Monitoring and Prediction of Surface Subsidence by Combining SSA-LSTM and TS-InSAR - A Case Study of Kunming Urban Area

doi:10.21203/rs.3.rs-5229068/v1

To enhance our understanding of urban surface deformation mechanisms and to prevent geohazards, this study utilizes two time-series Interferometric Synthetic Aperture Radar (InSAR) methods with Sentinel-1 data: Persistent Scatterer-InSAR (PS-InSAR) and Small Baseline Subset-InSAR (SBAS-InSAR). These complementary methods jointly validate surface subsidence data in Kunming's urban area from 2020 to 2022. Utilizing this data, the study introduces and implements a Long Short Term Memory (LSTM) network model, which is optimized by the Sparrow Search Algorithm (SSA), to forecast and analyze future surface subsidence trends in Kunming. The results reveal that: (1) Kunming's urban area is undergoing persistent, large-scale surface subsidence, with cumulative subsidence measured at 122.8 mm. (2) Geographical location significantly influences the subsidence areas. (3) The subsidence in Area B is predominantly influenced by vehicular traffic. (4) The SSA-LSTM model accurately predicts the future trajectory of surface subsidence in Kunming's urban environment. (5) The complexity of the causes of surface settlement in Kunming is linked to natural factors, including geography, climate, and geology, as well as human activities such as rapid urbanization, groundwater extraction, subsurface construction, and mining operations. In conclusion, through a thorough, multifaceted analysis employing various methods, this study offers fresh insights and a robust scientific foundation for grasping the dynamics of surface subsidence in Kunming and for the anticipation and prevention of geological disasters. Subsequent research will continue to investigate the myriad factors influencing surface subsidence to more precisely forecast and mitigate the risks of geohazards. This work is vital for informed urban planning and the promotion f sustainable development.

Earth and environmental sciences/Natural hazards

Earth and environmental sciences/Environmental sciences

SSA-LSTM

TS-InSAR

surface settlement

Kunming

Sentinel-1

In the development and construction of modern cities, urban surface settlement has become a significant issue for many large urban centers due to the dense distribution of buildings, the development and utilization of urban groundwater resources, and particularly soft foundations. This concern has escalated into a prominent geohazard, as over time, surface settlement can lead to severe consequences such as the collapse of urban roads, structural cracks in buildings, and even the tilting and potential collapse of structures [1][2]. Additionally, surface settlement not only threatens urban stability and the safety of buildings but may also indirectly lead to the depletion of groundwater sources and environmental pollution [3].

Traditional surveying methods, such as geodetic leveling and Global Navigation Satellite System (GNSS) measurements, offer high accuracy but are labor-intensive, costly, and time-consuming, with long data acquisition cycles and limited monitoring areas, resulting in a relatively sparse data collection [4]. In contrast, Interferometric Synthetic Aperture Radar (InSAR) technology, which leverages global microwave remote sensing satellites, has gained widespread application in monitoring urban surface settlement [5].

Ground subsidence in Kunming City has become a notable case of urban ground subsidence in the fracture subsidence basins found on China's western plateau. This phenomenon is characterized by a broad range and a significant rate of occurrence. In recent years, the expansion and acceleration of this issue have become apparent. However, related research commenced relatively late and mainly focused on identifying the extent of the subsidence area, which has not kept pace with the needs of Kunming's ambitious "one lake and four pieces" construction project. The application of time-series InSAR methods, such as PS-InSAR and SBAS-InSAR, for surface settlement monitoring has been extensively implemented [6][7][8]. Recent studies have demonstrated that using these methods in conjunction with Sentinel-1 radar data successfully captures surface settlement data in Kunming City. The data collected not only elucidates the distribution characteristics of surface subsidence funnels and patterns in Kunming's urban area but also confirms the feasibility and practicality of PS-InSAR and SBAS-InSAR methods in detecting minor urban surface deformations [9]. However, there remains a notable time lag between the satellite imaging, the generation of monitoring results, and real-time surface settlement, making accurate prediction of surface settlement trends a research priority [10].

The emergence of neural network models has introduced new dimensions to the field of surface settlement prediction. Long Short Term Memory (LSTM) networks, a class of neural networks designed for time-series data, offer distinct advantages in learning temporal data features [11][12][13]. However, parameter selection in traditional LSTM models can present challenges that affect the stability of the prediction model. The Sparrow Search Algorithm (SSA), an optimization method inspired by the foraging behavior of sparrows, with strong global search capabilities and convergence properties, can be used to optimize the parameters of the LSTM model, thus enhancing the model's performance and the accuracy of its predictions [14].

This study aims to apply SSA-LSTM and TS-InSAR methods to explore new techniques for enhanced monitoring and prediction of surface settlement, focusing on a case study in the Kunming urban area. This approach aims to establish a solid scientific foundation for urban planning and geological disaster prevention and control. Real-time surface deformation data for Kunming's urban area, spanning from 2020 to 2022, was gathered and cross-validated using PS-InSAR and SBAS-InSAR methodologies. Utilizing the TS-InSAR monitored surface deformation data, this study proposes the SSA-LSTM as a Surface Settlement Prediction Model to forecast the trends of surface subsidence. To ensure the accuracy of the prediction model, RMSE, MAE, and R² have been adopted as performance evaluation indicators. The data sources for this research include Sentinel-1A data provided by the European Space Agency (ESA) and Precise Orbit Determination (POD) files used for image refinement.

2.1. Study Area

Kunming, the capital of Yunnan Province, is a key urban center in Southwest China and a renowned tourist destination. It sits within a prototypical plateau fault depression basin, where the topography rises in the northern, eastern, and southeastern sectors, creating an extensive corniche landscape [15]. The groundwater in this basin is relatively static, owing to the high viscosity of the geological strata, leading to foundations that are often semi-saturated with water and rich in organic material. These characteristics render the area particularly prone to subsidence.

The urban core of Kunming is predominantly situated in the central part of the Kunming Basin and spans approximately 525.2 square kilometers. This region serves as the city’s political, cultural, and economic center, with varied land use patterns reflecting its diverse functions. The city's growth, combined with assorted environmental factors, has led to ongoing alterations in the soil composition and geological layers. These changes are intricately linked to the local geological structure, landform features, and the underlying hydrogeological context, making the Kunming Basin a complex urban setting for the surveillance and governance of surface subsidence.

A thorough, longitudinal examination of observational data has disclosed extensive ground subsidence across the urban expanse of Kunming, especially in the downtown area. Here, the swift pace of urban development, excessive extraction of groundwater, and soil erosion have collectively undermined the city’s foundation. The resultant soil displacement and ground subsidence have noticeably affected the stability of urban structures and impeded regional economic growth [9].

2.2. Data

The Sentinel-1A satellite, part of the Copernicus programme managed by the European Space Agency, is equipped with a C-band Synthetic Aperture Radar (SAR) that provides persistent SAR imagery to users [16]. With a revisit interval of 12 days, it systematically captures fine-scale urban land movements. For this study, 41 Sentinel-1A TOPS SAR images, spanning the period from January 2020 to October 2022 and covering the Kunming urban area, were selected to discern patterns of average subsidence rates and cumulative deformation. The specific parameters of these SAR images are detailed in Table 1. The ALOS World 3D-30m (AW3D30) elevation model (DEM), supplied by JAXA, was employed to adjust for phase discrepancies caused by topographic variations [17]. Precision Orbit Definition (POD) data, provided by the European Space Agency, were used for orbit refinement and phase unwrapping. These images have a temporal baseline extending up to 989 days.

Table 1

Basic Parameters of the SAR Image
Parameter	Value
Band (wavelength in cm)	C(5.6cm)
Imaging mode	Sentinel-1A IW
Incident angle	44.01
Polarization mode	VV
Spatial resolution	5×50
Revisit period/d	12
Number of images	41
Pass direction	Ascending
Time span	From 11 January 2020 to 27 October 2022

3.1 PS-InSAR

The PS-InSAR technique, an advanced remote sensing method for tracking terrestrial surface deformation, is known for its high spatial resolution and temporal stability. Introduced by Ferretti et al. In 2001 [18], it operates by examining scatterers with consistent electromagnetic signatures over time, referred to as permanent scatterers. These are typically associated with built structures and natural geological features such as rocky outcrops. To apply PS-InSAR, a collection of SAR images that survey the same geographic territory across N + 1 time frames is assembled. From this set, one image is designated as the master, against which the remaining N images are aligned and resampled. This meticulous step is crucial to preserving the geophysical signals within the extensive time series datasets. Figure 2 illustrates the spatial and temporal baselines of the study.

Utilizing a reference DEM along with precise satellite orbit data, the interferogram's phase components associated with flat earth and terrain variations are effectively mitigated. Consequently, the interferometric phase φ_x for the ith interference at a specific pixel location x can be computed. i satisfies:

$$\:{\phi\:}_{x,i}={{\phi\:}_{defo}}_{,x,i\:}+{{\phi\:}_{topo}}_{,x,i\:}+{{\phi\:}_{const}}_{,x,i\:}+{{\phi\:}_{slope}}_{,x,i\:}+{{\phi\:}_{atm}}_{,x,i\:}+{{\phi\:}_{noise}}_{,x,i\:}$$

1

The image from October 8, 2021, was selected as the super-master image. Initially, an inversion was carried out to estimate the residual height correction factor and the average deformation rate of the potential PS points using a linear model. The area was divided into sub-regions of 2 m², and reference points were selected based on the amplitude dispersion index. Each sub-region was processed individually, and the results were later combined to form a cohesive outcome. In the subsequent inversion, a linear model was used to determine the atmospheric interference phase component. This atmospheric component was then removed using a temporal high-pass filter (365 days) and a spatial low-pass filter (1200 meters). The refined data were geocoded to the geographic coordinate system, yielding the line-of-sight (LOS) deformation rate and the cumulative deformations over time, as depicted in Fig. 3. These steps are essential for analyzing surface deformations and their causative factors. The LOS deformation was also converted to vertical deformation using the radar incidence angle [19] with the following conversion equation:

$$\:{d}_{v}=\frac{{d}_{Los}}{{cos}\theta\:}$$

2

where d_Los is the line-of-sight deformation result, d_v is the vertical deformation result, and θ is the radar incidence angle.

3.2 SBAS-InSAR

The SBAS-InSAR technique, introduced by Berardino et al. (2002) [20], fundamentally incorporates SAR image data based on spatial and temporal baselines to create a set of interferometric pairs with small baselines. This ensures that the baseline distances within each subset are minimal. The time series of surface deformation for each subset is calculated using the least squares approach, as shown in Fig. 4.

The image from May 5, 2021, was chosen as the reference image (Fig. 5). The first inversion employs a linear model to invert for the residual height correction factor and the initial average deformation rate at points of high coherence, while also integrating small baseline data to derive the surface deformation time series using the SVD method. In the second inversion, the same linear model is used to estimate the atmospheric phase component, which is then removed using a temporal high-pass filter (365 days) and a spatial low-pass filter (1200 meters). The LOS deformation rate and time-series deformation accumulation are subsequently obtained through geocoding. Additionally, Eq. (2) is employed to convert LOS deformation to vertical deformation, based on the radar incidence angle, as illustrated in Fig. 5.

3.3 Integrated PS-InSAR and SBAS-InSAR methods

PS-InSAR is adept at modeling linear deformation, whereas SBAS-InSAR is tailored for nonlinear deformation [21]. It has been argued that high-resolution small baseline approaches are as effective as persistent scatterer approaches, which focus on cells dominated by a single scatterer. However, the PS technique has two key advantages. Firstly, it allows for the creation of all interferograms in reference to a single master image [22], reducing noise contributions from the master image across all interferograms before phase unwrapping. Secondly, it avoids spatial filtering, preventing the amplification of noise contributions from non-dominant scatterers due to resolution degradation. Therefore, these methods are often considered complementary, particularly when one dataset encompasses pixels with specific scattering characteristics. By merging the two techniques, a broader array of scattering signals can be harnessed from the complete dataset, markedly refining errors caused by spatial sampling. Furthermore, the signal-to-noise ratio benefits from the selective backscattering coefficients employed by both methods.

3.4 Long Short Term Memory networks

LSTM is a variant of the Recurrent Neural Network (RNN) tailored to process and model extensive sequence data. LSTMs have an edge over traditional RNNs as they counteract the challenges of vanishing and exploding gradients by incorporating a gating mechanism, which is adept at capturing long-term dependencies [23]. The fundamental components of an LSTM unit include the Memory Cell and three regulatory gates: the Input Gate, Forget Gate, and Output Gate. These elements regulate the flow of information and the update mechanics of the cell state through adjustable weights, thus enabling a more effective handling of input data [24]. At each time step, the LSTM considers both the present input data and the hidden state from the prior step, calculating a fresh hidden state and output that take into account the current input and the memory from the previous step. The gating system allows the LSTM to selectively retain or disregard information and to manage the output of hidden states as well as the generation of outputs with precision. The internal structure is shown in Fig. 6

Applying LSTMs to the analysis of TS-InSAR data can provide substantial predictive insights into the patterns and tendencies of surface subsidence. By delving into historical datasets of surface subsidence, LSTM models are equipped to detect and decipher intricate patterns and correlations within time-series data, which enables them to project future dynamic shifts. Owing to their proficiency in modeling long-term dependencies, LSTMs demonstrate heightened sensitivity in recognizing the nonlinear and extended temporal features of time-series data, outperforming conventional statistical methods and significantly improving predictive accuracy.

Nevertheless, the standard LSTM model encounters several hurdles in its application to the forecasting of surface settlements:
a. Despite being engineered to tackle long-term dependencies in time-series data, LSTMs might fail to capture correlations over prolonged periods if not properly adjusted and calibrated, leading to memory cells that do not effectively maintain pertinent information, thereby diminishing the precision of predictions.
b. In the sphere of surface settlement monitoring, the quality and completeness of observational data are crucial for model training. Real-world monitoring data can be incomplete or contain gaps, which presents a challenge for LSTM models that depend on uninterrupted data. Such data gaps can hinder the model's ability to learn accurate patterns of surface subsidence, detrimentally impacting predictions.
c. Given their extensive parameterization, standard LSTMs are prone to overfitting when trained on complex datasets. The model might learn the noise within the data rather than the underlying signal, which could degrade its predictive performance on new, unobserved monitoring data.
d. A critical component of deploying an LSTM model is selecting appropriate network structure parameters, learning rate, and iteration number.

To surmount these obstacles and fully leverage the capabilities of LSTMs in the prediction of surface subsidence, researchers must implement specific optimization strategies. Such strategies encompass the application of regularization methods to curb overfitting, the use of data interpolation techniques to manage missing information, and meticulous parameter tuning along with network structure refinement to better capture the complexities of the geological environment. By adopting these measures, the LSTM model is anticipated to reinforce its standing as a formidable instrument in the realm of surface subsidence prediction.

3.5 Sparrow Search Algorithm

The Sparrow Search Algorithm (SSA) is influenced by the meticulous observation of natural sparrow flocking behavior, and its distinctive communal behavior offers innovative algorithmic approaches for tackling complex optimization problems. SSA has displayed substantial benefits in numerous domains, such as heightened stability, exceptional global search capabilities, rapid convergence, and remarkable robustness [25]. In the simulation of the algorithm, the position of each sparrow represents a potential solution for optimization, while the sparrow's quest for sustenance is conceptualized as the search mechanism within the problem space.

Sparrow colonies demonstrate three primary foraging behaviors:
a. "Finders" guide the flock towards food sources, providing critical information on the search area and direction.
b. "Followers" rely on the "finders" for sustenance and improve the quality of their solutions by emulating the movements of the "finders."
c. At times, a "discoverer," recognizing that a food source is exhausted or insufficient for the group, opts to abandon it, prompting the colony to search for new sustenance.

To replicate the intricate dynamics of these natural behaviors, the roles of discoverers and followers within sparrow colonies are fluid, alternating through a specific mechanism, with discoverers generally comprising 10 to 20 percent of the colony. The discoverer steers the followers with its advanced search abilities and continually refines its position based on memory to enhance the quality of the food source. Meanwhile, followers enhance their solutions by imitating the discoverer, aiming to achieve or surpass the discoverer's level of fitness.

3.6 Predictive Modelling

Predicting surface subsidence is a sophisticated task steeped in time series analysis. Traditional LSTM models may face difficulties, such as being trapped by local optima and experiencing slower convergence rates when addressing such complex problems. To surmount these challenges, the training performance of LSTM can be improved by integrating the SSA with LSTM models, thus harnessing SSA's superior global search ability. This integration helps the LSTM model to delve deeper and better comprehend the long-term dependencies within the data, subsequently enhancing the model's accuracy and adaptability in forecasting the future dynamics of surface subsidence. A schematic of the prediction model is shown in Fig. 7

The combined technique not only strengthens LSTM's global search performance but also refines the model's ability to adaptively tune and model long-term dependencies, especially when dealing with intricate time-series challenges in urban surface settlement monitoring and prediction. By applying SSA to optimize LSTM's parameters, we anticipate a marked enhancement in the model's convergence speed and generalization capacity[26].

Utilizing surface settlement monitoring data gathered by TS-InSAR from six different sub-study areas, this research has constructed a stacked prediction model for accurately projecting time-series settlement data at each monitoring point. As a result, a point-array surface settlement prediction map has been generated. The predictive accuracy of the model will be quantitatively evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and the Coefficient of Determination (R-squared, R²). These metrics will enable a comparative analysis of the SSA-LSTM model's effectiveness against that of a traditional LSTM model in the prediction of surface settlement.

3.7 Subsidence Zone

According to the distribution law of historical settlement area in Kunming urban area, the study area will be divided into 6 sub-study areas respectively:A,B,C,D,E,F. The results of the study will be analysed sets based on these 6 sub-study areas, and the results of the zoning are shown in Fig. 1.

4.1 Cross-validation of PS, SBAS methods

Table 2

Correlation between PS and SBAS
Settlement Area	Linear regression equation	R2	Significant impact
A	y = 0.7585x-1.5397	0.94	**
B	y = 0.5245x-0.0798	0.88	**
C	y = 0.8686x-0.9393	0.98	**
D	y = 1.0332x-1.4303	0.92	**
E	y = 2.0254x-2.4997	0.88	**
F	y = 1.5544x-5.8799	0.98	**
Note: Significance influence levels are represented by the number of s in the analysis: ρ < 0.01 is denoted as * (very significant).

In this study, mean values across 41 time points were systematically compared using PS-InSAR and SBAS-InSAR techniques at all monitoring locations within each subsidence zone to achieve cross-validation regarding time-series alterations. Correlation analyses of time series monitoring outcomes employing PS-InSAR and SBAS-InSAR methodologies for each subsidence zone indicate extremely high statistical significance in the cross-validation, with correlation coefficients exceeding 0.88 across all zones. Zones C and F, in particular, display a correlation of 0.98, with P values below 0.01. These findings are illustrated in Fig. 8 and Table 2, confirming the effectiveness and reliability of both the PS and SBAS techniques in the time-series monitoring of surface settlement.

Specifically, PS-InSAR is more suitable for stable scattering points and densely built-up areas, while SBAS-InSAR excels at detecting low-frequency and non-linear deformations. By leveraging the strengths of both methods, this research distinguished between Settlement B and Settlement E as case studies to illustrate the outcomes of the SBAS method. The PS method outcomes were applied in other areas to demonstrate the cumulative surface settlement from January 2020 to October 2022 in the Kunming urban sector.

The analysis of the spatial distributions of monitoring results from PS-InSAR and SBAS-InSAR reveal that the size and contours of sedimentation funnels yielded by both methods demonstrate high congruity. Nevertheless, the spatial correlation of the forementioned funnels in Sedimentation C zone is less than the correlations in other zones, as depicted in Fig. 9.

4.2 Monitoring results

4.2.1 Sedimentation area distribution results

Analysis of recent monitoring outcomes in Kunming's urban area indicates that the spatial extent of subsidence funnels has remained relatively stable. Building upon previous research and current TS-InSAR monitoring results, the urban area of Kunming with significant surface settlements is divided into six distinct sub-areas:

Subsidence Zone A encompasses the Rongchuang-Xinhe community-Wangjiadui region in Xishan District.

Subsidence Zone B includes recently developed settlement funnels over the past three years, spanning Metro Line 5, the South Connector Highway, Qianwei Road, Hairun Road, Ziyuan Road, Guangfu Road, and Haize Road.

Subsidence Zone C, the largest settlement distribution area, comprises several parts of Guandu District, including Liujia Village, Caojiaqiao, Niuqiao Village, Dianchi Convention and Exhibition Centre, Sijia Community, Luo Nga Community, Wangjia Village, and Longma Community.

Subsidence Zone D, also located in Guandu District, covers the same regions as Zone C.

Subsidence Zone E is situated near communities along Luoyang Street in Chenggong District.

Subsidence Zone F incorporates areas around Zhonghe Village, Luojia Village, and East Huancheng Road in Danyu Street, Chenggong District.

It is important to note that subsidence zones A, C, and F are adjacent to Dianchi Lake, while zones B, D, and E are situated further inland.

The presentation of monitoring results enabled the identification of areas in Kunming with extensive coverage of subsidence funnels, particularly Subsidence Zones A and C. The Subsidence B area exhibited significant subsidence in the latest monitoring cycle, forming a contiguous subsidence zone with the Subsidence A and C areas. Detailed information is provided in Fig. 10.

4.2.2 Settling rate results

The dynamic development of surface subsidence is influenced by environmental conditions and human activities. The subsidence rate analysis suggests a spatially heterogeneous process, with some areas showing rapid rates and localized stress concentrations, while others are relatively stable.

The distribution pattern calls for a differential assessment of potential risks, especially the risk of geological hazards in Area D, due to the excessively large funnel openings. The funnel pattern and expansion trend in Zone E highlight the need for careful urban construction and land use management. This study indicates that the spatial distribution of subsidence rates forms a funnel-shaped pattern, with concerns that the openings of the funnels in Zones A, B, and C are diverging and becoming more pronounced, suggesting a multi-funnel phenomenon. In contrast, Zone D is at risk of geological hazards due to large funnel openings. The subsidence in Zone E is notable for its distinct funnel shape and expansion trend. Meanwhile, the subsidence rate in Zone F is lower than in other zones and exhibits a trend toward stabilization, suggesting that subsidence may be nearing an end. For further details, refer to Fig. 11.

Table 3

Time-Series In-SAR Monitoring Results
Settlement Area	Cumulative Settlement (mm)	Average Settlement of Settlement Area (mm)	Maximum Settlement Rate (mm·a^− 1)	Settlement Area (km²)
A	0 ~ 99.4	19.0	47.564	19.104
B	0 ~ 83.9	11.6	38.829	42.554
C	0 ~ 105.5	16.4	50.468	49.334
D	0 ~ 67.7	15.1	25.086	13.485
E	0 ~ 67	8.3	28.184	6.796
F	0 ~ 86.9	15.9	26.804	3.121

Table 3 details the TS-InSAR monitoring results from January 2020 to October 2022, providing comprehensive statistics on cumulative settlement, average settlement, maximum settlement rate, and the spatial extent of the subsidence zones in the Kunming urban area. The data reveal a trend of intensifying surface subsidence, with subsidence zones A and C being most affected, which presents considerable challenges to urban planning and infrastructure development. In these zones, the highest cumulative subsidence reaches 105.5mm and 99.4 mm respectively, and the average cumulative subsidence in these areas is also significant, at 19.0 mm and 16.4 mm respectively. The maximum rates of subsidence in zones A, B, and C are 47.564 mm·a^− 1, 38.829 mm·a-¹, and 50.468 mm·a^− 1, respectively, which are concentrated at the outlets of smaller subsidence funnels. Despite representing a smaller fraction of the total subsidence zone, these high-velocity subsidence areas necessitate close and continuous monitoring. Meanwhile, the maximum subsidence rates in zones D, E, and F are 25.086 mm·a^− 1, 28.184 mm·a^− 1, and 26.804 mm·a^− 1, respectively; these rates, while lower, affect larger areas and thus their cumulative effects over time also require careful consideration.

4.3 SSA-LSTM urban ground settlement prediction model

The problem of single-point, multi-period subsidence is inherently a time series challenge, as surface subsidence often exhibits a funnel-shaped progression. Monitoring results, derived from TS-InSAR evaluations, give rise to an array of surface time series data points. This type of problem, featuring multiple inputs and attributes, is well-suited to the LSTM network approach. In this study, a neural network training scheme spans 41 time periods, with early time steps (1 through x periods) used for training and subsequent ones (the x + 1 periods) for validation, progressively extending the training window further into the sequence. At each time step, the LSTM computes the hidden state and memory unit value, taking into account both the current input and the prior time step's information, thereby dynamically refining the learning process.

Table 4

Accuracy assessment of the SSA-LSTM prediction model
Settlement Area		RMSE	MAE	R2
Zone A	SSA-LSTM	3.5095	2.7904	0.93656
Zone A	LSTM	3.6497	2.9654	0.90658
Zone B	SSA-LSTM	2.7673	1.9183	0.93645
Zone B	LSTM	2.7912	2.1459	0.89663
Zone C	SSA-LSTM	2.9101	2.1245	0.93999
Zone C	LSTM	3.3663	2.3681	0.92371
Zone D	SSA-LSTM	2.9599	2.1528	0.93743
Zone D	LSTM	3.1022	2.2583	0.91951
Zone E	SSA-LSTM	2.6824	1.7587	0.83776
Zone E	LSTM	2.7466	1.9611	0.80341
Zone F	SSA-LSTM	3.3395	2.5678	0.96067
Zone F	LSTM	3.3427	2.6021	0.88364

The SSA-LSTM and LSTM prediction models were applied to forecast TS-InSAR monitoring results for the six subsidence zones under study. Correlation analyses were conducted between the input and output sequences. The models' performance for each subsidence zone was evaluated using RMSE, MAE, and R² scores. The results show that after the optimisation of SSA, the performance and robustness of LSTM for surface subsidence prediction are improved to some extent. Correlation coefficients for subsidence zones A to D all exceeded 0.93, with Zone F achieving the highest correlation of 0.96067, while Zone E recorded the lowest at 0.83776, as further detailed in Table 4. These results underline the SSA-LSTM's effectiveness as an analytical tool for time series examination of geophysical data.

Discrepancies between predicted and actual values are illustrated in Fig. 12, which displays the distribution of error intervals across the six subsidence zones. The majority of errors cluster within the 0–5 mm range and diminish progressively beyond this point, forming a bar-shaped distribution emanating from the center. Settlement Zone E exhibits a lower concentration of errors within the 5 mm threshold and a more widespread error distribution, leading to reduced prediction accuracy for this particular zone.

Input data from 41 monitoring periods were utilized to predict future surface settlements over a 24-day cycle, culminating with the last recorded monitoring on October 22, 2022. To mitigate the impact of cumulative errors on model precision, predictions were made for the subsequent 10 periods. These predictions encompass cumulative surface settlements from January 2020 to August 2023. Initially, a graphical depiction of the time series data for surface subsidence in Kunming's urban area is presented, alongside the SSA-LSTM model's predictions. As shown in Fig. 13, time is represented on the horizontal axis and the magnitude of surface subsidence on the vertical axis. The observed data, indicated by a solid blue line, reveals distinct patterns and trends in surface subsidence. In contrast, predictions from the SSA-LSTM model, marked by a dashed red line, closely mimic these patterns, showing a high degree of congruence with the actual data.

Using the monitoring data, the SSA-LSTM prediction model forecasts the trend of surface subsidence, taking into account that the subsidence rate at each stress point within the subsidence funnel varies. The overall prediction results ( Fig. 14) suggest that the bottom of the subsidence funnel is gradually spreading and becoming more defined. Moreover, portions of the subsidence funnel in Areas B and F exhibit larger subsidence amplitudes and faster rates. When predicting future surface subsidence, it is crucial to pay close attention to areas prone to the dispersal and refinement of the subsidence funnel, as significant surface subsidence may occur in these zones.

5.1 Analysis of Rainfall Variability and Time-Series Sedimentation Variability

There is a clear correlation between groundwater extraction and surface subsidence. When groundwater is removed, the pore pressure initially present in the groundwater system decreases, leading to a reduction in the volume of pore water and a reconfiguration of the subsurface sediments, which results in ground subsidence [27][28]. This subsidence is usually permanent, and even if the water table is later restored, the voids created by the subsidence are unlikely to "reinflate" to their original volume upon rehydration. This is particularly the case if the subsurface strata are composed of compressible materials such as clay and silt. Seasonal precipitation can influence changes in groundwater levels.

Data on inter-monthly precipitation in Kunming were collected over the past three years to examine the impact of seasonal rainfall on surface settlement in the plateau basin region. Located in the central part of Yunnan Province, China, Kunming's high elevation (approximately 1900 meters) and its subtropical plateau monsoon climate contribute to significant monsoon-driven seasonal precipitation patterns. The dry season extends from November to April, with relatively low rainfall, while the wet season spans from May to October, bringing substantial precipitation. Summer alone accounts for 60% of the annual rainfall. The temporal deformation patterns in the six subsidence zones reveal that surface subsidence fluctuates, with episodes of surface uplift occurring intermittently, although the general trend is one of subsidence, with uplift and subsidence alternating.

As depicted in Fig. 15, the dotted line graph represents the monthly precipitation from 2020 to 2022, while the curved graph illustrates the time series of surface deformation in subsidence zones A through F. In June, when annual precipitation peaks, normal seasonal or multi-year average groundwater level fluctuations often cause minor subsidence and uplift. However, these surface changes are typically reversible, and the soil particles can return to a more stable state when the water table rises, due to the reestablishment of water lubrication and interparticle pressure.

5.2 Analysis of changes in the settlement zone

An analysis of the time series of subsidence across eight distinct periods from 2020 to 2022 reveals widespread surface subsidence in the Kunming urban area, particularly on the east side of Dianchi Lake. This area has consistently shown stable subsidence throughout the study period. Over time, the subsidence has expanded irregularly, radiating outward from the center of the subsidence funnel across the six most affected zones. The diffusion pattern is shaped by local factors such as surface moisture, built structures, and geological formations.

The six subsidence areas (A, B, C, D, E, and F) have displayed different subsidence rates and diffusion behaviors over time. Areas A, B, and C tend to diffuse outward, serving as surface sources of subsidence. In contrast, Areas D, E, and F do not exhibit significant vertical changes but display an increased rate and speed of settlement, acting as point sources of subsidence. The specifics of this are elucidated in Fig. 16.

In the northeastern part of the Dianchi Lake area, the impact on the settlement can be attributed to both the soft geological nature of the wetlands adjacent to Dianchi and the development and construction of buildings. Moving from the banks of Dianchi, particularly heading from wetlands towards the land, surface settlement gradually decreases due to the influence of Dianchi and progressively increases due to urban development activities.

5.3 Trends in sedimentation funnels

The cumulative data on surface settlement in the urban area of Kunming from January 2020 to October 2023 were compiled by integrating the monitoring results with projections from the predictive model. Settlement data from February 2020 and November 2021, along with predictions for August 2023, were selected for analysis. January 2020 served as the baseline with zero settlement to analyze the trend of surface settlement in the urban area of Kunming over the four-year period.

As illustrated in Fig. 17, the top pair of 3D graphs display the monitoring results, while the lower 3D graph presents the predictive outcomes. Settlement zones labeled 'A' and 'D' show a general trend of settlement, 'B' depicts a cylindrical settlement funnel, while 'C' represents a conical settlement funnel. 'E' shows slight irregular surface deformation, with the most significant settlement center located at the margins of the settlement area. 'F' is identified as the smallest settlement area within the studied region.

In our study, we monitored and predicted surface settlement events in the urban expanse of Kunming, aiming to discern the patterns of surface settlement and the trends of their causative changes. The study suggests that surface settlement poses a potential hazard to the structural integrity of buildings, particularly when there is a significant difference in the magnitude of local settlement deformation compared to nearby areas. This can lead to uneven stress distributions on the foundations of buildings, potentially causing structural cracking, collapse, and other safety issues.

Analyses indicate that small, initially dispersed settlement funnels are highly susceptible to damaging surface structures. Due to the relatively minor deformation initially, the buildings may exhibit only insignificant cracking, typically within the buildings' structural capacity. However, as these dispersed settlement funnels merge to form larger subsidence areas, they might not lead to substantial differential settlement impacting individual structures, but they pose a serious risk to territorial integrity and the stability of transportation infrastructure. Although road systems are designed to tolerate some degree of deformation, the potential for significant infrastructural damage from expansive areas of subsidence is real. As the phenomenon advances, extensive subsidence areas reemerge with dispersive activities, accompanied by the emergence of additional smaller subsidence funnels. This indicates a notable variance in subsidence magnitudes and results in more pronounced impacts on buildings. Consequently, surface subsidence exhibits periodic fluctuations, manifesting as a spiral dynamic process of alternating uplift and subsidence. This cyclical trend is also evident in the spatial distribution pattern: it begins with minor subsidence events, progresses to the development of large-scale subsidence zones, and ultimately disperses into smaller subsidence occurrences, a sequence that may repeat.

Through the preceding analyses, we have not only identified the cyclical pattern of urban surface settlement but also gained deeper insight into the impacts that may be triggered by the settlement phenomenon. These insights are essential for predicting future settlement trends and for the development of preventive measures and mitigation strategies. These can serve as a scientific basis for urban planning and infrastructure development, thus reducing the risk of damage to buildings and enhancing urban safety.

5.4 Kunming Sedimentation Zones and Wetland Distribution Analysis.

As illustrated in Fig. 18, there are wetland parks located primarily alongside the Dianchi Pond at the sedimentation centres of Sedimentation Areas A, C, and F. Dry Gouwei Wetland Park, Yungang River Lakeside Ecological Wetland, Caohai Tunnel Park, and Yuchang Wetland are found in Subsidence Area A. Daguan Park and Lakeside Ecological Wetland Park (Lantai House section), meanwhile, are situated in Subsidence Area B. Haihong Wetland Park and Xinghai Peninsula Lakeside Ecological Wetland Park are in Subsidence Area C, with Layu River Wetland Park sited in Subsidence Area F.

The establishment of wetland parks, given the inherently soft terrain characteristic of wetlands, may significantly influence the hydrological conditions in these areas. The construction within the wetland parks, coupled with the wetlands' pliable substrate, can lead to surface subsidence that might alter the hydrology of the area, potentially affecting the stability of the wetland ecosystem. For example, surface subsidence can change the soil properties within the air pockets, impacting the balance of groundwater recharge, runoff, and discharge, and thus the overall hydrological equilibrium.

When the water table falls, the river's base flow can decrease, and soil water content may drop below the level necessary to sustain the field capacity, adversely affecting the growth of native vegetation. Conversely, when the water table rises, evaporation can leave deposits of inorganic salts on the surface, potentially increasing soil salinity and leading to land salinization. Additionally, the root systems of plants that are intolerant to waterlogged conditions may die due to a lack of oxygen [29][30].

The sedimentation, pollution, and degradation of wetland ecosystems could collectively contribute to the decline of water quality in Dianchi [31]. Ground subsidence may result in inadequate flood discharge capacity and the accumulation of pollutants in the wetlands [32][33]. Consequently, these pollutants might seep into Dianchi because of changes in runoff patterns caused by the subsidence, impacting the water quality and affecting the lives of urban and rural residents who rely on Dianchi as a water source.

5.5 Traffic Roads and Surface Settlement: A Relationship Analysis

Urban roadways have a considerable impact on the occurrence of surface settlement [34][35]. The construction and subsequent operation of Kunming Metro Line 5, while enhancing urban transportation convenience, have likely affected the surface stability of the urban area, according to monitoring data from Subsidence Area B. Figure 19 shows the detailed distribution of traffic roads within Subsidence Area B. Notably, in major arteries and densely populated areas of Subsidence Area B—such as the G56 Hangrui Expressway, Guangfu Road, and along Metro Line 5—small subsidence depressions appear to align with the paths of roads and Metro lines. These subsidence patterns may be closely linked to changes in surface water content caused by the underground construction activities, which, in turn, affect the soil structure. Such changes have led to compromised road leveling and stability, inflicting damage to utilities, particularly underground pipelines and drainage systems. Over time, this could necessitate increased maintenance, raise urban repair costs, and pose safety hazards. The risks of surface settlement related to the construction and operation of Kunming Metro Line 5 also threaten the structural integrity of buildings, with metro stations and surrounding public facilities facing heightened risks. This situation not only jeopardizes structural integrity but also directly affects the safety and assets of residents (Fig. 20). As such, specific measures must be implemented to address these structural risks, including regular maintenance checks, prompt repairs, and, where necessary, structural reinforcement.

In the near term, the surface settlements caused by the construction and operation of metro lines demand improved monitoring and management to preclude broader impacts. For the future, the potential consequences on urban ecology, transportation safety, building integrity, and economic activities must be thoroughly evaluated. The development of robust preventive and response strategies is crucial, such as adjustments to metro line designs or operational methods, improved foundation treatments, and soil consolidation techniques. Crucially, a detailed investigation into how metro construction and operation influence urban surface settlement will aid in identifying and addressing related issues. This requires cross-disciplinary collaboration among geology, engineering, and urban planning experts. By establishing proper planning and construction standards, the risks of subsidence can be minimized, thus preserving the city's livability and sustainability.

5.6 SSA-LSTM evaluation analysis

Through conducting a comparative analysis of the SSA alongside other meta-heuristic optimization algorithms, we can attain a deeper understanding of the algorithm's strengths and weaknesses in tackling real-world problems [36]. Initially, we evaluate the performance of the SSA against a range of standard optimization test functions. Compared to classical metaheuristic methods such as Particle Swarm Optimization (PSO), Genetic Algorithms (GA), and Ant Colony Optimization (ACO), SSA shows substantial advantages in terms of search space exploitation and convergence rates. This reflects SSA's capacity to effectively balance exploration and exploitation by emulating the social behaviors observed in sparrow populations, including foraging, vigilance, and flight. Nevertheless, SSA may become trapped in local optima or show sensitivity to parameter settings in certain complex situations, indicating areas where enhancement is possible. Regarding real-world applications, recent research illustrates SSA's efficacy across various sectors, including engineering optimization, machine learning model parameter tuning, and supply chain management. Especially noteworthy is SSA's proficiency in addressing both continuous and discrete optimization problems through straightforward encoding methods within practical contexts. However, when facing complex issues in intricate domains, relying solely on SSA may be inadequate, thus necessitating further research to hone and broaden the algorithm's capabilities.

Considering the limitations of the SSA as discussed, future research could investigate several paths: firstly, to improve SSA's local search capacity, integrating strategies to navigate past local optima—akin to those used in simulated annealing or taboo search—appears promising. A second avenue is to intensify the focus on algorithm parameter tuning, seeking optimal parameter configurations for diverse problems or adopting adaptive parameter adjustment strategies to enhance the algorithm's adaptability. Thirdly, combining SSA with other optimization algorithms presents a chance to improve optimization outcomes while preserving the algorithm's simplicity and ease of implementation. Despite the challenges associated with local optima and parameter tuning, this opens up new prospects for addressing real-world issues. By refining SSA and integrating various search strategies, there is potential for resolving a broad spectrum of real-world optimization problems with increased efficiency in future endeavors.

The current research successfully integrated the SSA with the LSTM network, utilizing TS-InSAR techniques to monitor and predict surface settlements in the metropolitan area of Kunming. After a thorough analysis of rainfall variation, sequential changes in settlements, time-series variations within the settlement region, the dynamics between traffic routes and surface settlements, and predictive analytics, the following conclusions can be drawn:

(1) The study verifies the cyclical nature of surface subsidence in the urban area of Kunming and highlights potential issues such as damage to transportation infrastructure, deterioration of buildings, and degradation of wetland ecosystems due to this phenomenon.

(2) There is a significant relationship between rainfall variability and surface subsidence. An analysis of monthly precipitation data in Kunming over the past three years, along with the temporal evolution of subsidence zones, has revealed a pattern of alternating surface uplift and subsidence.

(3) Through time-series analysis, it has been observed that certain settlement zones within Kunming's urban limits exhibit different rates and patterns of settlement, which could affect infrastructure and buildings to varying degrees.

(4) The study also examines the interactions between transportation networks and surface settlement, identifying potential risks related to increased surface loads from subway construction and urban infrastructure development.

(5) The SSA-LSTM model's predictive analysis provides a scientific basis for urban planning and infrastructure development, essential for accurately forecasting future settlement trends, minimizing the risk of structural damage, and enhancing urban safety.

In summary, this study sheds light on the regular patterns of the subsidence phenomenon and its subsequent effects through the observation and prediction of surface settlements within Kunming's urban limits. These insights are of considerable importance for improving urban construction and governance and for the prevention and mitigation of issues caused by subsidence. Nonetheless, further refinement of the research methodology is needed, along with an expansion of its application scope, and focused efforts to improve the SSA are imperative for more effective monitoring and prediction of surface subsidence.

Author Contribution

All authors contributed to the manuscript. B.L. and Y.P. constructed the TS-InSAR model, while B.L. and S.T. worked on the SSA-LSTM model. J.L. developed the overall idea of the paper. B.L. wrote the manuscript with valuable input from all authors. S.T. reviewed and evaluated the manuscript. B.W. collected the Sentinel-1 data and the rainfall data. Y.P. provided important insights on soil science and hydrology.

Acknowledgement

We would like to express our gratitude to the European Space Agency (ESA) for providing us with the Sentinel-1A images and satellite precise orbit data. We also extend our thanks to the Japan Aerospace Exploration Agency (JAXA) for supplying the AW3D30 data. Additionally, we acknowledge the National Meteorological Science Data Centre of China for generously sharing the rainfall data. We are grateful to the anonymous reviewers of Applied Science for their invaluable feedback, comprehensive review, and insightful comments and suggestions.

Data Availability

Data incorporated in this research is available free through these webpages: Sentinel-1 (https://dataspace.copernicus.eu/), AW3D30 DEM(http://www.eorc.jaxa.jp/ALOS/en/aw3d30/index.htm).

Wang, R.; Wu, K.; He, Q.; He, Y.; Gu, Y.; Wu, S. A novel method of monitoring surface subsidence law based on probability integral model combined with active and passive remote sensing data. Remote Sens. 2022, 14, 299. https://doi.org/10.3390/rs14020299
Hu, R.; Yue, Z.; Wang, L.U.; Wang, S. Review on current status and challenging issues of land subsidence in China. Eng. Geol. 2004, 76, 65–77. https://doi.org/10.1016/j.enggeo.2004.06.006
Ma, T.; Du, Y.; Ma, R.; Xiao, C.; Liu, Y. Water-rock interactions and related eco-environmental effects in typical land subsidence zones of China. Hydrogeol. J. 2018, 26, 1339–1349. https://doi.org/10.1007/s1004 0-017-1708-8
Aobpaet, A.; Cuenca, M.C.; Hooper, A.; Trisirisatayawong, I. InSAR time-series analysis of land subsidence in Bangkok, Thailand. Int. J. Remote Sens. 2013, 34, 2969–2982. https://doi.org/10.1080/ 01431161.2012.756596
Qu, F.; Lu, Z.; Zhang, Q.; Bawden, G.W.; Kim, J.W.; Zhao, C.; Qu, W. Mapping ground deformation over Houston–Galveston, Texas using multi-temporal InSAR. Remote Sens. Environ. 2015, 169, 290–306. https://doi.org/10.1016/j.rse.2015.08.027
Yao, J.; Yao, X.; Liu, X. Landslide detection and mapping based on SBAS-InSAR and PS-InSAR: A case study in Gongjue County, Tibet, China. Remote Sens. 2022, 14, 4728. https://doi.org/10.3390/rs14194728
Dehghan-Soraki, Y.; Sharifikia, M.; Sahebi, M.R. A comprehensive interferometric process for monitoring land deformation using ASAR and PALSAR satellite interferometric data. Gisci. Remote Sens. 2015, 52, 58–77. https://doi.org/10.1080/15481603.2014.989774
Galloway, D.L.; Burbey, T.J. Regional land subsidence accompanying groundwater extraction. Hydrogeol. J. 2011, 19, 1459–1486. https://doi.org/10.1007/s10040-011-0775-5
Guo, S.; Zhang, W.; Kang, W.; Zhang, T.; Li, Y. The Study on Land Subsidence in Kunming by Integrating PS, SBAS and DS InSAR. Remote Sens. Technol. Appl. 2022, 37, 460–473. https://doi.org/ 10.11873/j.issn.1004-0323.2022.2.0460
Gunn, M.J. The prediction of surface settlement profiles due to tunnelling. In Predictive soil mechanics: Proceedings of the Wroth Memorial Symposium held at St Catherine's College, Oxford, Thomas Telford Publishing, 27-29 July 1992; pp. 304–316. https://doi.org/10.1680/psm.19164.0020
Mahmoodzadeh, A.; Mohammadi, M.; Daraei, A.; Ali, H.F.H.; Al-Salihi, N.K.; Omer, R.M.D. Forecasting maximum surface settlement caused by urban tunneling. Autom. Constr. 2020, 120, 103375. https://doi.org/10.1016/j.autcon.2020.103375
Hu, Y.; Gu, C.; Meng, Z.; Shao, C.; Min, Z. Prediction for the settlement of concrete face rockfill dams using optimized LSTM model via correlated monitoring data. Water 2022, 14, 2157. https://doi.org/ 10.3390/w14142157
Li, C.; Li, J.; Shi, Z.; Li, L.; Li, M.; Jin, D.; Dong, G. Prediction of surface settlement induced by large-diameter shield tunneling based on machine-learning algorithms. Geofluids 2022, 2022, 1–13. https:// doi.org/10.1155/2022/4174768
Yang, S.; Jin, A.; Nie, W.; Liu, C.; Li, Y. Research on SSA-LSTM-based slope monitoring and early warning model. Sustainability 2022, 14, 10246. https://doi.org/10.3390/su141610246
Yu, L.; Oldfield, F.; Yushu, W.; Sufu, Z.; Jiayi, X. Paleoenvironmental implications of magnetic measurements on sediment core from Kunming Basin, Southwest China. J. Paleolimnol. 1990, 3, 95–111. https://doi.org/10.1007/BF00414065
Clerici, N.; Valbuena Calderón, C.A.; Posada, J.M. Fusion of Sentinel-1A and Sentinel-2A data for land cover mapping: a case study in the lower Magdalena region, Colombia. J. Maps 2017, 13, 718–726. https://doi.org/10.1080/17445647.2017.1372316
Grohmann, C.H. Evaluation of TanDEM-X DEMs on selected Brazilian sites: Comparison with SRTM, ASTER GDEM and ALOS AW3D30. Remote Sens. Environm. 2018, 212, 121–133. https://doi.org/10.1016/ j.rse.2018.04.043
Ferretti, A.; Prati, C.; Rocca, F. Permanent scatterers in SAR interferometry. IEEE Trans. Geosci. Remote Sens. 2001, 39, 8–20. https://doi.org/10.1109/36.898661
He, Y.; Wang, W.; Yan, H.; Zhang, L.; Chen, Y.; Yang, S. Characteristics of surface deformation in Lanzhou with Sentinel-1A TOPS. Geosciences 2020, 10, 99. https://doi.org/10.3390/geosciences10030099
Berardino, P.; Fornaro, G.; Lanari, R.; Sansosti, E. A new algorithm for surface deformation monitoring based on small baseline differential SAR interferograms. IEEE Trans. Geosci. Remote Sens. 2002, 40, 2375–2383. https://doi.org/10.1109/TGRS.2002.803792
Wang, Q.; Gao, Y.; Gong, T.; Liu, T.; Sui, Z.; Fan, J.; Wang, Z. Dam Surface Deformation Monitoring and Analysis Based on PS-InSAR Technology: A Case Study of Xiaolangdi Reservoir Dam in China. Water 2023, 15, 3298. https://doi.org/10.3390/w15183298
An, B.; Jiang, Y.; Wang, C.; Shen, P.; Song, T.; Hu, C.; Liu, K. Ground infrastructure monitoring in coastal areas using time-series inSAR technology: the case study of Pudong International Airport, Shanghai. Int. J. Digit. Earth 2023, 16, 355–374. https://doi.org/10.1080/17538947.2023.2171144
Hochreiter, S.; Schmidhuber, J. Long short-term memory. Neural Comput. 1997, 9, 1735–1780. https://doi.org/10.1162/neco.1997.9.8.1735
Greff, K.; Srivastava, R.K.; Koutník, J.; Steunebrink, B.R.; Schmidhuber, J.R. LSTM: A search space odyssey. IEEE Trans. Neural Netw. Learn. Syst. 2016, 28, 2222–2232. https://doi.org/10.1109/tnnls.59687
Tuerxun, W.; Chang, X.; Hongyu, G.; Zhijie, J.; Huajian, Z. Fault diagnosis of wind turbines based on a support vector machine optimized by the sparrow search algorithm. IEEE Access 2021, 9, 69307–69315. https://doi.org/10.1109/ACCESS.2021.3075547
Madiniyeti, J.; Chao, Y.; Li, T.; Qi, H.; Wang, F. Concrete Dam Deformation Prediction Model Research Based on SSA–LSTM. Appl. Sci. 2023, 13, 7375. https://doi.org/10.3390/app13137375
Budhu, M.; Adiyaman, I.B. Mechanics of land subsidence due to groundwater pumping. Int. J. Numer. Anal. Methods Geomech. 2010, 34, 1459–1478. https://doi.org/10.1002/nag.863
Erban, L.E.; Gorelick, S.M.; Zebker, H.A. Groundwater extraction, land subsidence, and sea-level rise in the Mekong Delta, Vietnam. Environ. Res. Lett. 2014, 9, 084010. https://doi.org/10.1088/1748-9326/9/8/ 084010
Arnold, J.G.; Muttiah, R.S.; Srinivasan, R.; Allen, P.M. Regional estimation of base flow and groundwater recharge in the Upper Mississippi river basin. J. Hydrol. 2000, 227, 21–40. https://doi.org/ 10.1016/S0022-1694(99)00139-0
Henriksen, H.J.; Troldborg, L.; Højberg, A.L.; Refsgaard, J.C. Assessment of exploitable groundwater resources of Denmark by use of ensemble resource indicators and a numerical groundwater–surface water model. J. Hydrol. 2008, 348, 224–240. https://doi.org/10.1016/j.jhydrol.2007.09.056
O'Geen, A.; Budd, R.; Gan, J.; Maynard, J.; Parikh, S.; Dahlgren, R. Mitigating nonpoint source pollution in agriculture with constructed and restored wetlands. Adv. Agron. 2010, 108, 1–76. https://doi.org/10.1016/S0065-2113(10)08001-6
Lee, S.; Dunn, R.; Young, R.; Connolly, R.; Dale, P.; Dehayr, R.; Lemckert, C.; Mckinnon, S.; Powell, B.; Teasdale, P. Impact of urbanization on coastal wetland structure and function. Austral Ecol. 2006, 31, 149–163. https://doi.org/10.1111/j.1442-9993.2006.01581.x
Shutes, R.B.E. Artificial wetlands and water quality improvement. Environ. Int. 2001, 26, 441–447. https://doi.org/10.1016/S0160-4120(01)00025-3
Borzi, G.; Santucci, L.; Tanjal, C.; Carol, E. Assessment of surface runoff conditioned by road works and urban settlements in large plain basins. Nat. Hazards 2020, 101, 981–994. https://doi.org/10.1007/ s11069-020-03905-0
Yang, R.; Xu, Q.; Xu, X.; Chen, Y. Rural settlement spatial patterns and effects: Road traffic accessibility and geographic factors in Guangdong Province, China. J. Geogr. Sci. 2019, 29, 213–230. https://doi.org/ 10.1007/s11442-019-1593-2
Hasanpour, R.; Chakeri, H.; Ozcelik, Y.; Denek, H. Evaluation of surface settlements in the Istanbul metro in terms of analytical, numerical and direct measurements. Bull. Eng. Geol. Environ. 2012, 71, 499–510. https://doi.org/10.1007/s10064-012-0428-5

No competing interests reported.

Monitoring and Prediction of Surface Subsidence by Combining SSA-LSTM and TS-InSAR - A Case Study of Kunming Urban Area

Status:

Version 1

Abstract

Figures

1. Introduction

2. Study area and data