Assessments of land subsidence along Rizhao-Lankao High-speed Railway at Heze,

The Heze section of Rizhao-Lankao High-speed Railway (RLHR-HZ) has been under 12 construction since 2018 and will be operative by the end of 2021. However, there is a concern that land 13 subsidence in Heze region may affect the normal operation of RLHR-HZ. In this study, we investigate the 14 contemporary ground deformation in the region between 2015 and 2019 by using more than 350 C-band 15 interferograms constructed from two tracks of Sentine-1 data over the region. The Small Baselines Subset 16 (SBAS) technique is adopted to compile the time series displacement. We find that the RLHR-HZ runs 17 through two main subsidence areas: One is located east of Heze region with rates ranging from -4 cm/yr to 18 -1 cm/yr, and another one is located in the coal field with rates ranging from -8 cm/yr to -2 cm/yr. A total 19 length of 35 km of RLSR-HZ are affected by the two subsidence basins. Considering the previous 20 investigation and the monthly precipitation, we infer that the subsidence bowl east of Heze region is due to 21 massive extraction of deep groundwater. Close inspections of the relative locations between the second 22 subsidence area and the underground mining reveals that the subsidence there is probably caused by the 23 groundwater outflow and fault instability due to mining, rather than being directly caused by mining. The 24 InSAR-derived ground subsidence implies that it’s necessary to continue monitoring the ground 25 deformation along RLSR-HZ. 26

14 https://doi.org/10.5194/nhess-2020-176 Preprint. Discussion started: 24 June 2020 c Author(s) 2020. CC BY 4.0 License. noise and accumulative deformation on wrapped phase. Then, all the interferograms are inspected to 23 distinguish the noisy interferograms which will be excluded from the network. In addition, the temporal 24 intervals of interferograms are not very uniform. There are some gaps being more than 84 days, which 25 results in a disconnected network for the auto-generated interferograms. Therefore, some interferograms 26 with high coherence are manually appended to the original stacks to form a connected network which is 27 essential for StaMPS-SB. Finally, two connected networks containing 192 interferograms for S1-40 and 28 171 interferograms for S1-142 are generated (shown in Fig. 2). In these networks, each image is connected 29 to at least two other images, by which the temporal sampling rate is increased by an average factor of 2.9. 30 Taking advantage of the adequate interferograms, the model parameters such as deformation, topographic 31 errors and atmospheric artifacts can be estimated more accurately comparing with single master approach. bandwidth is more than 95% in most cases. Additionally, the range bandwidth is larger than 40 MHz and 4 almost all of the geometric baselines are shorter than 150 m. As a result no spectral filtering is applied to 5 avoid coarsening the resolution. In contrast, an adaptive spatial filtering algorithm is introduced and 6 implemented to improve the interferometric phase and coherence while preserving the image details 7 (Ferretti et al., 2011; Goel and Adam, 2011; Parizzi and Brcic, 2011). We identify the statistically 8 homogenous pixels (SHP) using the Kolmogorov-Smirnov (KS) test based on the amplitude of images. A 9 rectangular window with a dimensions of 19  13 (azimuth  range) pixels is used to identify SHP, on 10 which a spatial filtering will be implemented if the number of SHP is more than 18. 11 StaMPS-SB selects coherent targets based on the phase characteristics (Hooper 2008), 12  is the wrapped phase of candidate pixel x in the i th filtered interferogram. is the spatially uncorrelated terms due to look angler error 15 which is correlated with the perpendicular baseline. N is the number of interferograms. Candidate pixels 16 are firstly selected using the amplitude difference dispersion specifying (a threshold of 0.6 is adopted). 17 Then, these candidates will be filtered in small patches, such as 50 m  50 m, to estimate  Once the coherent targets have been determined, the differential interferometric phase of each coherent 22 target is corrected using the estimated spatially uncorrelated terms (i.e.
, the residual phase mainly consisted of ground deformation, atmospheric artifacts and orbit error, 24 can then be unwrapped using the three-dimensional (3-D) algorithm. Firstly, the difference phase between 25 neighboring coherent targets is preliminarily unwrapped in time under the Nyquist assumption. A priori 26 probability density function (PDF) can be built in each interferogram based on the unwrapped difference 27 phase in time dimension. These PDFs are then used to search for the optimization routines of unwrapping 28 in space to achieve the final unwrapped results (Hooper 2010). Subsequently, the time series deformation of 29 each coherent target can be extracted by temporal and spatial filtering based on the characteristics of each 30 https://doi.org/10.5194/nhess-2020-176 Preprint. Discussion started: 24 June 2020 c Author(s) 2020. CC BY 4.0 License. Fig. 3 shows the displacement rates derived from S1-40 and S1-142. The negative displacement indicates 11 that the ground was subjected to subsidence during this period.

16
As no in-situ data is publicly available, we assess the consistency and precision of InSAR results by a 17 cross comparison (shown in Fig. 4) of the displacement rates derived from the S1-40 and S1-142. Although 18 the physical parameters of sensor is same with each other, the location of two sets of measurement points 19 (MPs, consisting of PS and DS pixels) is slightly different due to the different geometric parameters. In 20 order to identify the common MPs, both results from S1-40 and S1-142 are resampled to a grid with a 21  S1-40 and S1-142 are shown, which leads to a denser MPs in the overlapping area. Two ellipses show the major subsidence 3 area. The triangles (i.e., pink triangles named as R1-R7, red triangles named as U1-U4 and black triangle named as M1-M4) 4 delineate the displacement features in rural area, urban area and mining area discussed in detail in Fig. 6-8

6
To make further investigation on the ongoing settlement, we estimate the displacement profile along the 7 RLSR-HZ and show the results in Fig.5b. It can be seen that the deformation pattern is not homogenous 8 along the RLSR-HZ. The ground experiences strongest subsidence in the underground mining area, 9 followed by the urban area. We also observe slightly deformation in rural area, where the displacement 10 rates primarily are in the range of -2 cm/yr to 1 cm/yr. However, another subsidence areas with the 11 maximum rates up to -3 cm/yr appear near the 116.16°. This area is offset from the coalfield by 3.5 km, so 12 the subsidence phenomenon there might be related to the underground mining. It suggests that the groundwater is still excessively exploited and the groundwater recharge is insufficient to 23 supply the exploitation. Xu et al., (2017) pointed out that more than 1 billion tons of groundwater is over 24 extracted annually by more than 137000 wells in Heze region. 25 Besides, Fig .6 shows that the behavior of the time series displacement significantly differs at rural area 26 and urban area. Comparing with rural area, the ground experiences more serious subsidence in urban area. 27 One reason for the large settlement in urban area is that the deep groundwater (over 200 m depth) is 28 massively extracted to meet the industrial, except for shallow groundwater to domestic use. Approximately    Fig. 8 show that the accumulative subsidence above coal does not exceed 50 cm during more 14 than four years. In addition, Fig. 8 reveals a significant linear displacement trend. In fact, the 15 mining-induced subsidence is generally nonlinear with high rates, which can reach up to several meters in a 16 short time (e.g, several months). These characteristics indicate that the subsidence near M1~M4 may not be 17 directly induced by underground mining. 18 Note that the underground mining inevitably produces many fracture fields which may be interconnected 19 with the faults, forming a large interconnected network. The groundwater will flow into the working planes 20  In Fig. 9, six interferograms of box (a) in Fig. 3a are shown. It is observed that there are several active 3 subsidence areas, represented by dense fringes (ellipses in Fig. 9), locate in 5~8 km north of M1 and M2. 4 The loss of signal in the central of these subsidence areas is probably due to the displacement being too 5 large to be detected by MT-InSAR with the C-band Sentinel-1 data. This conforms to the mining-induced 6 displacement field which presents a bowl-shaped subsidence pattern, as observed in other studies (He et al., 7 1994;Litwiniszyn 1956;Zhu et al., 2020). The dense fringes, i.e., strong deformation, suggests there is 8 active underground mining just below this subsidence basin (referred as primary subsidence basin and 9 denoted by ellipses in Fig. 9). In contrast, all the six interferograms and the corresponding displacement 10 rates shown in Fig. 3 reveal a relatively slight subsidence near M1 and M2 (referred as secondary 11 subsidence basin) in comparison to the primary subsidence basin. Close inspections of the distribution of 12 fringes in Fig. 9, the ground deformation seems to progress to M1 and M2 area along two 'galleries' 13 (dashed line in Fig. 9). A plausible explanation of this deformation near M1 and M2 could be the 14 groundwater outflow to working panels and is drained out, which finally triggers the ground subsidence. 15 16 Fig. 9 .Example of interferograms produced from S1-40 and S1-142 of box (a) in Fig. 3. The ellipses show the primary 17 subsidence area induced by underground mining. One color fringe stands for approximately 2.8 cm displacement in the LOS.

18
The dashed lines show the secondary subsidence area caused by complicated factors, e.g, groundwater drain out and/or fault 19 activation due to underground mining.

20
In addition, a similar phenomenon is observed in an area (box (b) in Fig.3a) 5-11 km south of RLSR-HZ. 21 Fig. 10 shows six interferograms of this area. In the right (ellipses in Fig. 10), the dense fringes in this 22 primary subsidence area imply active underground mining. However, an arc-shape secondary subsidence 23 area (dashed line in Fig. 11) with a length of about 14 km presents in 3 km to the left of the primary 24 subsidence area. The arc-shape subsidence does not agree with the law of the ground movement induced by 25 underground mining. Fig.3 shows the arc-shape subsidence is consistent with the fault, which implies that 26 by this subsidence. 23 Considering the previous investigation coupled with information of human activities, we conclude that 24 the subsidence is mainly caused by extraction of groundwater and underground mining: 25 -Combining the known previous investigations and the monthly precipitation, we find two patterns of 26