST observes a mixture of the vege- structure/geothermal effects, and penetration/radiation ef- tation canopy, snow cover, and ground surface, depending on fects, which cause a discontinuity in the plane and depth of the region and seasons. The snow cover and vegetation might the continuous distribution of permafrost (Zhou et al., 2000). have a significant influence on the relationship between the Due to the large streamflow and high water temperature of GST and MODIS LST, depending on the snow depth and TTH, TTH’ and Buqu (flow through YSP) rivers, the pen- duration (Zhang, 2005), and vegetation height and coverage. etrative taliks not only developed on the riverbed and high The snow cover distribution is spatially variable over the TP floodplain, but also expanded to the first or second terrace (Fig. 8a), with the most persistently snow-covered areas oc- (the width generally reached 5–10 km). Additionally, a bare curring in the south-eastern and western edge of the TP and ground, gravel layer exists, and a higher mean annual air in some alpine regions with elevations higher than 6000 m temperature was beneficial to precipitation infiltration, which (Qin et al., 2006; Pu et al., 2007). Overall, the snow cover created active thermal transfer conditions. Therefore, the sea- is rare, thin (< 3 cm) and has a short duration (mostly ex- sonally frozen grounds existing in TTH and YSP were also isting less than 1 day for a single snow event) due to the affected by penetration/radiation effects. However, for the strong solar radiation and wind in the vast interior and the rivers with less streamflow and at higher latitudes, such as the northern TP (Che et al., 2008; Huang et al., 2017), where CMEH and BHL rivers, the non-penetrative taliks are much the permafrost is most developed. Therefore, although thin smaller (generally < 100 m) and thus almost impossible to snow cover might have a cooling effect on GST due to the identify. The seasonally frozen ground in northern WL was high albedo of fresh snow and a rapid process of snowmelt mainly affected by regional geologic structure/geothermal (Zhang, 2005), the cooling effect may be of short duration effects, which has been validated by the results of engineer- and have very little effect on our study. The vegetation types www.the-cryosphere.net/11/2527/2017/ The Cryosphere, 11, 2527–2542, 2017 2538 D. Zou et al.: A new map of permafrost distribution on the Tibetan Plateau Figure 7. Spatial difference among the three maps (96: TP-1996, 06: TP-2006, 16: TP-2016; SFG: seasonally frozen ground, P: permafrost). Figure 8. Annual average snow depth (a edited after Che et al., 2008) and vegetation types of the permafrost region (b edited after Wang et al., 2016) on the Tibetan Plateau. in the alpine ecosystem of the permafrost region on the TP air temperature isotherms and modified in several regions (Fig. 8b) are all composed of grassland and characterized (such as Qinghai–Xizang Highway, Qinghai–Yunnan High- by dwarf and sparsely distributed plants (Wang et al., 2016). way, and the Hengduan Mountains) using field data based on The vegetation cover across most of the permafrost region the authors’ knowledge. The threshold was determined by was less than 30 % (Lehnert et al., 2015) and even less than the empirical statistical relationship between permafrost oc- 10 % in the middle and western TP. In view of the conditions currence and meteorological observations in the eastern TP of both snow cover and vegetation on the TP, there are only (Li and Cheng, 1996), while the universality of the thresh- slight differences between the GST and MODIS LST on av- old is questionable in the western TP due to insufficient data. erage, and even smaller differences in FDD and TDD in our In addition, high uncertainty exists in the use of air tempera- study area. In addition, the HANTS algorithm might cause ture interpolation because of the scarce, unevenly distributed some bias under cloudy-sky conditions. Further evaluation of monitoring sites. There were more sites in the eastern TP and the algorithm was not performed in this study, because it has fewer in the western TP, more sites at lower elevations and been proved to be an effective approach for filling in gaps in fewer at higher elevations, and very few sites in permafrost the MODIS LST data for the TP where clear-sky conditions regions. This resulted in the low accuracy of extrapolated air dominated (Xu et al., 2013). temperature on the TP (Lin et al., 2002; Li et al., 2003), The data set used in the earliest maps (compiled in the especially in the permafrost region. The permafrost maps 1980s and 1990s) included air temperature, field data, aerial were compiled with conventional cartographic techniques photographs, satellite images, and many relevant maps (Tong that plotted the permafrost boundaries on the topographic and Li, 1983; Shi and Mi, 1988; Li and Cheng, 1996). The maps by hand (Tong and Li, 1983; Shi and Mi, 1988; Li and permafrost boundary was mainly based on a threshold of Cheng, 1996). The artefactual errors were very difficult to The Cryosphere, 11, 2527–2542, 2017 www.the-cryosphere.net/11/2527/2017/ D. Zou et al.: A new map of permafrost distribution on the Tibetan Plateau 2539 control and depended on the knowledge and skill of the map- equilibrium of permafrost under ongoing climate warming, per. These factors led to significant uncertainties in the maps. and thereby any map based on a contemporary climate forc- These maps place much emphasis on the broad concept of ing is likely to underestimate the extent of permafrost. How- “possible” permafrost regions and this overestimated the ac- ever, permafrost bodies have a long response time to atmo- tual permafrost areas (Wang et al., 2016). The permafrost spheric conditions (Riseborough, 2007; Romanovsky et al., mapping of TP-2006 was based on the MAGT that consid- 2010; Smith et al., 2010). The increasing rates of ground ered the characteristics of high-altitude permafrost. The re- temperature were much lower in the TP than that in the cir- gional MAGT was interpolated based on the relationship be- cumpolar regions and much lower for the warm permafrost tween elevation/latitude and the borehole observations along (Wu and Zhang, 2008; Smith et al., 2005; Zhao et al., 2010), the Qinghai–Xizang Highway (Nan et al., 2002). The MAGT which is mostly distributed near the permafrost boundaries. model performed more accurately in the central TP than Moreover, the degradation of permafrost in these regions was in the eastern and western TP, which was demonstrated in characterized by a deepening of the active layer, rather than the validation of the three transects. In view of the medium the disappearance of permafrost. The changes of the per- spatio-temporal resolution and sensitivity to spatial tempera- mafrost distribution on the TP might be very limited in the ture heterogeneity of the MODIS LST data used in the map- past several decades. Therefore, the spatial difference among ping of TP-2016, it can accurately represent the spatial pat- the three maps might be mainly induced by the differences tern of LST on the TP. In addition, the MODIS LST data in methods and data sources. The TP-2016 could be used as were calibrated using ground-based LST observations ob- the benchmark map for permafrost distribution on the TP, al- tained from automatic weather stations in typical permafrost though more work is needed to improve the accuracy of sur- regions (Zou et al., 2014), which correspond to actual climate face forcing and the soil parameters. In addition, although the conditions of the permafrost region. Moreover, the subsur- approach based on the relationship between current climate face thermal properties derived from soil investigation data and permafrost occurrence is useful for mapping the distribu- were also considered in the TTOP model. The improvement tion of the TP permafrost, it should be cautious when apply- in upper boundary conditions of the permafrost model and ing the transient responses of permafrost to climate change the use of large quantities of reliable in situ observed data to modelling. sets led to higher modelling accuracy. In the earliest maps, only observational data from the field 5 Conclusions sites along Qinghai–Xizang Highway were used for map evaluation (Tong and Li, 1983; Shi and Mi, 1988; Li and This study exploited the advantages of the medium spatio- Cheng, 1996). For TP-2006, the threshold of 0.5 C MAGT temporal resolution of MODIS LST products to construct was determined by a sensitivity analysis of comparison with a database of mean daily LST of the TP. The permafrost the TP-1996, without independent validation (Nan et al., distribution is simulated by the TTOP model combined 2002). The validation in this study showed that the TP-2006 with ground observations and soil investigated data sets. accuracy was higher than that of TP-1996. However, TP- The model was validated against the permafrost distribu- 2006 highlights the excessive elevation effects in the MAGT tion obtained from the borehole temperature data, five IRs interpolation and masks the effects of local factors to some and three transects and compared to two recent benchmark degree. The better performance of TP-2006 in the B–Q and maps. From the validation with borehole temperature data, WQ IRs might be explained by geomorphology similar to the suggested method of permafrost boundary identifica- the Qinghai–Xizang Highway, because these two IRs were tion shows a better result than the two maps, especially for closer to the highway than the other three IRs. This suggests the seasonally frozen ground in valleys and around lakes. that the MAGT model could reflect the permafrost distribu- The accuracy of the method validation shows that the TP- tion when there are sufficient borehole ground temperature 2016 case has the highest kappa coefficients for the five IRs observations, and this is why we used it to model the per- and three transects. The average coefficients were 0.57 and mafrost distribution of five IRs. The validation results of the 6 2 0.68. The modelling estimation shows that 1.06 10 km five IRs emphasized the performance when recognizing per- 6 6 2 of permafrost (0.97 10 –1.15 10 km , 90 % confidence mafrost boundaries and that of the three transects emphasized 6 2 interval), 1.46 10 km of seasonally frozen ground, and the overall evaluation of the three maps. Overall, the valida- 6 2 0.03 10 km of unfrozen ground could be on the TP. Com- tion results of both the five IRs and three transects suggested pared with two recent benchmark maps, the TTOP model that TP-2016 performed the best and achieved the highest is superior in recognizing the boundary of permafrost, espe- accuracy among the three maps. The results provide a stan- cially in the seasonally frozen ground areas caused by local dard permafrost distribution map on the TP based on current factors. The new permafrost distribution map represents a ba- climate conditions. sic data set for future permafrost research. The ground temperatures of permafrost on the TP have in- creased during the past several decades (Wu and Liu, 2004; Wu and Zhang, 2008; Zhao et al., 2010). This means a dis- www.the-cryosphere.net/11/2527/2017/ The Cryosphere, 11, 2527–2542, 2017 2540 D. Zou et al.: A new map of permafrost distribution on the Tibetan Plateau Data availability. The data of permafrost distribution on the Ti- Cohen, J.: A Coefficient of Agreement for Nominal Scales, Educ. betan Plateau generated in this paper is provided in the supplement. Psychol. Meas., 20, 37–46, 1960. Coll, C., Caselles, V., Galve, J. M., Valor, E., Niclòs, R., Sánchez, J. M., and Rivas, R.: Ground measurements for the validation of land surface temperatures derived from AATSR and MODIS The Supplement related to this article is available online data, Remote Sens. Environ., 97, 288–300, 2005. at https://doi.org/10.5194/tc-11-2527-2017-supplement. Gillespie, A.: Land Surface Temperature, Encyclopedia of Remote Sensing, NY, Springer New York, New York, 314–319, 2014. Gruber, S.: Derivation and analysis of a high-resolution estimate of global permafrost zonation, The Cryosphere, 6, 221–233, https://doi.org/10.5194/tc-6-221-2012, 2012. Competing interests. The authors declare that they have no conflict Guo, W., Liu, S., Yao, X., Xu, J., Shangguan, D., Wu, L., Zhao, J., of interest. Liu, Q., Jiang, Z., Wei, J., Bao, W., Yu, P., Ding, L., Li, G., Li, P., Ge, C., and Wang, Y.: The second glacier inventory dataset of China (version 1.0), Cold and Arid Regions Science Data Center Acknowledgements. We thank the Atmosphere Archive and at Lanzhou, 2014. Distribution System for providing MODIS land products, and Cold Hachem, S., Allard, M., and Duguay, C.: Using the MODIS land and Arid Regions Science Data Center for data on glacier, lake and surface temperature product for mapping permafrost: an appli- snow depth on the Tibetan Plateau. This research was financially cation to northern Québec and Labrador, Canada, Permafrost supported by the National Major Scientific Project of China Periglac., 20, 407–416, 2009. “Cryospheric Change and Impacts Research” (2013CBA01803), Haeberli, W. and Hohmann, R.: Climate, glaciers and permafrost in the Creative Research Groups of National Natural Science Foun- the Swiss Alps 2050: scenarios, consequences and recommen- dation of China (no. 41421061), the key project of the Chinese dations, in: Proceedings Ninth International Conference on Per- Academy of Sciences (KJZD-EW-G03-02). We also acknowledge mafrost, 1, 607–612, 2008. Jerry Brown, Stephan Gruber, and Sergey Marchenko for their Huang, X., Deng, J., Wang, W., Feng, Q., and Liang, T.: Impact constructive suggestions. of climate and elevation on snow cover using integrated remote sensing snow products in Tibetan Plateau, Remote. Sens. Envi- Edited by: Tobias Bolch ron., 190, 274–288, 2017. Reviewed by: two anonymous referees Jin, H., Wei, Z., Wang, S., Yu, Q., Lü, L., Wu, Q., and Ji, Y.: Assess- ment of frozen-ground conditions for engineering geology along the Qinghai-Tibet highway and railway, China, Eng. Geol., 101, 96–109, 2008. References Kääb, A.: Remote sensing of permafrost-related problems and haz- ards, Permafrost Periglac., 19, 107–136, 2008. Brown, J., Ferrians Jr., O. J., Heginbottom, J. A., and Melnikov, Kersten, M. S.: Laboratory research for the determination of the E. S.: Circum-Arctic Map of Permafrost and Ground-Ice Con- thermal properties of soils, ACFEL Technical Report, 23, AD ditions, Circum-Pacific Map Series CP-45, scale 1V 10 000 000, 712516, 1949. 1 sheet, US Geological Survey in Cooperation with the Circum- Klene, A. E., Nelson, F. E., and Shiklomanov, N. I.: The N-factor Pacific Council for Energy and Mineral Resources, Washington, in natural landscapes: variablility of air and soil-surface temper- DC, 1997. atures, Kuparuk river basin, Alaska, USA, Arct. Antarct. Alp. Brown, J., Ferrians Jr., O. J., Heginbottom, J. A., and Melnikov, Res., 33, 140–148, 2001. E. S.: Circum-Arctic Map of Permafrost and Ground Ice Condi- Langer, M., Westermann, S., and Boike, J.: Spatial and temporal tions, Digital media, National Snow and Ice Data Center/World variations of summer surface temperatures of wet polygonal tun- Data Center for Glaciology, Boulder, CO, 1998. dra in Siberia – implications for MODIS LST based permafrost Che, T., Xin, L., Jin, R., Armstrong, R., and Zhang, T.: Snow depth monitoring, Remote Sens. Environ., 114, 2059–2069, 2010. derived from passive microwave remote-sensing data in China, Lehnert, L., Meyer, H., Wang, Y., Miehe, G., Thies, B., Reuden- Ann. Glaciol., 49, 145–154, 2008. bach, C., and Bendix, J.: Retrieval of grassland plant coverage Chen, J., Zhao, L., Sheng, Y., Li, J., Wu, X. D., Du, E. J., Liu, G. Y., on the Tibetan Plateau based on a multi-scale, multi-sensor and and Pang, Q. Q.: Some Characteristics of Permafrost and Its Dis- multi-method approach, Remote Sens. Environ., 164, 197–207, tribution in the Gaize Area on the Qinghai-Tibet Plateau, China, Arct. Antarct. Alp. Res., 48, 395–409, 2016. Li, K., Chen, J., Zhao, L., Zhang, X. M., Pang, Q. Q., Fang, H. Cheng, G. D.: Problems on zonation of high-altitude permafrost, B., and Liu, G. Y.: Permafrost distribution in typical area of Ac. Geogr. Sin., 39, 185–193, 1984. west Kunlun Mountains derived from a comprehensive survey, Cheng, G. D. and Wu, T. H.: Responses of permafrost J. Glaciol. Geocryol., 34, 447–454, 2012 (in Chinese). to climate change and their environmental significance, Li, S. and Cheng, G.: Map of Frozen Ground on Qinghai-Xizang Qinghai-Tibet Plateau, J. Geophys. Res., 112, F02S03, Plateau, Gansu Culture Press, Lanzhou, 1996. https://doi.org/10.1029/2006JF000631, 2007. Li, W., Zhao, L., Wu, X., Wang, S., Nan, Z., Fang, H., and China Meteorological Administration.: The ground meteorological Shi, W.: Distribution of Soils and Landform Relationships in observation specification, China Meteorological Press, Beijing, China, 2003. The Cryosphere, 11, 2527–2542, 2017 www.the-cryosphere.net/11/2527/2017/ D. Zou et al.: A new map of permafrost distribution on the Tibetan Plateau 2541 Permafrost Regions of the Western Qinghai-Xizang (Tibetan) Qin, D., Liu, S., and Li, P.: Snow cover distribution, variability, and Plateau, China, Soil. Sci., 179, 348–357, 2014. response to climate change in western China, J. Clim., 19, 1820– Li, W., Zhao, L., Wu, X., Wang, S., Sheng, Y., Ping, C., Zhao, Y., 1833, 2006. Fang, H., and Shi, W.: Soil distribution modeling using induc- Qin, Y., Wu, T., Li, R., Xie, C., Qiao, Y., Chen, H., Zou, D., and tive learning in the eastern part of permafrost regions in Qinghai- Zhang, L.: Application of ERA Product of Land Surface Temper- Xizang (Tibetan) Plateau, Catena, 126, 98–104, 2015a. ature in Permafrost Regions of Qinghai-Xizang Plateau, Plateau Li, W., Zhao, L., Wu, X., Zhao, Y., Fang, H., and Shi, W.: Distribu- Meteorol., 34, 666–675, 2015 (in Chinese). tion of soils and landform relationships in the permafrost regions Qiu, G. Q.: The principle of geocryological regionalization and the of Qinghai-Xizang (Tibetan) Plateau, Chinese Science Bulletin, compilation of frozen ground map, in: Geocryology in China, 60, 2216–2226, 2015b. edited by: Zhou, Y. W., Guo, D. X., Qiu, G. Q., Cheng, G. D., and Li, X.: The cryosphere information system and its application re- Li, S. D., Science Press, Beijing, 162–170, 2000 (in Chinese). search, Lanzhou Institute of Glaciology and Geocryology, Chi- Ran, Y., Li, X., Cheng, G., Zhang, T., Wu, Q., Jin, H., and Jin, R.: nese Academy of Sciences, 99 pp., 1998. Distribution of Permafrost in China: An Overview of Existing Li, X. and Cheng, G.: A GIS-aided response model of high-altitude Permafrost Maps, Permafrost Periglac., 23, 322–333, 2012. permafrost to global change, Sci. China Ser. D, 42, 72–79, 1999. Riseborough, D.: The effect of transient conditions on an equilib- Li, X., Cheng, G., and Lu, L.: Comparison study of spatial interpo- rium permafrost-climate model, Permafrost Periglac., 18, 21–32, lation methods of air temperature over Qinghai-Xizang Plateau, 2007. Plateau Meteorology, 22, 565–573, 2003 (in Chinese). Riseborough, D., Shiklomanov, N., Etzelmüller, B., Gruber, S., and Li, X., Cheng, G., Jin, H., Kang, E., Che, T., Jin, R., Wu, L., Nan, Marchenko, S.: Recent advances in permafrost modelling, Per- Z., Wang, J., and Shen, Y.: Cryospheric Change in China, Glob. mafrost Periglac., 19, 137–156, 2008. Planet. Change, 62, 210–218, 2008. Roerink, G. J., Menenti, M., and Verhoef, W.: Reconstructing cloud- Lin, Z., Niu, F., Liu, H., and Lu, J.: Hydrothermal processes of free NDVI composites using Fourier analysis of time series, Int. Alpine Tundra Lakes, Beiluhe Basin, Qinghai-Tibet Plateau, J. Remote Sens., 21, 1911–1917, 2000. Cold. Reg. Sci. Technol., 65, 446–455, 2011. Romanovsky, V., Burgess, M., and Smith, S.: Permafrost Tempera- Lin, Z. H., Mo, X. G., Li, H. X., and Li, H. B.: Comparison of Three ture Records: Indicators of Climate Change, EOS Transactions, Spatial Interpolation Methods for Climate Variables in China, American Geophysical Union, 83, 589–594, 2002. Acta Geographica Sinica, 1, 47–56, 2002. Romanovsky, V., Smith, S., and Christiansen, H.: Permafrost ther- Lunardini, V. J.: Theory of n-factors and correlation of data, in: Pro- mal state in the polar Northern Hemisphere during the interna- ceedings of Third International Conference on Permafrost, 1, 40– tional polar year 2007–2009: a synthesis, Permafrost Periglac., 46, 1978. 21, 106–116, 2010. Nan, Z., Li, S., and Liu, Y.: Mean annual ground temperature distri- Sheng, Y., Cao, Y. B., Li, J., Wu, J. C., Chen, J., and Feng, Z. L.: bution on the Tibetan Plateau: permafrost distribution mapping Characteristics of Permafrost along Highway G214 in the East- and further application, J. Glaciol. Geocryol., 24, 142–148, 2002 ern Qinghai-Tibet Plateau, J. Mt. Sci., 12, 1135–1144, 2015. (in Chinese). Shi, Y. and Mi, D.: Map of snow, ice and frozen ground in China Nelson, F. and Outcalt, S.: A Frost Index Number for Spatial Predic- (1V 4 000 000), Beijing, China, Cartographic Publishing House, tion of Ground-Frost Zones, in: Permafrost-Fourth International 1988 (in Chinese). Conference Proceedings, Washington, DC, National Academy Smith, M. W. and Riseborough, D. W.: Permafrost Monitoring and Press, vol. 1, 907–911, 1983. Detection of Climate Change, Permafrost Periglac., 7, 301–309, Nguyen, T. N., Burn, C. R., King, D. J., and Smith, S. L.: Estimat- 1996. ing the extent of near-surface permafrost using remote sensing, Smith, M. W. and Riseborough, D. W.: Climate and the limits of per- Mackenzie Delta, Northwest Territories, Permafrost Periglac., mafrost: a zonal analysis, Permafrost Periglac., 13, 1–15, 2002. 20, 141–153, 2009. Smith, S. L., Burgess, M. M., Riseborough, D., and Mark Nixon, Niu, F., Lin, Z., Liu, H., and Lu, J.: Characteristics of thermokarst F.: Recent trends from Canadian permafrost thermal monitoring lakes and their influence on permafrost in Qinghai-Tibet Plateau, network sites, Permafrost Periglac., 16, 19–30, 2005. Geomorphology, 132, 222–233, 2011. Smith, S. L., Romanovsky, V. E., Lewkowicz, A. G., Burn, C. R., Pang, Q., Zhao, L., Li, S., and Ding, Y.: Active layer thickness varia- Allard, M., Clow, G. D., Yoshikawa, K., and Throop, J.: Thermal tions on the Qinghai-Tibet Plateau under the scenarios of climate state of permafrost in North America: a contribution to the inter- change, Environ. Earth Sci., 66, 849–857, 2011. national polar year, Permafrost Periglac., 21, 117–135, 2010. Prince, S. D., Goetz, S. J., Dubayah, R. O., Czajkowski, K. P., and Tong, B., and Li, S.: The mapping principles and method of the per- Thawley, M.: Inference of surface and air temperature, atmo- mafrost distribution along the Qinghai-Xizang (Tibetan) Plateau spheric precipitable water and vapor pressure deficit using Ad- (1V 600 000), in: Proceedings of the Second Chinese Permafrost vanced Very High-Resolution Radiometer satellite observations: Conference, Lanzhou, Gansu People Press, 75–80, 1983. comparison with field observations, J. Hydrol., 212, 230–249, Wan, Z.: New refinements and validation of the MODIS land- 1998. surface temperature/emissivity products, Remote Sens. Environ., Pu, Z., Xu, L., and Salomonson, V. V.: MODIS/Terra 112, 59–74, 2008. observed seasonal variations of snow cover over the Wan, Z.: Collection-5 MODIS Land Surface Temperature Products Tibetan Plateau, Geophys. Res. Lett., 34, L06706, Users’ Guide, available at: http://www.icess.ucsb.edu/modis/ https://doi.org/10.1029/2007GL029262, 2007. LstUsrGuide/MODIS_LST_products_Users_guide_C5.pdf (last access: 10 March 2015), 10–20, 2009. www.the-cryosphere.net/11/2527/2017/ The Cryosphere, 11, 2527–2542, 2017 2542 D. Zou et al.: A new map of permafrost distribution on the Tibetan Plateau Wan, Z. and Dozier, J.: A generalized split-window algorithm for Wu, Q., Zhu, Y., and Liu, Y.: Application of the permafrost ta- retrieving land-surface temperature from space, IEEE. T. Geosci. ble temperature and thermal offset forecast model in the Tibetan Remote, 34, 892–905, 1996. Plateau, J. Glaciol. Geocryol., 24, 614–617, 2002a (in Chinese). Wan, Z. and Li, Z.: Radiance-based validation of the V5 MODIS Wu, Q., Zhu, Y., and Liu, Y.: Evaluation model of permafrost ther- land-surface temperature product, Int. J. Remote Sens., 29, mal stability and thawing sensibility under engineering activity, 5373–5395, 2008. Cold. Reg. Sci. Technol., 34, 19–30, 2002b. Wan, Z., Zhang, Y., Zhang, Y., and Li, Z.: Validation of the land- Wu, X., Zhao, L., Fang, H., Zhao, Y., Smoak, J. M., Pang, Q., and surface temperature products retrieved from Moderate Resolu- Ding, Y.: Environmental controls on soil organic carbon and ni- tion Imaging Spectroradiometer data, Remote Sens. Environ., 83, trogen stocks in the high-altitude arid western Qinghai-Tibetan 163–180, 2002. Plateau permafrost region, J. Geophys. Res.-Biogeo., 121, 176– Wan, Z., Zhang, Y., Zhang, Y., and Li, Z.: Quality assessment and 187, 2016. validation of the global land surface temperature, Int. J. Remote Xu, Y., Shen, Y., and Wu, Z.: Spatial and Temporal Variations of Sens., 25, 261–274, 2004. Land Surface Temperature Over the Tibetan Plateau Based on Wang, T., Wang, N., and Li, S.: Map of the glaciers, frozen ground Harmonic Analysis, Mt. Res. Dev., 33, 85–94, 2013. and desert in China, 1V 4 000 000, Chinese Map Press, Beijing, Yang, M., Nelson, F. E., Shiklomanov, N. I., Guo, D., and Wan, G.: China, 2006. Permafrost degradation and its environmental effects on the Ti- Wang, W., Rinke, A., Moore, J. C., Cui, X., Ji, D., Li, Q., Zhang, N., betan Plateau: A review of recent research, Earth-Sci. Rev., 103, Wang, C., Zhang, S., Lawrence, D. M., McGuire, A. D., Zhang, 31–44, 2010. W., Delire, C., Koven, C., Saito, K., MacDougall, A., Burke, E., Zhang, T.: Influence of the seasonal snow cover on the ground ther- and Decharme, B.: Diagnostic and model dependent uncertainty mal regime: An overview, Rev. Geophys., 43, 1–23, 2005. of simulated Tibetan permafrost area, The Cryosphere, 10, 287– Zhang, X. M., Nan, Z. T., Wu, J. C., Du, E. J., Wang, T., and You, 306, https://doi.org/10.5194/tc-10-287-2016, 2016. Y. H.: Modeling permafrost distribution in Wenquan Area over Wang, Z., Nan, Z., and Zhao, L.: The Applicability of MODIS Land Qinghai-Tibet Plateau by using multivariate adaptive regression Surface Temperature Products to Simulating the Permafrost Dis- splines, J. Glaciol. Geocryol., 33, 1088–1097, 2011 (in Chinese). tribution over the Tibetan Plateau, J. Glaciol. Geocryol., 33, 132– Zhang, X. M., Sheng, Y., Zhao, L., Wu, J. C., Chen, J., Du, E. J., and 143, 2011 (in Chinese). You, Y. H.: Permafrost distribution using sub-region classifica- Wang, Z., Wang, Q., Zhao, L., Wu, X., Yue, G., Zou, D., Nan, Z., tion and multivariate data in the Wenquan area over the Qinghai- Liu, G., Pang, Q., Fang, H., Wu, T., Shi, J., Jiao, K., Zhao, Y., and Tibet Plateau, Scientia Geographica Sinica, 32, 1513–1520, 2012 Zhang, L.: Mapping the vegetation distribution of the permafrost (in Chinese). zone on the Qinghai-Tibet Plateau, J. Mt. Sci., 13, 1035–1046, Zhao, L., Ping, C. L., Yang, D., Cheng, G., Ding, Y., and Liu, S.: 2016. Changes of climate and seasonally frozen ground over the past 30 Westermann, S., Langer, M., and Boike, J.: Spatial and temporal years in Qinghai-Xizang (Tibetan) Plateau, China, Glob. Planet. variations of summer surface temperatures of high-arctic tundra Change, 43, 19–31, 2004. on Svalbard – Implications for MODIS LST based permafrost Zhao, L., Wu, Q., Marchenko, S. S., and Sharkhuu, N.: Thermal monitoring, Remote Sens. Environ., 115, 908–922, 2011. state of permafrost and active layer in Central Asia during the in- Westermann, S., Langer, M., and Boike, J.: Systematic bias of aver- ternational polar year, Permafrost Periglac., 21, 198–207, 2010. age winter-time land surface temperatures inferred from MODIS Zou, D., Zhao, L., Wu, T., Wu, X., Pang, Q., and Wang, Z.: at a site on Svalbard, Norway, Remote Sens. Environ., 118, 162– Modeling ground surface temperature by means of remote 167, 2012. sensing data in high-altitude areas: test in the central Tibetan Westermann, S., Østby, T. I., Gisnås, K., Schuler, T. V., and Et- Plateau with application of moderate-resolution imaging spec- zelmüller, B.: A ground temperature map of the North At- troradiometer Terra/Aqua land surface temperature and ground- lantic permafrost region based on remote sensing and reanalysis based infrared radiometer, J. Appl. Remote Sens., 8, 083516, data, The Cryosphere, 9, 1303–1319, https://doi.org/10.5194/tc- https://doi.org/10.1117/1.JRS.8.083516, 2014. 9-1303-2015, 2015. Zhou, Y., Guo, D., Qiu, G., Cheng, G., and Li, S.: China Permafrost, Wu, Q. and Liu, Y.: Ground temperature monitoring and its recent Science Press, Beijing, 145–151, 2000. change in Qinghai-Tibet Plateau, Cold. Reg. Sci. Technol., 38, 85–92, 2004. Wu, Q. and Zhang, T.: Recent permafrost warming on the Qinghai- Tibetan Plateau, J. Geophys. Res., 113, 1–22, 2008. The Cryosphere, 11, 2527–2542, 2017 www.the-cryosphere.net/11/2527/2017/
The Cryosphere – Unpaywall
Published: Nov 8, 2017