Key Points:
Abstract
The comprehensive understanding of the occurred changes of permafrost, including the changes of mean annual ground temperature (MAGT) and active layer thickness (ALT), on the Qinghai-Tibet Plateau (QTP) is critical to project permafrost changes due to climate change. Here, we use statistical and machine learning (ML) modeling approaches to simulate the present and future changes of MAGT and ALT in the permafrost regions of the QTP. The results show that the combination of statistical and ML method is reliable to simulate the MAGT and ALT, with the root-mean-square error of 0.53°C and 0.69 m for the MAGT and ALT, respectively. The results show that the present (2000−2015) permafrost area on the QTP is 1.04 × 106km2 (0.80−1.28 × 106km2), and the average MAGT and ALT are -1.35 ± 0.42°C and 2.3 ± 0.60 m, respectively. According to the classification system of permafrost stability, 37.3% of the QTP permafrost is suffering from the risk of disappearance. In the future (2061−2080), the near-surface permafrost area will shrink significantly under different Representative Concentration Pathway scenarios (RCPs). It is predicted that the permafrost area will be reduced to 42% of the present area under RCP8.5. Overall, the future changes of MAGT and ALT are pronounced and region-specific. As a result, the combined statistical method with ML requires less parameters and input variables for simulation permafrost thermal regimes and could present an efficient way to figure out the response of permafrost to climatic changes on the QTP.
Keywords: permafrost; mean annual ground temperature; active layer; climate change; Qinghai-Tibet Plateau
1. Introduction
Frozen ground is an important component of the cryosphere, which exerts strong influences on regional ecology, hydrology and infrastructure engineering (Westermann et al., 2015; Wang et al., 2018a). The Qinghai-Tibet Plateau (QTP) is underlain by typical high-altitude permafrost region, which is undergoing more dramatic climatic warming than its surrounding regions (Wang et al., 2019a). A growing number of studies have reported the present status and predicted degradation of permafrost under various global warming scenarios (Pang et al., 2010, 2012; Zhang and Wu, 2012a; Guo and Wang, 2017; Xu et al., 2017a; Wang et al., 2018a). The degradation of permafrost may trigger the release of organic carbon into the atmosphere (Cheng and Wu 2007; Wu et al., 2017a; Chang et al., 2018; Wang et al., 2018b; Ran et al., 2018). It is also a potential threat to engineering construction and maintenance. However, most of these studies are based on linear statistical models and equilibrium models, and mainly focused on identifying the extent of permafrost, while researches on the present and future change of ground thermal regimes (including: the mean annual ground temperature, MAGT, and the active layer thickness, ALT) are relatively rare (Zhang et al., 2012a; Wang et al., 2019a). The changes of MAGT and ALT could affect the ecosystem of the QTP by altering the ground ice evolution, hydrological processes, vegetation dynamics and carbon cycling, etc. (Yang et al., 2010a; Wu et al., 2016; Niu et al., 2019; Hu et al., 2020). Therefore, it is of great importance to investigate present and future changes of the MAGT and ALT in the permafrost region (Qin et al., 2017; Zhang et al., 2018).
Permafrost is a thermally-defined subsurface phenomenon (Westermann et al 2015). Satellite sensors could obtain limited surface information, and only portion of the microwave remote sensing could penetrate several centimetres underground (Zhao et al., 2011; Michaelides et al., 2018; Qu et al., 2019). In general, it is difficult to use remote sensing to directly obtain information on changes in the physical state of permafrost (Yang et al., 2019). The current research on permafrost thermal regime is mostly focus on either in situ observing or modeling using atmospheric circulation models (Westermann et al., 2015). Most of the existing modeling frameworks require ground-based measurements as model inputs, while the in situ observations of permafrost are relatively sparse and highly non-uniform in cold regions. The long-term and continuous in situ observation sites for permafrost on the QTP are mostly located along the Qinghai-Tibet Highway and Railway, and other regions are less well distributed (Hu et al., 2015; Qin et al., 2017; Zheng et al., 2019). The absence of observation data would greatly weakens the accuracy of simulation results. Therefore, it is challenging to select reliable modeling approaches with limited data to obtain the occurrence of permafrost and its projection due to climate change.
At present, the simulation studies on the ALT and soil thermal state of the QTP fall into two categories, including equilibrium models and mechanistic transient models. (Riseborough et al., 2008; Qin et al., 2017; Aalto et al., 2018). The most commonly used equilibrium models include Stefan formula (Zhang and Wu 2012a; Xu et al., 2017a), Kudryavtsev formula (Pang et al., 2009; Wang et al., 2020a), the N factor (Nan et al., 2012), and the Temperature at the Top of the Permafrost model (TTOP) (Zou et al., 2017). The form of the equilibrium model is relatively simple and requires fewer driving data for input (Riseborough et al., 2008; Pang et al., 2009). However, this type of model tend to show poor portability. In contrast, mechanistic transient models consider more details of the hydrothermal exchange processes between the atmosphere and ground. Examples of this model include the Community Land Model (CLM; Oleson et al., 2010; Fang et al., 2016; Chen et al., 2017), Noah (Gao et al., 2015; Chen et al., 2015), the Geomorphology-based Eco-hydrological Model (GBEHM; Zheng et al., 2019), the SHAW model (Guo et al., 2011; Liu et al., 2013), and the CoupModel (Zhang et al., 2012b; Hu et al., 2013). Nevertheless, the processes of these models are complex and often insufficiently account for the hydrothermal dynamics, with the understanding of the soil physical mechanisms increase, the parameterization processes will become more complex (Harris et al., 2009; Hu et al., 2015; Guo and Wang, 2016). In addition to the transient models mentioned above, in recent years, the fine-scale tightly coupled hydro-thermal modeling of permafrost has also made great progress (e.g., models like ATS, Jafarov et al., 2018; and SUTRA, Walvoord et al., 2019, etc.), These models are typically based on a multidimensional solution to address fully coupled surface/subsurface permafrost thermal hydrology, which have played an important role to study the permafrost of local scale and microtopography (Painter et al., 2016).
Physics-based mechanistic models are currently the popular methods to study the permafrost, and the simulation results can show high accuracy. However, even with significant improvements in computer technology and algorithm simulation (Westermann et al., 2016), the current modeling still exists a trade-off between modeling resolution and size of the geographical domain (Etzelmüller, 2013). Especially in the case of lack of data and insufficient computing resources, the extensive application of physics-based mechanistic models would be limited. Whereas, the combined statistical method with machine learning (ML) can make up these deficiencies. In recent years, their great power in permafrost modeling has been confirmed (Xu et al., 2017b; Chadburn et al., 2017; Aalto et al., 2018). The main purpose of statistical and ML model is to identify the relationship between a dependent variable and one or more explanatory variables (Wheeler et al., 2013). They can easily explain environmental conditions related to topography and land cover, whereas these factors may be difficult to express with physical parameters (Etzelmüller, 2013). Due to the good coupling between air temperature (often characterized by mean annual air temperature or cumulative temperature sums) and ground thermal regime (Chadburn et al., 2017; Aalto et al., 2018), the subsurface (<10−20 m) soil thermal conditions respond well to climate change at the decadal scale (Aalto et al., 2018). In addition, precipitation type (e.g., snow, rain and sleet) and local environmental predictors (e.g., topography, underlying surface condition and soil texture condition) have great impacts on soil hydrothermal dynamics and the surface radiation budget (Lee et al., 2013; Zhu et al., 2019).
Therefore, in this study, we employed statistical and ML methods to investigate the MAGT and ALT across the QTP. The objective is to verify the applicability of the combined method on the QTP and quantitatively assess the present and future status of QTP permafrost. Firstly, we identified the critical factors which determining the occurrence of permafrost. Secondly, we used the combined modeling approaches integrated with field observation data, meteorological data and geospatial environmental predictors to calculate the present MAGT and ALT. Thirdly, the present results were benchmarked against in situ measurements of ALT and ground temperatures. Finally, the optimal modeling framework was used to predict future MAGT and ALT forced by different RCPs. The projection of the MAGT and ALT can serve as a useful reference and provide important information for the study of climate change, hydrology, ecology, and geohazards resulted from permafrost degradation on the QTP.
2. Data and Methods