The three monitoring sections selected for this research are placed in the permafrost section of the QTR between the Erdaogou–Ando valley, as shown in Figure 1.

The information on the embankment section is listed in Table 1. Three monitoring sections are located in the warm permafrost area with an annual average ground temperature higher than −1°C. The P1 section is TE, situated in the western part of the Erdaogou valley, and the mean average ground temperature (MAGT) of permafrost under the embankment is −0.65°C. The P2 section is UCRE, located on the west bank of the Buqu River, and the MAGT of permafrost under the embankment is −0.46°C. The embankment basement is filled by the crushed-rock layer with a thickness of 1 m, and the diameter of crushed rock is 10–15 cm. The thickness of the crushed rock on the left and right sides of the slopes is 1.0 m and 0.6 m, respectively, and the diameter is 10–15 cm. The P3 section is the CRRE, located on the western slope of the Toju Valley near the southern boundary of permafrost. The MAGT of permafrost under the embankment is −0.20°C, the thickness of the crushed rock on the left and right sides of the slope is 1.2 and 0.8 m, respectively, and the diameter of the crushed rock is 15–20 cm. The purpose of setting different thicknesses of rock revetment on the left and right sides is to reduce the temperature difference between the left and right sides of the embankment by relying on the active cooling effect of the rock layer. It has been confirmed in previous studies (Lai et al., 2004; Zhang et al., 2016).

As determined by the drilling survey, the geological condition of permafrost under the three embankment sections is depicted in Figure 2. The TE natural permafrost table (NPT) is about 3.8 m, the volume ice content is between 15% and 40% from 3.8 to 10 m, and the soil consists primarily of gravel sand and gravel clay from the top to bottom. The NPT of the UCRE is roughly 2.4 m, and the volume ice content is between 15% and 35% from 2.4 to 10 m. The soil is composed mainly of silty sand, weathered mudstone, and sandstone from the top to bottom. The NPT of the CRRE is about 2.6 m, and the volume ice content is >50% from 2.6 to 2.8 m. The volume ice content of the soil ice layer ranges between 15% and 30% from 2.8 to 10 m, and the soil consists mainly of silty clay, gravel clay, and weathered mudstone from the top to bottom.

This study addresses three different embankment structures in the warm permafrost zone using the embankment ground temperature and deformation data obtained from the long-term monitoring system of the QTR. Figure 3 illustrates the layout of ground-temperature boreholes and deformation monitoring points in each monitoring section. Figure 3A shows the layout of the ground-temperature boreholes. The distance of the natural borehole from the left side of the embankment is 20 m. The depth of the natural borehole is 18 m. The depth of both the left- and right-shoulder boreholes is 20 m. Thermistors are placed in the boreholes at 0.5-m intervals from the ground surface to the depth of 10 m and then at 1.0-m intervals to depths below 10 m. These thermistors were produced by the State Key Laboratory of Frozen Soil Engineering, China, and the measurement accuracy is ± 0.05°C. The ground temperature is automatically collected by the data logger (DT500) on a daily basis. Figure 3B shows the layout of deformation monitoring points at each observation profile. Eight deformation monitoring points and one reference point were installed on the embankment surface. The deformation monitoring points were fixed 20 cm under the embankment surface and embedded into the embankment shoulder; deformation observation points were arranged in four rows, and the distance between rows was 20 m. The reference point was also located on the embankment shoulder, and a longer metal rod was embedded into the embankment by a depth of 20 m with the upper end exposed on the embankment surface. A level instrument with a precision of ± 0.1 mm was utilized to manually measure the difference displacement between the monitoring points and the reference point with at frequency of once a month. Wu et al. (2008), Yu (2008) and Ma et al. (2011) introduced the relevant work on the ground temperature and embankment deformation measurement. A negative value of deformation indicates settlement, while a positive value represents frost heave. In addition, the left and right sides of the embankment are defined in the direction from Golmud to Lhasa; that is, the left side of the embankment corresponds to the sunny side, and the right side corresponds to the shady side. Roughly 16 years of continuous deformation monitoring data were collected from October 2005 to December 2020.

The settlement values of the left and right shoulders in the three monitoring sections for eight monitoring points and the corresponding average values are illustrated in Figure 4. During the 16 years of operation of the TE, the accumulated settlement of the left and right shoulders was 307–382 mm and 126–251 mm, with an average value of 356 mm and 181 mm, respectively. In addition, the accumulated settlement difference between the left and right shoulders was 175 mm (Figure 4A). There is no evident tendency of frost heave or settlement slowdown, and the settlement has been continuously developed without any slowdown trend. For the UCRE (Figure 4B), the cumulative average settlement value of the left shoulder during the 16 years of operation was 92–150 mm, with an average value of 120 mm. In contrast, the cumulative average settlement value of the right shoulder ranged from 45 mm to 65 mm, with an average value of 54 mm, while the cumulative settlement difference between the left and right shoulders was 66 mm. Before 2010, the cumulative deformation curves of the left and right shoulders exhibited a slightly increasing tendency due to the influence of frost heave during the cold season. After 2011, the settlement growth rate in the right shoulder reduced considerably. For the CRRE (Figure 4C), the cumulative settlement value of the left shoulder ranged from 87 mm to 171 mm, with an average value of 127 mm, and the cumulative settlement value of the right shoulder was between 58 and 131 mm, with an average value of 94 mm, and the cumulative settlement difference between the left and right shoulders was only 31 mm. Compared to the UCRE, the CRRE settlement curve fluctuates more. This indicates that the thicker and larger crushed-rock slope absorbs more cold energy from the outside environment in the cold season, and frost heave is larger in the cold season. This phenomenon is consistent with the findings in the literature (Zhang, et al., 2006; Liu, et al., 2017), and it can also be related to external air temperature variation, the water content of the frozen soil under the embankment, the direction of the embankment, and the dominant wind direction of the annual year. However, additional field monitoring data are required to support this hypothesis.

The statistical values of the annual settlement between the left and right shoulders in the three monitored sections from 2006 to 2020 are displayed in Figure 5. As demonstrated in Figure 5A, the annual settlement of the TE varies widely and has no discernible pattern over 15 years, and the annual settlement of the left shoulder is in the range of 25.0 mm–29.7 mm during the 2016–2019, while the annual settlement of the left shoulder is in the range of 9.7 mm–21.7 mm in all other years. The development pattern of the annual settlement of the right shoulder differed slightly from that of the left shoulder, especially in 2016–2019. In the 3 years, the settlement of the left shoulder was larger (all exceeding 29 mm), while the settlement of the right shoulder was within the range of 11.0 mm–15.2 mm, and the difference in the annual settlement was particularly pronounced. As illustrated in Figure 5B, the settlement in the left shoulder was maintained within the range of 8.3 mm–12.3 mm. In addition, the settlement in the right shoulder was maintained in the range of 1.2 mm–6.2 mm for the first 5 years (2006–2010) of the UCRE, and the disparity in the annual settlement between the left and right shoulders was larger. From 2011 to 2015, the left shoulder settlement remained within the range of 3.5 mm–7.2 mm, while the right shoulder settlement was simultaneously in the range of 0.4 mm–3.8 mm. The annual settlement steadily declines, and the annual settlement of the right shoulder is always smaller than that of the left one in the corresponding years. The annual settlement between the left and right shoulders increased from 2016 to 2020, and the reasons for this phenomenon must be analyzed using ground temperature field data. In contrast, in the case of the approximate water content of permafrost and a greater embankment height (the UCRE is higher than the TE, which indicates a greater self-weight load of the embankment), the annual settlement of the UCRE has been kept at a lower value than that of the TE. It illustrates that the UCRE is effective in controlling the settlement. As depicted in Figure 5C, the settlement of the CRRE was larger in the first 2 years (2006–2007). It demonstrates that the traffic load and the self-weight load brought by the highest embankment (the embankment height is 7 m) caused a greater compression deformation between the active and warm permafrost layers under the embankment. The settlement in the left and right shoulders decreased after 2008 and preserved a more stable settlement rate. Compared to the UCRE, the settlement difference between the left and right shoulders of the CRRE is smaller.

The APT on the left and right shoulders and the difference between them to evaluate the settlement mechanism of the three embankment structures are provided in Figure 6. As shown in Figure 6A, the APT in the left and right shoulders of the TE exhibits a trend of a relatively steady increase along the operational year. It indicated that the downward shift of the APT is a significant factor in the development of the annual settlement in the TE. In addition, except for 1 year (the difference was 1.0 m in 2006), the difference in APT between the left and right shoulders was within the range of 0.5 m, and the impact of the shady–sunny slope on the APT of the TE was not essential. Since the first year of operation was 2006, the lagging impact of thermal disturbance in the preconstruction process persisted, resulting in a greater discrepancy between the APTs of the left and right shoulders. For the remainder of the time, the smaller difference between APTs of the left and right shoulders can be due to the shorter length of the embankment slope in the TE (the height of the embankment is only 0.9 m) and the slight heat difference between shaded and sunny slopes resulting from external factors such as solar radiation. For the UCRE (Figure 6B), the APT in the left and right shoulders varies within a small range over a period of 16 years (6–7 m for the left shoulder, 4.2–4.5 m for the right shoulder, and the difference between APTs of the left and right shoulders is 1.6–2.5 m). It indicated that from 2006 to 2020, the change in the APT in the UCRE was not apparent. It was not the primary reason for the variation in an annual settlement in the UCRE in the operational year. For the CRRE (Figure 6C), the APT under the left shoulder decreases yearly (from 8.6 m in 2005 to 6.8 m in 2020), while the APT under the right shoulder remains relatively constant between 4.7 m and 5.5 m. The difference in APT between the left and right shoulders similarly declines gradually and corresponds to a yearly reduction of the APT in the left shoulder. This implies that the CRRE greatly increases the APT as the operational years increase, and the annual settlement trend positively correlates with the APT change trend.

The influence of the shady–sunny slope effect on the settlement characteristics of embankment is an issue that needs to be addressed. Figure 7 depicts the cumulative settlement values of the left and right shoulders and their corresponding differential settlement values for the three monitored sections. For the TE (Figure 7A), the differential settlement of the left and right shoulders due to the shady–sunny slope impact has existed since the first operational year, 2006. In addition, the difference in the cumulative average settlement values between the left and right shoulders after 16 years of operation exceeds 160 mm. The differential settlement in the left and right shoulders between October 2005 and December 2018 maintains a rough annual growth rate (approximately 11 mm/a), while the variation of the cumulative settlement values between the left and right shoulders after December 2018 remains between 155 and 163 mm with a significantly slower annual growth rate. This can be due to the warming climate after 2018, when the settlement rates of the left and right shoulders tended to be comparable. After 16 years of operation, the average cumulative settlement value in the left shoulder of the UCRE is 125 mm, while that in the right shoulder is 58 mm (Figure 7B). The difference between the average cumulative settlement values in the left and right shoulders is more than 60 mm. In the second half of 2005, there was no notable difference in the settlement of the left and right shoulders, and the variation between 2006 and 2018 was approximately 11 mm/a. This indicates that the annual growth rate of settlement of the left and right shoulders caused by the effect of the shady–sunny slope during this period does not change significantly with the increase in the operational year. However, the growth rate of the differential settlement after 2018 significantly slowed down, which is similar to the results of the differential settlement of the TE after 2018, and the influence of climate on the differential settlement of shady–sunny slopes is verified. The differential settlement between the left and right shoulders of the CRRE remains within the range of 35 mm after 16 years of operation (Figure 7C). Before 2008, the differential settlement of the left and right shoulders constantly fluctuated at zero, which indicates that the shady–sunny slope effect of the settlement deformation of the block-berm roadbed in the first 3 years of operation was weak. From 2008 to 2012, the differential settlement deformation maintains an annual growth rate of approximately 3.5 mm/a. The influence of the shady–sunny slope effect on the settlement was prominent, and after 2012, the annual growth rate of the differential settlement slowed. In addition, comparing the results of the UCRE and the CRRE reveals that although the accumulated settlement of the left shoulder is the same in both monitoring sections (125 mm and 123 mm), the difference in the settlement of the right shoulder (shady slope side) is greater than 35 mm. This indicates that in addition to the influence of the local hydroclimatic environment on the settlement, UCRE has a greater controlling effect on the settlement than CRRE. Before 2008, the impact of the shady–sunny slope on the settlement of the CRRE was negligible. In contrast, after 2008, the effect gradually became more pronounced, which is different from the TE and the UCRE.

The evolutionary characteristics of the ground temperature field and the settlement under the left and right shoulders of the three monitored sections with the year of operation are depicted in Figure 8. Figure 8A demonstrates that the APT in the left shoulder of the TE decreases significantly as the operational year increases, and the APT also drops annually. However, the seasonal maximum frost depth and the APT cannot be connected after 2014, and the thawed soil layer is created between the seasonal maximum frost depth interface and the APT (Wang et al., 2021). The development of the settlement rate in the years after the formation of the thawed soil is critical, which is verified under the left and right shoulders. Notably, the formation of the thawed interlayer begins at different times in the left and right shoulders of the TE (in the left shoulder, it began at 2014, and in the right shoulder, it began at 2009) (Figures 8A, B), and the thickness of the thawed interlayer gradually grows as the operational year increases. This indicates that under the combined influence of climate warming and engineering activities, the energy balance system of the original natural ground surface is broken, and the net annual heat absorption of permafrost beneath the embankment is higher than before. As the APT declines, the active layer in the permafrost zone becomes thicker. When the shady–sunny slope effect occurs due to the alignment of the embankment, the thawed interlayer under the embankment is affected by the asymmetric temperature field, resulting in the formation of an asymmetric thawed interlayer. The cooling effect of the UCRE has a substantial protective effect on the ground temperature (Figures 8C, B), and the APT under the left shoulder changes very little after 16 operational years, while the APT under the right shoulder does not vary significantly until 2013 and tends to decrease after 2013. This demonstrates that the UCRE considerably impacts permafrost protection as the number of operational years increases. In addition, the settlement rate leads to no significant disturbance in the ground temperature field for both the left and right shoulders. The APT under the left shoulder of the CRRE varies little from year to year (Figure 8E), whereas the thawed interlayer appearance annually disrupts the settlement development. In years without a thawed interlayer, the fluctuation amplitude of the settlement curve is slightly larger, which can be related to the variation in the winter temperature between years. The cooling effect of the CRRE has a pronounced effect on the ground temperature under the right shoulder (Figure 8F). During the 16 years of operation, the APT varied minimally, and the settlement developed at a more constant deformation rate after 2008, except for small fluctuations in the first 3 years of operation. Hence, the change control of the APT under the embankment is more obvious for the CRRE, irrespective of whether the slope is sunny or shady. However, the trend of the accelerated settlement of the embankment due to the formation of a thawed interlayer on the sunny slope side must be considered.