Identification of Potential Landslide in Jianzha Counctry Based on InSAR and Deep Learning

Identification of Potential Landslide in Jianzha Counctry Based on InSAR and Deep Learning | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Identification of Potential Landslide in Jianzha Counctry Based on InSAR and Deep Learning Xianwu Yang, Dannuo Chen, Yihang Dong, Yamei Xue, Kexin Qin This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4642799/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 10 You are reading this latest preprint version Abstract Landslide disasters have characteristics of frequent occurrence, widespread impact, and high destructiveness, posing serious threats to human lives, property, and the ecological environment. Timely and accurate early identification of landslides remains an urgent issue within the disaster prevention field. This study focuses on Jianzha County, Qinghai Province, integrating PS-InSAR、SBAS-InSAR and optical remote sensing techniques to delineate potential landslide-prone areas. Utilizing Google Earth imagery and existing landslide datasets, potential landslide points were identified through a deep learning model. The results indicate that: (1) In Jianzha County, the variation trend of the average surface velocity monitored by PS-InSAR and SBAS-InSAR technology is consistent, and the deformation monitoring results are reliable. (2) Utilizing the deep learning model, 56 potential landslide points were identified, comprising 39 high-risk points and 17 medium-risk points. By integrating the spatial distribution data of historical geological disaster points, it was found that 10 out of 13 previously occurred landslide disaster points were located at the identified high-risk landslide points, achieving a detection accuracy of 76.92%. (3) The spatial distribution of landslide points exhibits clustering, with slopes ranging from 10–40°, elevations between 15–30 m, and slope orientations predominantly towards the northeast. (4) Landslide formation is correlated with seasonal precipitation concentrations and temperature fluctuations. This method can provide a crucial basis for large-scale surface deformation monitoring and early identification of landslide risks. Earth and environmental sciences/Natural hazards Earth and environmental sciences/Environmental sciences/Environmental impact InSAR Landslide identification Visual analysis Deep learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 Figure 13 Figure 14 1. Introduction Geological hazards such as landslides occur frequently in Jianzha County, Qinghai Province, severely impacting the ecological environment and social stability. Consequently, it is imperative to monitor and provide early warnings for landslides, and to implement effective preventive and responsive measures. Due to various inducing factors of landslide disasters, such as river scouring, heavy rainfall, earthquake, human activities, etc., it will not only cause serious casualties, property losses, traffic interruption, environmental damage and other direct hazards, but also produce secondary disasters such as debris flow and dammed lake [ 1 ], which greatly restricts the development and utilization of resources [ 2 ]. Landslides are characterized by sudden onset, high difficulty of management, and frequent occurrences in groups [ 3 ], all of which complicate the early identification of landslide hazards. Timely, accurate, and efficient early identification of mountain landslides has become a pressing issue in the field of disaster prevention. Field surveys and regular updates of landslide inventory maps are conventional methods for investigating potential landslides. However, relying solely on traditional manual field surveys or expert interpretation introduces subjective biases and consumes considerable resources. Additionally, these methods exhibit inefficiencies and inaccuracies in identifying and monitoring potential or subtly moving landslides [ 4 – 6 ], making widespread implementation challenging. In contrast, optical remote sensing data offer extensive coverage and high identification accuracy, enabling the identification of landslide locations and areas with significant deformations, thereby improving the accuracy of potential landslide monitoring. Consequently, landslide monitoring using optical imagery has gained significant attention. Nevertheless, optical remote sensing data alone are insufficient for detecting potential landslides with minor deformations and no destructive hazards [ 7 ]. Radar data exhibit features of continuous operation, all-weather capability, and high precision. Research has illustrated their expansive application prospects and growth potential in surface deformation detection. Interferometric synthetic aperture radar (InSAR) techniques, a novel spatial surface measurement technology, remains unaffected by weather conditions and offers benefits such as high precision, low cost, extensive coverage, and continuous operation, rendering it widely adopted for automated landslide hazard monitoring. Persistent Scatterer InSAR (PS-InSAR) and small baseline subset InSAR (SBAS-InSAR) are commonly used in long time series InSAR Technology [ 8 – 18 ]. PS-InSAR mitigates temporal-spatial baseline self-coherence and atmospheric interference in surface deformation monitoring over extended periods. SBAS-InSAR inherits the advantages of D-InSAR technology and enables the acquisition of extensive and contiguous surface deformation information. Relying solely on a single surface deformation monitoring technique for potential landslide hazard identification entails uncertainties. Integrating both techniques can enhance identification outcomes to a certain extent. Elevation data are utilized in InSAR data processing for interferogram unwrapping, image registration, geocoding, etc., and are also employed for extracting subsequent terrain factors to further delineate landslide-prone areas. Utilizing multi-source remote sensing technology to study the dynamic changes in the development of landslide-type debris flows [ 19 , 20 ]. Wu [ 21 ] utilized InSAR technology and optical remote sensing for early identification and monitoring of landslides in Guizhou Province, a crucial step in improving the area's geological disaster prevention and control capabilities. Piroton [ 22 ] utilized a combination of drone-based imagery, radar, and optical remote sensing technologies to detect terrain elevation changes associated with rapid and slow surface displacements, along with meteorological analyses, to identify triggering conditions leading to slope instability. Casagli [ 23 ] described the application of RSTS in landslide analysis and management, and the analysis showed that remote sensing technology has great potential in landslide detection, monitoring and prediction. Bouali [ 24 ] utilized Synthetic Aperture Radar (SAR), optical remote sensing, and Global Positioning System (GPS) to quantify the incremental and average deformation of the Portuguese Bend landslide in California from 2007–2017. While machine learning methods are widely applied in landslide modeling due to their simplicity and strong interpretability [ 25 ], their limited capability to explore correlations between input variables hampers their ability to extract deep features from the data. Furthermore, these models lack the autonomy to perform feature learning [ 26 ]. Different from traditional machine learning methods, Convolutional Neural Networks (CNNs) are machine learning model containing multiple convolution layers and pooling layers, which reflects the structure of human brain neural network. By emulating the neural networks of the human brain, CNNs can comprehend learning objectives and automatically analyze data, facilitating the identification of landslide-prone areas in larger scenes without the requirement for visual interpretation [ 27 – 34 ]. Through extensive training of a large number of data sets, deep features can be obtained from complex data. Currently, the identification of landslide disasters through intelligent means has become a prevailing trend. However, there is a lack of deep neural network models specifically trained for joint InSAR technology in landslide image recognition. It is crucial to leverage multi-source databases and intelligent methods to construct high-precision models tailored for potential landslide disaster identification in this region, thereby enabling the intelligent identification of large-scale landslide distribution. This study, grounded in terrain visibility analysis, aims to integrate PS-InSAR, SBAS-InSAR, and optical remote sensing technologies to delineate potential landslide areas. Subsequently, by employing Google imagery in conjunction with deep learning models, it seeks to identify potential landslide points. Finally, the study integrates terrain, precipitation, and temperature data to analyze the distribution patterns and underlying causes of landslide development. This research provides novel approaches for the large-scale, automated early identification of mountain landslides and landslide susceptibility mapping, significantly contributing to disaster prevention and mitigation efforts. 2. Materials and Methods 2.1. Study Area Jianzha county (101°38′-102°06′E, 35°40′-36°10′N) is located in the southeast of Qinghai Province, with a total area of 1714km2 and an altitude of 1995-4263m. It is located in the transition zone between the Qinghai Tibet Plateau and the Loess Plateau in China [ 35 ]. The Yellow River traverses the county from north to south, extending 96 kilometers within the county boundaries, and exhibits multiple river terraces and basin-hill landforms, characterized by complex geological conditions. The terrain of the study area is high in the South and North, low in the middle, high in the West and low in the East. The unique geological conditions, landform environment, climatic influences, and issues of land desertification in the upper reaches of the Yellow River have led to the development of numerous landslide hazards in this region. Landslide hazards in Jianzha County are predominantly found near the Yellow River, particularly in Kanbula, Nengke, Cuozhou, and Jianzha Township. The study area in Jianzha County is illustrated in Fig. 1 . 2.2. Data Collection and Processing The Sentinel-1A data utilized in this study were sourced from NASA ASF ( https://search.asf.alaska.edu/ ), while the Precise Orbit Data (POD) were procured from the Copernicus Data Space Ecosystem ( https://dataspace.copernicus.eu/ ). A dataset comprising 60 monthly images, spanning from January 2018 to December 2022, was selected as the primary data source. All data were acquired along the same ascending orbit, featuring a spatial resolution of 5m×20m, a 12-day revisit period, C-band (wavelength 5.6 cm), an incidence angle of 37.13°, and the Interferometric Wide (IW) swath imaging mode with VV polarization. The precise orbit data effectively eliminate the orbital errors associated with Sentinel-1A. High-resolution Google satellite images (3m spatial resolution) from 2020 to 2022 were employed to construct a landslide recognition model using a deep learning network, thus providing substantial support for feature learning with an extensive landslide sample dataset. The dataset used for deep learning examples is the Bijie_Landslide_Dataset, curated by Wuhan University. The dataset data is obtained from TripeSat satellite data with 0.8m resolution in panchromatic band and 3.2m resolution in multispectral band. This study primarily utilizes 770 landslide images and their corresponding binary mask data derived from satellite imagery. In these masks, landslide sample values are uniformly set to 0, while non-landslide sample values are set to 1. The SRTM DEM was obtained from USGS ( https://earthexplorer.usgs.gov/ ), utilizing SRTM1 data with a spatial resolution of 30m as elevation reference data. These data were used for geographic coordinate reference in surface deformation and extraction of environmental factors influencing landslide development. Historical landslide validation data were sourced from the Spatial Distribution of Geological Disaster Points dataset provided by the Chinese Academy of Sciences' Resource and Environmental Science and Data Center ( https://www.resdc.cn/ ). Lithological data were obtained from ISRIC ( https://www.isric.org/ ) for the lithological analysis of landslide hazard point distribution. Precipitation and temperature data were obtained from National Earth System Science Data Center ( https://www.geodata.cn/ ) with a spatial resolution of 1km. The possible effects of precipitation and air temperature on landslide surface deformation were investigated using monthly data from 2018 to 2022. 2.3. Methodology This paper proposes a method for identifying potential landslide points based on InSAR technology and deep learning techniques. The main approach is as follows: First, the visibility of the area is analyzed by combining SAR imagery and DEM data to verify the feasibility of InSAR technology. Secondly, the PS-InSAR technology and SBAS-InSAR technology were used to delineate a large area of surface deformation from sentinel-1a images of 60 scenes in Jianzha county from January 2018 to December 2022. Given the close correlation between surface deformation areas and potential landslide hazards, deep learning and visual interpretation are applied to Google imagery to identify potential landslide distribution points within these areas. Finally, the topographical features of landslide development in the region are extracted using known landslide points. The topographical characteristics of historical landslide points serve as the basis for assessing landslide hazard levels. Points matching InSAR surface deformation characteristics, deep learning identification features, and topographical attributes are classified as high-hazard landslide points, while the remaining points are categorized as medium-hazard landslide points. Finally, the factors contributing to landslide deformation are analyzed by integrating topographical, precipitation, and temperature data. The technical workflow is illustrated in Fig. 2 : 2.3.1. InSAR deformation monitoring Synthetic Aperture Radar Interferometry (InSAR) is an advanced space-to-earth observation technology. The radar system interferes with the changing phase information generated by more than two multi-temporal SAR images in the same area, and finally can obtain a large range, high precision and high resolution surface deformation information in the time series [ 9 , 21 , 36 ]. Due to the side-looking imaging of SAR sensors, the incident angle formed when the radar hits ground objects affect the image quality. As the incident angle decreases, the image's echo signal is enhanced, resulting in brighter pixels [ 37 ]. This unique imaging method makes the radar images highly susceptible to displaying anomalies. When the beam is directed to the slope, because the irradiation distance of the bottom surface is larger than that of the top, the top is imaged before the bottom, resulting in an inverted overlay phenomenon. When the slope changes greatly, the beam is difficult to illuminate the back slope, so that the sensor is difficult to receive the reflection of the ground object on the back slope, and the shadow area will be shown on the map. In mountainous regions with significant relief degree, these phenomena will become more pronounced, potentially leading to visual blind spots. Perspective foreshortening refers to the distortion of object size and shape in the image due to the spatial relationship between the sensor and the ground, affecting the accurate measurement of distance and dimensions. To mitigate the impact of these phenomena on landslide point identification, a visibility analysis of the area must be conducted prior to detecting potential landslides using PS-InSAR and SBAS-InSAR technologies. The fundamental principle of PS-InSAR involves statistically analyzing the amplitude information of radar images from the same area and identifying permanent scatterers that remain unaffected by temporal and spatial baseline decorrelation and atmospheric delays to perform phase modeling and deformation calculation [ 12 , 38 ]. For N + 1 SAR images of the study area, the primary image is selected based on image quality and the distribution of spatial and temporal baselines, and differential interferometric processing is performed on the registered images to generate N interferograms. By analyzing the differences in Permanent Scatterer (PS) points across different image pairs, the deformation rate and elevation error of each PS point relative to the main reference point can be calculated, accounting for factors such as atmospheric effects, orbital errors, and surface deformation. After obtaining the differential interferometric phase, phase unwrapping is necessary to extract the actual physical displacement information. The unwrapped linear phase residuals can be used to calculate the actual surface deformation. However, this technique may result in poor coherence for some interferometric pairs due to long spatial and temporal baselines when selecting one image as the common primary image and using the remaining images as secondary images. In mountainous regions with significant relief degree land surface and low coherence, supplementary techniques are required to validate the deformation monitoring results of PS-InSAR. SBAS-InSAR is an InSAR time series method based on multiple master images. Its fundamental principle involves calculating the spatial-temporal baselines of images from different times within the region using the short baseline principle, selecting appropriate spatial-temporal thresholds to form interferograms, and performing multi-look processing to reduce noise. Using the singular value decomposition (SVD) method, the spatial small baseline subset data are combined into a time series to compute the least squares solution within the subset [ 11 , 39 ]. Finally, the residual phase is used to invert the atmospheric phase and non-linear phase to derive the deformation time series for the given time phase. Compared to PS-InSAR, SBAS-InSAR can further mitigate the effects of temporal and spatial decorrelation, thus obtaining higher precision deformation information. If the study area has N + 1 SAR images, the registered images are arranged in chronological order and divided into several sets based on spatial-temporal baselines for differential interferometric processing, resulting in M interferograms. The resulting M interferograms must satisfy the constraint: $$\frac{\text{N+1}}{\text{2}}\text{≤M≤}\frac{\text{N(N+1)}}{\text{2}}$$ 1 The expression for the interferometric phase of the i-th interferogram pair is: $${\text{φ}}_{\text{i}}\text{=}{\text{φ}}_{\text{topo}}\text{+}{\text{φ}}_{\text{flat}}\text{+}{\text{φ}}_{\text{orb}}\text{+}{\text{φ}}_{\text{def}}\text{+}{\text{φ}}_{\text{atm}}\text{+}{\text{φ}}_{\text{scat}}\text{+}{\text{φ}}_{\text{noise}}$$ 2 $${\text{φ}}_{\text{topo}}\text{=-}\frac{\text{4π}{\text{B}}_{\text{┴}}\text{h}}{\text{λRsinθ}}$$ 3 $${\text{φ}}_{\text{flat}}\text{=-}\frac{\text{4π}{\text{B}}_{\text{┴}}}{\text{λ}}$$ 4 In the equation, φ topo represents the topographic phase; φ flat represents the flat-earth phase; φ orb represents the orbital error phase; φ def represents the deformation phase, which includes both deformation and non-deformation components; φ atm represents the atmospheric error phase; φ scat represents the phase due to changes in point target scattering characteristics; and φ noise represents the noise phase. B ┴ represents the perpendicular baseline length, h represents the elevation error, λ represents the radar wavelength, R represents the slant range, and θ represents the incidence angle. 2.3.2. Landslide recognition based on deep learning This study utilizes a deep learning framework based on the open-source Python machine learning library PyTorch to build and train a ResNet50 network model for learning landslide features. After extracting classification information, the trained network is validated using test data. Landslide classification divides the sample set into landslide and non-landslide targets, i.e., a binary classification, and learns to delineate landslide boundaries. During the adjustment of learning rate and loss weight, the optimal parameter combination is iteratively updated to obtain the best-trained network. Finally, the landslide recognition model is evaluated using binary classification metrics: Precision, Accuracy, and Recall. Precision represents the accuracy of the positive part predicted by the model; Accuracy is the accuracy of the model. The higher the accuracy, the better the model effect; Recall rate is the correct proportion predicted by the model. The calculation formulas are as follows: $$\text{Precision=TP/(TP+FP)}$$ 5 $$\text{Accuracy=(TP+TN)/(TP+TN+FP+FN)}$$ 6 $$\text{Recall=TP/(TP+FN)}$$ 7 Where True Positive (TP) represents the number of samples correctly predicted as positive by the model; False Positive (FP) represents the number of samples incorrectly predicted as positive by the model; True Negative (TN) represents the number of samples correctly predicted as negative by the model; and False Negative (FN) represents the number of samples incorrectly predicted as negative by the model. 3. Results 3.1 Visibility analysis Spaceborne radar satellites, due to their side-looking imaging mode, often result in images with layover, shadow, and foreshortening effects when observing the ground using radar beams. Based on elevation data of the study area, the county exhibits significant elevation differences, with variations reaching over 2000 meters. The dramatic topographical relief makes the area highly susceptible to tropospheric atmospheric delays when monitored using InSAR technology. In this region, the satellite incidence angle is 37.13° [ 40 ]. The visibility classification standards are based on Sentinel-1A's incidence angle and topographical factors, dividing the area into three categories: non-visibility, low sensitivity, and high visibility. Non-visibility areas are primarily located on slopes facing east, southeast, and northeast. When the slope angle is less than 37.13 °, perspective shrinkage will occur, and when the slope angle is greater than 37.13 °, overlap will occur. Visibility areas include high visibility and low sensitivity regions. High visibility areas are mainly on slopes facing west, southwest, and northwest, where shadowing occurs when the slope angle exceeds 52.87°. Low sensitivity areas are primarily on the south and north slopes, which are less sensitive to surface deformation. The classification of visibility types and the geometric distortion areas of the study region are shown in Fig. 3 . From the mapping results, visibility areas cover 1252.42 km², accounting for 73.07% of the area. Layover and shadow areas are sparse, covering only 6.14 km² (0.36%). Foreshortening areas are mostly on southeast-facing back slopes, covering 455.41 km² (26.57%). Historical landslide point distributions reveal that landslides frequently occur in valleys and low mountain hills near the Yellow River. These areas are minimally affected by layover, shadow, and foreshortening, indicating that the selected ascending radar imagery has good visibility and high reliability for landslide monitoring. 3.2 PS-InSAR deformation monitoring results The image from January 4, 2020, was selected as the primary image, with the remaining 59 images serving as secondary images. These were used to generate SAR data pairs and connection graphs for subsequent differential interferometry processing. The temporal baseline between the primary image and all secondary images ranges from − 133–131d, and the spatial baseline ranges from − 92–131m, both below the critical baseline. The longest temporal baseline is 133 days, corresponding to the image from June 10, 2022. PS-InSAR involves processing each interferometric pair individually for registration and interferometry, resulting in unwrapped phase maps of the residual phase. During registration, the secondary images are aligned with the primary image, with a range-to-azimuth ratio set at 4:1. After completion, quick-look images are examined to verify the registration and interferometry results of all pairs. Interferometric pairs with poor unwrapping or coherence are excluded, ensuring that only correctly processed pairs are used. The first inversion method automatically selects reference points with minimal deformation and high coherence, and then analyzes the phase changes over the time series to obtain displacement rates and residual elevation. The second inversion converts the phase shifts from the first inversion results into deformation information in the geographic coordinate system, yielding the final deformation rates. The deformation data from the first and second inversions are geocoded to produce a PS point vector file, which is then interpolated to generate a point target deformation rate map (Fig. 4 ). The line-of-sight (LOS) deformation rate in the study area, as measured by PS-InSAR technology, ranges from − 68 − 28 mm/a. From the deformation rate distribution map, it is observed that the PS-InSAR deformation monitoring area is primarily located in relatively flat valley areas near the Yellow River channel. Fewer valid PS points are extracted from densely vegetated areas, valleys with foreshortening effects, and mountainous regions, making it challenging to assess surface deformation. Significant subsidence is observed in the northern and eastern directions in towns such as Kanbula, Kangyang, Dangshun, and Maketang, while localized uplift is evident in Cuozhou Township. These contrasting phenomena arise due to the concentrated rainfall in the study area, leading to extensive overall subsidence caused by water erosion. Additionally, the expansion of human activities prompts the construction of production-friendly structures near residential areas, resulting in scattered local uplift. 3.3 SBAS-InSAR deformation monitoring results To ensure the highest quality of interferograms and improve result accuracy, the maximum temporal baseline threshold was set to 120d, and the maximum spatial baseline threshold to 2%. The image from August 7, 2020, was selected as the super master image, with the remaining 59 images serving as secondary images, resulting in 228 freely combined interferometric pairs. The temporal baseline between the super master image and all secondary images ranges from − 116–106d, and the spatial baseline ranges from − 110–152m. Image pairs with short temporal and spatial baselines undergo SBAS-InSAR inversion after interferometric processing. Statistical analysis of the connected pairs reveals that the maximum number of pairs for a single image is 11, while the minimum is 1, which is sufficient to meet the landslide identification requirements for Jianzha County. The SBAS-InSAR interferometric processing can flatten, filter, and unwrap the phase of the image pairs. Similarly, set the ratio of distance direction to azimuth direction to 4:1. Subsequently, the topographic phase is removed based on DEM data, and a polynomial model (Goldstein) is employed for filtering. Because the study area is a mountainous area with low coherence, 3D unwrapping is not carried out, and the minimum cost flow is selected. Finally, interference pairs with poor unwrapping effect and coherence are eliminated. The results are shown in Fig. 5 . Using the selected control points as a reference, residual phase and phase ramps remaining after unwrapping are removed, and finally, the unwrapped phase is converted into elevation or deformation values. The introduction of precise ground control points can effectively remove residual constant phase and flat-earth effect, thereby enhancing the validity of the results. The selection of ground control points should first ensure uniform coverage over a wide area with high coherence, good unwrapping results, and stability; residual topographic areas, phase jump regions, and deformation stripe areas should be avoided as much as possible [ 12 ]. Since the GCP files obtained from PS-InSAR cannot be directly used for SBAS-InSAR processing, automatically selected and geocoded GCP points from the PS-InSAR technique are chosen here. This reduces subjective errors in selecting control points manually in the SBAS-InSAR method while enhancing the accuracy of deformation estimation. The distribution of GCP points is shown in Fig. 6 (a). The distribution of GCP points indicates that the 266 selected GCP targets are uniformly distributed within the study area and are mostly located in high-coherence urban regions, which aligns with the actual conditions of the study area. The deformation data obtained from SBAS-InSAR is geocoded to produce a deformation rate map (Fig. 6 (b)). Apart from the mountainous regions with significant topographic relief in the southwest and northwest of Jianzha County, the coverage extracted using this monitoring method shows a significant improvement compared to the deformation rate results of PS-InSAR. Deformation information can be extracted even in some mountainous areas. The LOS deformation rate of SBAS-InSAR ranges from − 62 − 37 mm/a. Using SBAS-InSAR technology, it is found that, in addition to the slight uplift observed in Cuozhou Township, local uplift also occurs in Maketang Township and Angla Township. The reason is that the number of permanent scatterers extracted by the PS-InSAR technology is limited by the environment, resulting in a limited amount of deformation data obtained [ 38 ], making it difficult to meet the requirements for extracting large-area continuous surface deformation. In contrast, SBAS-InSAR compensates for this deficiency, making it more effective for extracting surface deformation information within the study area. Therefore, subsequent surface deformation information will primarily utilize SBAS-InSAR technology, supplemented by surface subsidence data extracted using PS-InSAR, to better identify potential landslide hazards. To evaluate the reliability of the combined results from PS-InSAR and SBAS-InSAR technologies, 400 identical points were selected from the annual average deformation rates obtained using Sentinel-1A data for both technologies. A statistical analysis of the linear relationship between the annual average deformation rates (Fig. 7 ) reveals that the deformation rate distribution of identical points in both technologies is essentially consistent, with an R² greater than 0.91. This indicates a high correlation between the two technologies, demonstrating the validity of combining the results from PS-InSAR and SBAS-InSAR technologies. 3.4 Landslide recognition results Due to the limited number of samples in the landslide dataset, and considering that convolutional neural networks require a large number of samples to extract feature information. To improve the accuracy of landslide sample recognition and achieve the best recognition results, data augmentation of the landslide samples is necessary. The samples in the data set were randomly rotated 90°, 180° and 270°, horizontally mirrored and vertically mirrored to increase the quality of the landslide data set. The image data after data amplification is 3850 pieces, and then the training set and the verification set are randomly selected according to the ratio of 7:3. Select any landslide sample to show the effect of data expansion process as shown in the figure. The Fig. 8 shows the processing examples of original sample, rotation 270°, rotation 270°+ horizontal flip, vertical mirroring, and 90° rotation. The landslide prediction samples are created using Google satellite images with a spatial resolution of 3 meters. The Google images of Jianzha County are preprocessed by cropping them into 256×256 pixels. The landslide hazard areas in the prediction samples are then identified and annotated sequentially, and finally, the annotated samples are merged to obtain the landslide prediction distribution map of the study area. Influenced by the topography, the surface deformation is mostly in the negative direction of LOS and has a large deformation rate, which is very easy to cause landslide disasters. Based on the surface deformation information extracted using the aforementioned PS-InSAR and SBAS-InSAR technologies, areas with significant deformation, reaching the threshold of 20 mm/a in the LOS direction, are delineated as landslide hazard zones. By integrating Google Earth images, the areas with significant deformation from 2020 to 2022 are summarized to obtain the distribution of landslide hazard zones. The model training is set to 1500 iterations, with a batch size of 32 and an initial learning rate of 0.0001. For the prediction part of the model, the potential landslide areas delineated in the Google images are cropped into 256×256 pixel files. The landslide boundaries are then mapped on the subdivided local images. Combining the augmented dataset, the test accuracy curve and training loss curve (Fig. 9 ) reveal that the test accuracy of the model rapidly increases from 43.6% at the beginning of training, reaching 85% for the first time after 80 iterations. The fluctuation amplitude decreases after 800 iterations, stabilizing around 91% after 1500 iterations. Training loss decreases from 0.52 at the beginning of training and stabilizes at 0.0557 after 1500 iterations. Considering both the test accuracy and iteration number as evaluation metrics for the weight files, the 770th weight file (test accuracy 94.4%, test precision 98.7%, recall 84.2%) is selected for subsequent landslide hazard monitoring. A total of 56 potential landslide areas are identified. As of 2020, 10 of the 13 historical landslide disaster points are located at the identified high hidden danger landslide points, and 3 landslides are missed, with the monitoring accuracy of 76.92%. For easier statistical analysis, the potential landslide hazard points are represented as dots. The distribution of historical and potential landslide monitoring points is shown in Fig. 10 . The topographic features of historical landslides are extracted to assess the 56 identified landslide hazard points, revealing 39 high-risk and 17 medium-risk landslide points. 4. Discussion Influence of topography on landslide distribution The frequent occurrence of landslides is closely related to geological structures. By integrating slope topography factors, the environmental elements for landslide development are extracted, including slope, aspect, relief degree, and lithology, to analyze the control conditions of the landslide development environment (Fig. 11 ). Since steep slopes increase the risk of soil or rock instability, combined with the differences in slope topography, humidity, and vegetation cover due to different slope aspects, slope angle is often considered an important factor in landslide prevention and mitigation planning. Using the equal interval method, the slope is divided into six categories to count the number of landslides: 0–10°, 10–20°, 20–30°, 30–40°, 40–50°, and above 50°. Aspect determines the wind direction, precipitation, surface water flow direction, and vegetation growth, which all affect slope stability. To visually display the relationship between landslides and aspect, the slope aspects are categorized into eight directions for statistical analysis. Relief degree refers to the vertical distance between the highest and lowest points within a specific area and is typically used to measure the unevenness of the terrain, reflecting the slope characteristics that foster landslide occurrences. Statistics reveal that the relief degree of landslide disaster points in this area is concentrated between 0 and 90 meters. Using the equal interval method, the relief degree is divided into six levels: 0-15m, 15-30m, 30-45m, 45-60m, 60-75m, and 75-90m. Research indicates that the distribution of landslides is inseparable from the characteristics of rock and soil masses. The rock or soil types and structures in landslide areas differ significantly from those in surrounding non-sliding slopes. Previously sliding rock layers or soils usually have a looser structure, making areas with frequent fault activities prone to landslides. Therefore, lithology is selected as one of the topographic factors for analyzing landslide development. Analyzing the development and distribution characteristics of potential landslide points using slope terrain factors. As illustrated in Fig. 12 (a), potential landslides are predominantly distributed within the 10–40° slope range, comprising 76.8% of the total. This indicates that there is no significant positive correlation between landslide distribution and slope steepness. Landslides are more likely to occur on natural slopes where environmental factors such as wind direction, precipitation, lithology, and soil quality are favorable. The relief degree statistics indicate that the relief degree in this area is significantly influenced by lithology. Potential landslide points are predominantly distributed at elevations of 15-45m, primarily in areas where rainfall and unsound engineering activities have caused environmental damage. 98.21% of landslide points exhibit an relief degree of less than 60m, and as relief degree increases above 60m, there is no observed trend of worsening landslide severity. Lithological statistics indicate that landslide points are predominantly found in areas composed of sandstone, mudstone, feldspathic sandstone, and aeolian deposits. These lithological formations belong to a group prone to sliding and, upon contact with water, form impermeable layers within the slope, thereby creating potential sliding surfaces. The statistical results of slope aspect are shown in Fig. 12 (b). It is found that potential landslides are concentrated on the northeast and Southeast slopes, and the probability of occurrence of the facing slope is significantly higher than that of the back slope. Combining the analysis of slope aspect characteristics, it is highly likely that the terrain causes windward slopes to receive more rainfall than leeward slopes, thus reducing slope stability and loosening the soil, making precipitation a dominant factor in landslide occurrence. Additionally, the study area's topography, characterized by higher elevations in the west and lower elevations in the east, provides natural east-facing slopes with effective free faces, thereby making landslides more likely to occur. From the observed landslide points, the aforementioned four types of terrain factors were extracted as potential landslide development elements. The thresholds for landslide-prone factors were identified as follows: elevation between 2066-2839m for river terraces and low hills, slope gradient of 5.19–39.78°, aspect of 39.69-358.53°, surface relief degree of 5-54m, with no significant differentiation in lithological factors. Therefore, it can be concluded that landslides are more likely to occur when the aforementioned slope terrain factors fall within the appropriate threshold ranges, combined with environmental triggers such as rainfall. Not all potential landslide points are equally prone to landslide disasters. Impact of precipitationand temperature Precipitation is a critical factor influencing landslide deformation [ 41 ]. Due to the influence of terrain and the southeast monsoon, precipitation is higher in the southwestern part of Jianzha County, characterized by mid- and high-mountain areas. In 2018, Jianzha County received an annual precipitation of 570.9mm, with summer precipitation averaging 323.7mm, comprising 56.7% of the annual total. The precipitation in spring and autumn were 101.5mm and 138.7mm respectively. The precipitation in winter is only 7mm. Synthesis of the monitoring data from the past five years indicates that the average downward deformation rates at the landslide points are consistently 0.17mm•(30d) -1 , 0.79mm•(30d) -1 , 3.71mm•(30d) -1 , 3.04mm•(30d) -1 , -7.05mm•(30d) -1 , 6.09mm•(30d) -1 , 6.52mm•(30d) -1 , 4.04mm•(30d) -1 , and 5.86mm•(30d) -1 respectively. From July to September 2018, the deformation rates at P2, P3, P4, and P7 reached 6-10mm•(30d) -1 , which was much higher than the average deformation rate. From June to September 2019, the deformation rates at P4, P6, P8, and P9 reached 3-5mm•(30d) -1 . From August to September 2020, the deformation rates at P6, P7, and P9 reached 3-5mm•(30d) -1 . From August to September 2021, the deformation rate at P4 reached 3mm•(30d) -1 . From June to September 2022, P1, P3, P4, and P5 experienced sudden deformation changes, reaching 4-7mm•(30d) -1 (Fig. 13 ). These peak deformation periods coincided with times of higher annual precipitation, indicating a correlation between landslides and rainfall, and suggesting that areas with past landslides are at risk of recurring events. Jianzha County is characterized by a plateau cool temperate semi-arid climate, with frequent spring droughts, hail, and frost events, large diurnal temperature variations, long sunshine duration, and intense solar radiation. The annual average temperature in the region is 7.9°C, with an extreme annual maximum temperature of 34.1°C, and annual sunshine duration totaling 4432.3 hours. The relationship between monitoring points and temperature indicates that from October 2018 to January 2019, the deformation rates at P1, P3, P4, and P6 reached 3-6mm•(30d) -1 . From October 2019 to January 2020, the deformation rates at P4 and P7 reached 4-7mm•(30d) -1 . From October 2020 to January 2021, the deformation rates at P1 and P5 reached 2-7mm•(30d) -1 , significantly higher than the average deformation rates at these points (Fig. 14 ). These periods of significant deformation coincide with the winter months, characterized by low precipitation and large temperature variations, indicating that, in addition to rainfall, temperature changes also affect the deformation rates of landslides. 5. Conclusions This paper combines InSAR technology with deep learning methods, utilizing InSAR technology to monitor surface deformation and convolutional neural networks to identify potential landslides in Jianzha County based on known landslide characteristics. The main conclusions are as follows: (1) PS-InSAR and SBAS-InSAR technologies exhibit a high correlation. Time series analysis of extracted feature points confirms that the surface deformation trends detected by both InSAR technologies are consistent. In this study, SBAS-InSAR technology is primarily used for surface deformation information, supplemented by surface subsidence data extracted by PS-InSAR, allowing for better identification of potential landslide hazards. (2) Based on the deep learning model and combined with existing landslide data, potential landslide points can be quickly and accurately identified, particularly the 39 high-risk landslide points, which have a high probability of experiencing actual landslide events in a short period. (3) Landslide points in Jianzha County are concentrated on the northeast and southeast slopes with angles ranging from 10° to 40°, and the relief degree is concentrated in river terraces and low hilly areas with elevations ranging from 15 to 30m. The left bank of the Yellow River is a key area for landslide occurrences. The formation of landslides is also related to precipitation and temperature. These studies demonstrate that combining InSAR technology with deep learning methods can quickly and accurately identify early potential landslide points, providing scientific evidence for landslide monitoring and management. Future research should focus on improving the landslide dataset, particularly by incorporating historical landslide data from the study area, to make the landslide identification results more targeted. Declarations Data availability The datasets generated and analyzed during the current study are not publicly available due to the funding responsibility but are available from the corresponding author on reasonable request. Acknowledgments The research was sponsored by the National Natural Science Foundation of China (No. 41701449 and No. 41930102), Nanhu Scholars Program for Young Scholars of XYNU. Postgraduate Education Reform and Quality Improvement Project of Henan Province (HNYJS2020JD14). Author Contributions Conceptualization, X.Y. and D.C.; methodology, X.Y. and D.C.; software, Y.D.; validation, Y.X.; formal analysis, K.Q.; investigation, Y.D.; resources, Y.X. and K.Q; data curation, D.C.; writing—original draft preparation, D.C.; writing—review and editing, X.Y.; visualization, D.C.; project administration, X.Y. All authors have read and agreed to the published version of the manuscript. All authors agree to submit this manuscript and affirm that it has not been published or submitted to any other journal. Conflicts of Interests The authors declare no conflict of interest. Additional information Correspondence and requests for materials should be addressed to X.Y. References Flentje P, Chowdhury R. Resilience and sustainability in the management of landslides[C]//Proceedings of the institution of civil engineers-engineering sustainability. Thomas Telford Ltd, 171(1): 3–14(2016). Constantin, M.; Bednarik, M.; Jurchescu, M.C.; Vlaicu, M. Landslide susceptibility assessment using the bivariate statistical analysis and the index of entropy in the Sibiciu Basin (Romania). Environ. Earth Sci, 63, 397–406(2011). [CrossRef] Chen Siming; Huo Aidi; Zhang Jia et al. Identification of potential landslides in the loess hilly area (Xiji County) of Ningxia with InSAR technology. Science Technology and Engineering, 22(12): 4721–4728(2022). Jin, Y.; Li, X.; Zhu, S.; Tong, B., Chen, F.; Cui, R.; Huang, J. Accurate landslide identification by multisource data fusion analysis with improved feature extraction backbone network. Geomatics, Natural Hazards and Risk, 13(1), 2313–2332(2022). Pang, D.; Liu, G.; He, J.; Li, W.; Fu, R. Automatic Remote Sensing Identification of Co-Seismic Landslides Using Deep Learning Methods. Forests, 13, 1213(2022). Kovács, I.P.; Czigány, S.; Dobre, B. et al. A field survey–based method to characterise landslide development: a case study at the high bluff of the Danube, south-central Hungary. Landslides, 16, 1567–1581(2019). Xin W.; Xuanmei F.; Qiang X.; Peijun D. Change detection-based co-seismic landslide mapping through extended morphological profiles and ensemble strategy, ISPRS Journal of Photogrammetry and Remote Sensing, 187, 225–239(2022). Zhang, T.; Zhang, W.; Cao, D.; Yi, Y.; Wu, X. A New Deep Learning Neural Network Model for the Identification of InSAR Anomalous Deformation Areas. Remote Sens., 14, 2690(2022). Jia, H.; Wang, Y.; Ge, D.; Deng, Y.; Wang, R. InSAR Study of Landslides: Early Detection, Three-Dimensional, and Long-Term Surface Displacement Estimation—A Case of Xiaojiang River Basin, China. Remote Sens., 14, 1759(2022). [CrossRef] Zhang, R.; Zhao, X.; Dong, X.; Dai, K.; Deng, J.; Zhuo, G.; Yu, B.; Wu, T.; Xiang, J. Potential Landslide Identification in Baihetan Reservoir Area Based on C-/L-Band Synthetic Aperture Radar Data and Applicability Analysis. Remote Sens., 16, 1591(2024). Dong, J., Niu, R., Li, B., Xu, H., & Wang, S. (2022). Potential landslides identification based on temporal and spatial filtering of SBAS-InSAR results. Geomatics, Natural Hazards and Risk, 14(1), 52–75(2022). [CrossRef] 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., 14, 4728(2022). [CrossRef] Huang, H.; Ju, S.; Duan, W.; Jiang, D.; Gao, Z.; Liu, H. Landslide Monitoring along the Dadu River in Sichuan Based on Sentinel-1 Multi-Temporal InSAR. Sensors, 23, 3383(2023). Zhang, J.; Gong, Y.; Huang, W.; Wang, X.; Ke, Z.; Liu, Y.; Huo, A.; Adnan, A.; Abuarab, M.E.-S. Identification of Potential Landslide Hazards Using Time-Series InSAR in Xiji County, Ningxia. Water, 15, 300(2023). Yi, Y.; Xu, X.; Xu, G.; Gao, H. Rapid Mapping of Slow-Moving Landslides Using an Automated SAR Processing Platform (HyP3) and Stacking-InSAR Method. Remote Sens., 15, 1611(2023). Hussain, S., Pan, B., Afzal, Z. et al. Landslide detection and inventory updating using the time-series InSAR approach along the Karakoram Highway, Northern Pakistan. Sci Rep, 13, 7485(2023). Lian, B.; Wang, D.; Wang, X.; Tan, W. Early Identification and Dynamic Stability Evaluation of High-Locality Landslides in Yezhi Site Area, China by the InSAR Method. Land, 13, 569(2024). Kalavrezou, I.-E.; Castro-Melgar, I.; Nika, D.; Gatsios, T.; Lalechos, S.; Parcharidis, I. Application of Time Series INSAR (SBAS) Method Using Sentinel-1 for Monitoring Ground Deformation of the Aegina Island (Western Edge of Hellenic Volcanic Arc). Land, 13, 485(2024). Guo, H.; Martínez-Graña, A.M. Susceptibility of Landslide Debris Flow in Yanghe Township Based on Multi-Source Remote Sensing Information Extraction Technology (Sichuan, China). Land, 13, 206(2024). Albanwan, H.; Qin, R.; Liu, J.-K. Remote Sensing-Based 3D Assessment of Landslides: A Review of the Data, Methods, and Applications. Remote Sens., 16, 455(2024). Wu, L.; Wang, J.; Fu, Y. Early identifying and monitoring landslides in Guizhou province with InSAR and optical remote sensing. Bulletin of Surveying and Mapping, (07), 98–102(2021). [CrossRef] Piroton V, Schlögel R, Barbier C, et al. Monitoring the recent activity of landslides in the Mailuu-Suu Valley (Kyrgyzstan) using radar and optical remote sensing techniques, Geosciences, 10(5): 164(2020). Casagli N, Intrieri E, Tofani V, et al. Landslide detection, monitoring and prediction with remote-sensing techniques, Nature Reviews Earth & Environment, 4(1): 51–64(2023). Bouali E H, Oommen T, Escobar-Wolf R. Evidence of Instability in Previously-Mapped Landslides as Measured Using GPS, Optical, and SAR Data between 2007 and 2017: A Case Study in the Portuguese Bend Landslide Complex, California, Remote Sensing, 11(8): 937–956(2019). Chen, C.; Shen, Z.; Weng, Y.; You, S.; Lin, J.; Li, S.; Wang, K. Modeling Landslide Susceptibility in Forest-Covered Areas in Lin’an, China, Using Logistical Regression, a Decision Tree, and Random Forests. Remote Sens, 15, 4378(2023). Sheng, Y.; Xu, G.; Jin, B.; Zhou, C.; Li, Y.; Chen, W. Data-Driven Landslide Spatial Prediction and Deformation Monitoring: A Case Study of Shiyan City, China. Remote Sens, 15, 5256(2023). Xiao L, Zhang Y, Peng G. Landslide susceptibility assessment using integrated deep learning algorithm along the China-Nepal highway, Sensors, 18(12): 4436(2018). Xiong, K.; Adhikari, B.R.; Stamatopoulos, C.A.; Zhan, Y.; Wu, S.; Dong, Z.; Di, B. Comparison of Different Machine Learning Methods for Debris Flow Susceptibility Mapping: A Case Study in the Sichuan Province, China. Remote Sens., 12, 295(2020). Huang F, Zhang J, Zhou C, et al. A deep learning algorithm using a fully connected sparse autoencoder neural network for landslide susceptibility prediction, Landslides, 17: 217–229(2020). Van Dao D, Jaafari A, Bayat M, et al. A spatially explicit deep learning neural network model for the prediction of landslide susceptibility, Catena, 188: 104451(2020). Shahabi, H.; Ahmadi, R.; Alizadeh, M.; Hashim, M.; Al-Ansari, N.; Shirzadi, A.; Wolf, I.D.; Ariffin, E.H. Landslide Susceptibility Mapping in a Mountainous Area Using Machine Learning Algorithms. Remote Sens., 15, 3112(2023). Huang, W., Ding, M., Li, Z. et al. Landslide susceptibility mapping and dynamic response along the Sichuan-Tibet transportation corridor using deep learning algorithms, Catena, 222: 106866(2023). Ali, N.; Chen, J.; Fu, X.; Ali, R.; Hussain, M.A.; Daud, H.; Hussain, J.; Altalbe, A. Integrating Machine Learning Ensembles for Landslide Susceptibility Mapping in Northern Pakistan. Remote Sens., 16, 988(2024). Zhang, Q.; Wang, T. Deep Learning for Exploring Landslides with Remote Sensing and Geo-Environmental Data: Frameworks, Progress, Challenges, and Opportunities. Remote Sens., 16, 1344(2024). Zhang, X.; Yang, L,; Song, X. Runoff and sediment load changes in the upper Yellow River and their influencing factors in recent 60 years. Journal of Lake Sciences, 36(2), 602–619(2024). Richard, W.; Daniel, A.; Mark, F. Volumetric interferometry for sparse 3D synthetic aperture radar with bistatic geometries. Electronics Letters, 59(12), 12851(2023). Chen Y, Xia J, Yu C and Chen B. Editorial: InSAR crustal deformation monitoring, modeling and error analysis. Front. Environ. Sci., 10:1009492(2022). Huang, X.; Li, X.; Li, H.; Yang, Y.; Duan, S.; Xiao, W.; Du, H.; Liu, H. Study on Rock Strata Movement Deformation and Surface Subsidence in Mining Area Based on PS-InSAR Technology. Preprints,06,2212(2023). [CrossRef] Karaca, Ş. O., Erten, G., Ergintav, S., Khan, S. D. Anthropogenic problems threatening major cities: Largest surface deformations observed in Hatay, Türkiye based on SBAS-InSAR. Bulletin of the Mineral Research and Exploration, 173(173), 235–252(2024). Guo R.; Li S.; Chen Y.; Yuan,L. A method based on SBAS-InSAR for comprehensive identification of potential landslide. Journal of Geo-information Science, 21(7):1109–1120(2019). She, X.; Li, D.; Yang, S.; Xie, X.; Sun, Y.; Zhao, W. Landslide Hazard Assessment for Wanzhou Considering the Correlation of Rainfall and Surface Deformation. Remote Sens., 16, 1587(2024). Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 26 Jul, 2024 Reviews received at journal 25 Jul, 2024 Reviews received at journal 23 Jul, 2024 Reviewers agreed at journal 16 Jul, 2024 Reviewers agreed at journal 16 Jul, 2024 Reviewers invited by journal 16 Jul, 2024 Editor assigned by journal 16 Jul, 2024 Editor invited by journal 02 Jul, 2024 Submission checks completed at journal 28 Jun, 2024 First submitted to journal 26 Jun, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4642799","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":328498666,"identity":"3656d956-dd82-4d51-b173-edfc7323bc38","order_by":0,"name":"Xianwu Yang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAvUlEQVRIiWNgGAWjYDACZjBpI8fPzHz4ASla0owl29nSDEix63DihvM8ChJEqTU4zvzsMW9bGuPmwzwMBgw1NtEEtUg2s5kb87bZMJsd5j3wgOFYWm4DIS38zAxm0rltaWxmh/kSDBgbDhPWwsbM/g2o5TCPcTOPgQRRWviZeUC2HJYwYCZWi2QzT5n0n3NpBhKHgYGcQIxfDM4f3yY5o8ymvr//8OEHH2psCGtBBQmkKR8Fo2AUjIJRgAsAAJdxNvGwhpFCAAAAAElFTkSuQmCC","orcid":"","institution":"Xinyang Normal University","correspondingAuthor":true,"prefix":"","firstName":"Xianwu","middleName":"","lastName":"Yang","suffix":""},{"id":328498667,"identity":"79780595-617b-4282-9c2f-ec27be5e3aa2","order_by":1,"name":"Dannuo Chen","email":"","orcid":"","institution":"Xinyang Normal University","correspondingAuthor":false,"prefix":"","firstName":"Dannuo","middleName":"","lastName":"Chen","suffix":""},{"id":328498670,"identity":"b3671765-66e2-48cf-8d5e-e89253af7f62","order_by":2,"name":"Yihang Dong","email":"","orcid":"","institution":"Xinyang Normal University","correspondingAuthor":false,"prefix":"","firstName":"Yihang","middleName":"","lastName":"Dong","suffix":""},{"id":328498672,"identity":"909e322b-00da-45b0-b219-b6e2e23d09c1","order_by":3,"name":"Yamei Xue","email":"","orcid":"","institution":"Xinyang Normal University","correspondingAuthor":false,"prefix":"","firstName":"Yamei","middleName":"","lastName":"Xue","suffix":""},{"id":328498674,"identity":"11699329-d787-4a5f-a254-436e08c9ece5","order_by":4,"name":"Kexin Qin","email":"","orcid":"","institution":"Xinyang Normal University","correspondingAuthor":false,"prefix":"","firstName":"Kexin","middleName":"","lastName":"Qin","suffix":""}],"badges":[],"createdAt":"2024-06-26 12:41:20","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4642799/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4642799/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":60725636,"identity":"f4885281-bb7f-4e12-8e85-75d047ecdde4","added_by":"auto","created_at":"2024-07-20 05:40:31","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":4634290,"visible":true,"origin":"","legend":"\u003cp\u003e(a) location of Qinghai Province. (b) image, water system and highway distribution of the study area; (c) digital elevation model of the study area.\u003c/p\u003e","description":"","filename":"Fig1.png","url":"https://assets-eu.researchsquare.com/files/rs-4642799/v1/a13a7b496146151b8ee726a6.png"},{"id":60726227,"identity":"eee426b7-c2f5-4751-a832-93b5b06ed12d","added_by":"auto","created_at":"2024-07-20 05:56:31","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":282423,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of the methodology.\u003c/p\u003e","description":"","filename":"Fig2.png","url":"https://assets-eu.researchsquare.com/files/rs-4642799/v1/66227221cdd4f778a3fc9e6f.png"},{"id":60725647,"identity":"91c6be85-496c-4766-80e4-e5b715214970","added_by":"auto","created_at":"2024-07-20 05:40:31","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":5233370,"visible":true,"origin":"","legend":"\u003cp\u003eGeometric distortion map of study area\u003c/p\u003e","description":"","filename":"Fig3.png","url":"https://assets-eu.researchsquare.com/files/rs-4642799/v1/620f5e1221c2165bb99e6e22.png"},{"id":60725662,"identity":"ecda268d-fbfe-4ef3-b8f4-b5708591433d","added_by":"auto","created_at":"2024-07-20 05:48:31","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":7163751,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePS-InSAR deformation rate distribution.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"Fig4.png","url":"https://assets-eu.researchsquare.com/files/rs-4642799/v1/b1dcfe39e0ed94a70f83a094.png"},{"id":60725638,"identity":"68ae3253-5bde-4e3b-bc45-351e03f414c3","added_by":"auto","created_at":"2024-07-20 05:40:31","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":352575,"visible":true,"origin":"","legend":"\u003cp\u003e(a) De flattening interferogram and (b) coherence coefficient diagram from February 7, 2018 to March 3, 2018.\u003c/p\u003e","description":"","filename":"Fig5.png","url":"https://assets-eu.researchsquare.com/files/rs-4642799/v1/dfaed40da75d0e29a77315aa.png"},{"id":60725661,"identity":"c99bc3c4-3e10-4853-b9f0-e9ff1a26e1b0","added_by":"auto","created_at":"2024-07-20 05:48:31","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":313993,"visible":true,"origin":"","legend":"\u003cp\u003e(a) GCP point selection of SBAS-InSAR. \u003cstrong\u003e(b) \u003c/strong\u003eSBAS-InSAR deformation rate distribution.\u003c/p\u003e","description":"","filename":"Fig6.png","url":"https://assets-eu.researchsquare.com/files/rs-4642799/v1/8fd7044d1e86e03198c98cd2.png"},{"id":60726229,"identity":"e9ffa669-6065-4747-9ee2-81b91014a409","added_by":"auto","created_at":"2024-07-20 05:56:31","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":81734,"visible":true,"origin":"","legend":"\u003cp\u003eLinear relationship diagram of annual average deformation rate at the same point of track lifting.\u003c/p\u003e","description":"","filename":"Fig7.png","url":"https://assets-eu.researchsquare.com/files/rs-4642799/v1/9fbf03f0a8d5a52fed64b26a.png"},{"id":60725667,"identity":"fa0b880d-0835-47a0-90a5-1c0038ab97d3","added_by":"auto","created_at":"2024-07-20 05:48:31","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":116021,"visible":true,"origin":"","legend":"\u003cp\u003eAmplification processing image of a sample data andcorresponding mask.\u003c/p\u003e","description":"","filename":"Fig8.png","url":"https://assets-eu.researchsquare.com/files/rs-4642799/v1/3ba44134da4c181bee4d56ef.png"},{"id":60725643,"identity":"8e7353b1-9320-4933-82d2-54a4d66720c6","added_by":"auto","created_at":"2024-07-20 05:40:31","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":426374,"visible":true,"origin":"","legend":"\u003cp\u003eTest accuracy and training loss of augmented dataset.\u003c/p\u003e","description":"","filename":"fig9.png","url":"https://assets-eu.researchsquare.com/files/rs-4642799/v1/cffada7bf8260fbefa995bc9.png"},{"id":60725649,"identity":"203c77a2-df81-416c-8702-930ff9a018b2","added_by":"auto","created_at":"2024-07-20 05:40:31","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":8167327,"visible":true,"origin":"","legend":"\u003cp\u003eDistribution of Historical and Potential Landslide Points.\u003c/p\u003e","description":"","filename":"fig10.png","url":"https://assets-eu.researchsquare.com/files/rs-4642799/v1/255336b824ef6a4939055a02.png"},{"id":60726426,"identity":"1ca3541f-d26e-4996-b106-bb6e80023021","added_by":"auto","created_at":"2024-07-20 06:04:31","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":259281,"visible":true,"origin":"","legend":"\u003cp\u003eTerrain factors and potential landslide points: (a) slope (b) aspect (c) relief degree (d) lithology.\u003c/p\u003e","description":"","filename":"fig11.png","url":"https://assets-eu.researchsquare.com/files/rs-4642799/v1/21c27fa6712be8922e5bf25c.png"},{"id":60726228,"identity":"c693741a-6e51-4777-bdb3-0b567958c16f","added_by":"auto","created_at":"2024-07-20 05:56:31","extension":"png","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":92456,"visible":true,"origin":"","legend":"\u003cp\u003eStatistics of landslide development and distribution characteristics: (a) relationship between potential landslide distribution and slope, relief degree (b) relationship between potential landslide distribution and aspect.\u003c/p\u003e","description":"","filename":"Fig12.png","url":"https://assets-eu.researchsquare.com/files/rs-4642799/v1/6d4131e967c6f23e867fd334.png"},{"id":60725646,"identity":"962897b6-7967-4c93-882c-18be2f9f2c54","added_by":"auto","created_at":"2024-07-20 05:40:31","extension":"png","order_by":13,"title":"Figure 13","display":"","copyAsset":false,"role":"figure","size":543391,"visible":true,"origin":"","legend":"\u003cp\u003eRelationship between Landslide Deformation and Precipitation Monitoring.\u003c/p\u003e","description":"","filename":"fig13.png","url":"https://assets-eu.researchsquare.com/files/rs-4642799/v1/8d04a9f75aac62baa1f63ce5.png"},{"id":60725648,"identity":"b0cf52fc-38ee-47a5-a472-f915eec10b1c","added_by":"auto","created_at":"2024-07-20 05:40:31","extension":"png","order_by":14,"title":"Figure 14","display":"","copyAsset":false,"role":"figure","size":542712,"visible":true,"origin":"","legend":"\u003cp\u003eRelationship between Landslide Deformation and Temperature Monitoring.\u003c/p\u003e","description":"","filename":"fig14.png","url":"https://assets-eu.researchsquare.com/files/rs-4642799/v1/356cf7ca5d6d4ef0b11fd6e0.png"},{"id":60726632,"identity":"3c594cfd-b191-4dd5-8e82-91d8b82e67ca","added_by":"auto","created_at":"2024-07-20 06:12:47","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":38122018,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4642799/v1/6508238a-eecf-4ae0-b2d0-86d7038d6b14.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Identification of Potential Landslide in Jianzha Counctry Based on InSAR and Deep Learning","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eGeological hazards such as landslides occur frequently in Jianzha County, Qinghai Province, severely impacting the ecological environment and social stability. Consequently, it is imperative to monitor and provide early warnings for landslides, and to implement effective preventive and responsive measures. Due to various inducing factors of landslide disasters, such as river scouring, heavy rainfall, earthquake, human activities, etc., it will not only cause serious casualties, property losses, traffic interruption, environmental damage and other direct hazards, but also produce secondary disasters such as debris flow and dammed lake [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e], which greatly restricts the development and utilization of resources [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Landslides are characterized by sudden onset, high difficulty of management, and frequent occurrences in groups [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e], all of which complicate the early identification of landslide hazards. Timely, accurate, and efficient early identification of mountain landslides has become a pressing issue in the field of disaster prevention.\u003c/p\u003e \u003cp\u003eField surveys and regular updates of landslide inventory maps are conventional methods for investigating potential landslides. However, relying solely on traditional manual field surveys or expert interpretation introduces subjective biases and consumes considerable resources. Additionally, these methods exhibit inefficiencies and inaccuracies in identifying and monitoring potential or subtly moving landslides [\u003cspan additionalcitationids=\"CR5\" citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e], making widespread implementation challenging. In contrast, optical remote sensing data offer extensive coverage and high identification accuracy, enabling the identification of landslide locations and areas with significant deformations, thereby improving the accuracy of potential landslide monitoring. Consequently, landslide monitoring using optical imagery has gained significant attention. Nevertheless, optical remote sensing data alone are insufficient for detecting potential landslides with minor deformations and no destructive hazards [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eRadar data exhibit features of continuous operation, all-weather capability, and high precision. Research has illustrated their expansive application prospects and growth potential in surface deformation detection. Interferometric synthetic aperture radar (InSAR) techniques, a novel spatial surface measurement technology, remains unaffected by weather conditions and offers benefits such as high precision, low cost, extensive coverage, and continuous operation, rendering it widely adopted for automated landslide hazard monitoring. Persistent Scatterer InSAR (PS-InSAR) and small baseline subset InSAR (SBAS-InSAR) are commonly used in long time series InSAR Technology [\u003cspan additionalcitationids=\"CR9 CR10 CR11 CR12 CR13 CR14 CR15 CR16 CR17\" citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. PS-InSAR mitigates temporal-spatial baseline self-coherence and atmospheric interference in surface deformation monitoring over extended periods. SBAS-InSAR inherits the advantages of D-InSAR technology and enables the acquisition of extensive and contiguous surface deformation information. Relying solely on a single surface deformation monitoring technique for potential landslide hazard identification entails uncertainties. Integrating both techniques can enhance identification outcomes to a certain extent. Elevation data are utilized in InSAR data processing for interferogram unwrapping, image registration, geocoding, etc., and are also employed for extracting subsequent terrain factors to further delineate landslide-prone areas. Utilizing multi-source remote sensing technology to study the dynamic changes in the development of landslide-type debris flows [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Wu [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e] utilized InSAR technology and optical remote sensing for early identification and monitoring of landslides in Guizhou Province, a crucial step in improving the area's geological disaster prevention and control capabilities. Piroton [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e] utilized a combination of drone-based imagery, radar, and optical remote sensing technologies to detect terrain elevation changes associated with rapid and slow surface displacements, along with meteorological analyses, to identify triggering conditions leading to slope instability. Casagli [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e] described the application of RSTS in landslide analysis and management, and the analysis showed that remote sensing technology has great potential in landslide detection, monitoring and prediction. Bouali [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e] utilized Synthetic Aperture Radar (SAR), optical remote sensing, and Global Positioning System (GPS) to quantify the incremental and average deformation of the Portuguese Bend landslide in California from 2007\u0026ndash;2017.\u003c/p\u003e \u003cp\u003eWhile machine learning methods are widely applied in landslide modeling due to their simplicity and strong interpretability [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e], their limited capability to explore correlations between input variables hampers their ability to extract deep features from the data. Furthermore, these models lack the autonomy to perform feature learning [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. Different from traditional machine learning methods, Convolutional Neural Networks (CNNs) are machine learning model containing multiple convolution layers and pooling layers, which reflects the structure of human brain neural network. By emulating the neural networks of the human brain, CNNs can comprehend learning objectives and automatically analyze data, facilitating the identification of landslide-prone areas in larger scenes without the requirement for visual interpretation [\u003cspan additionalcitationids=\"CR28 CR29 CR30 CR31 CR32 CR33\" citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. Through extensive training of a large number of data sets, deep features can be obtained from complex data.\u003c/p\u003e \u003cp\u003eCurrently, the identification of landslide disasters through intelligent means has become a prevailing trend. However, there is a lack of deep neural network models specifically trained for joint InSAR technology in landslide image recognition. It is crucial to leverage multi-source databases and intelligent methods to construct high-precision models tailored for potential landslide disaster identification in this region, thereby enabling the intelligent identification of large-scale landslide distribution.\u003c/p\u003e \u003cp\u003eThis study, grounded in terrain visibility analysis, aims to integrate PS-InSAR, SBAS-InSAR, and optical remote sensing technologies to delineate potential landslide areas. Subsequently, by employing Google imagery in conjunction with deep learning models, it seeks to identify potential landslide points. Finally, the study integrates terrain, precipitation, and temperature data to analyze the distribution patterns and underlying causes of landslide development. This research provides novel approaches for the large-scale, automated early identification of mountain landslides and landslide susceptibility mapping, significantly contributing to disaster prevention and mitigation efforts.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"2. Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1. Study Area\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eJianzha county (101\u0026deg;38\u0026prime;-102\u0026deg;06\u0026prime;E, 35\u0026deg;40\u0026prime;-36\u0026deg;10\u0026prime;N) is located in the southeast of Qinghai Province, with a total area of 1714km2 and an altitude of 1995-4263m. It is located in the transition zone between the Qinghai Tibet Plateau and the Loess Plateau in China [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]. The Yellow River traverses the county from north to south, extending 96 kilometers within the county boundaries, and exhibits multiple river terraces and basin-hill landforms, characterized by complex geological conditions. The terrain of the study area is high in the South and North, low in the middle, high in the West and low in the East. The unique geological conditions, landform environment, climatic influences, and issues of land desertification in the upper reaches of the Yellow River have led to the development of numerous landslide hazards in this region. Landslide hazards in Jianzha County are predominantly found near the Yellow River, particularly in Kanbula, Nengke, Cuozhou, and Jianzha Township. The study area in Jianzha County is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2. Data Collection and Processing\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe Sentinel-1A data utilized in this study were sourced from NASA ASF (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://search.asf.alaska.edu/\u003c/span\u003e\u003cspan address=\"https://search.asf.alaska.edu/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), while the Precise Orbit Data (POD) were procured from the Copernicus Data Space Ecosystem (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://dataspace.copernicus.eu/\u003c/span\u003e\u003cspan address=\"https://dataspace.copernicus.eu/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). A dataset comprising 60 monthly images, spanning from January 2018 to December 2022, was selected as the primary data source. All data were acquired along the same ascending orbit, featuring a spatial resolution of 5m\u0026times;20m, a 12-day revisit period, C-band (wavelength 5.6 cm), an incidence angle of 37.13\u0026deg;, and the Interferometric Wide (IW) swath imaging mode with VV polarization. The precise orbit data effectively eliminate the orbital errors associated with Sentinel-1A. High-resolution Google satellite images (3m spatial resolution) from 2020 to 2022 were employed to construct a landslide recognition model using a deep learning network, thus providing substantial support for feature learning with an extensive landslide sample dataset. The dataset used for deep learning examples is the Bijie_Landslide_Dataset, curated by Wuhan University. The dataset data is obtained from TripeSat satellite data with 0.8m resolution in panchromatic band and 3.2m resolution in multispectral band. This study primarily utilizes 770 landslide images and their corresponding binary mask data derived from satellite imagery. In these masks, landslide sample values are uniformly set to 0, while non-landslide sample values are set to 1.\u003c/p\u003e \u003cp\u003eThe SRTM DEM was obtained from USGS (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://earthexplorer.usgs.gov/\u003c/span\u003e\u003cspan address=\"https://earthexplorer.usgs.gov/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), utilizing SRTM1 data with a spatial resolution of 30m as elevation reference data. These data were used for geographic coordinate reference in surface deformation and extraction of environmental factors influencing landslide development. Historical landslide validation data were sourced from the Spatial Distribution of Geological Disaster Points dataset provided by the Chinese Academy of Sciences' Resource and Environmental Science and Data Center (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.resdc.cn/\u003c/span\u003e\u003cspan address=\"https://www.resdc.cn/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). Lithological data were obtained from ISRIC (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.isric.org/\u003c/span\u003e\u003cspan address=\"https://www.isric.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) for the lithological analysis of landslide hazard point distribution. Precipitation and temperature data were obtained from National Earth System Science Data Center (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.geodata.cn/\u003c/span\u003e\u003cspan address=\"https://www.geodata.cn/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) with a spatial resolution of 1km. The possible effects of precipitation and air temperature on landslide surface deformation were investigated using monthly data from 2018 to 2022.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3. Methodology\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThis paper proposes a method for identifying potential landslide points based on InSAR technology and deep learning techniques. The main approach is as follows: First, the visibility of the area is analyzed by combining SAR imagery and DEM data to verify the feasibility of InSAR technology. Secondly, the PS-InSAR technology and SBAS-InSAR technology were used to delineate a large area of surface deformation from sentinel-1a images of 60 scenes in Jianzha county from January 2018 to December 2022. Given the close correlation between surface deformation areas and potential landslide hazards, deep learning and visual interpretation are applied to Google imagery to identify potential landslide distribution points within these areas. Finally, the topographical features of landslide development in the region are extracted using known landslide points. The topographical characteristics of historical landslide points serve as the basis for assessing landslide hazard levels. Points matching InSAR surface deformation characteristics, deep learning identification features, and topographical attributes are classified as high-hazard landslide points, while the remaining points are categorized as medium-hazard landslide points. Finally, the factors contributing to landslide deformation are analyzed by integrating topographical, precipitation, and temperature data. The technical workflow is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e:\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec6\" class=\"Section3\"\u003e \u003ch2\u003e2.3.1. InSAR deformation monitoring\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eSynthetic Aperture Radar Interferometry (InSAR) is an advanced space-to-earth observation technology. The radar system interferes with the changing phase information generated by more than two multi-temporal SAR images in the same area, and finally can obtain a large range, high precision and high resolution surface deformation information in the time series [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. Due to the side-looking imaging of SAR sensors, the incident angle formed when the radar hits ground objects affect the image quality. As the incident angle decreases, the image's echo signal is enhanced, resulting in brighter pixels [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]. This unique imaging method makes the radar images highly susceptible to displaying anomalies. When the beam is directed to the slope, because the irradiation distance of the bottom surface is larger than that of the top, the top is imaged before the bottom, resulting in an inverted overlay phenomenon. When the slope changes greatly, the beam is difficult to illuminate the back slope, so that the sensor is difficult to receive the reflection of the ground object on the back slope, and the shadow area will be shown on the map. In mountainous regions with significant relief degree, these phenomena will become more pronounced, potentially leading to visual blind spots. Perspective foreshortening refers to the distortion of object size and shape in the image due to the spatial relationship between the sensor and the ground, affecting the accurate measurement of distance and dimensions. To mitigate the impact of these phenomena on landslide point identification, a visibility analysis of the area must be conducted prior to detecting potential landslides using PS-InSAR and SBAS-InSAR technologies.\u003c/p\u003e \u003cp\u003eThe fundamental principle of PS-InSAR involves statistically analyzing the amplitude information of radar images from the same area and identifying permanent scatterers that remain unaffected by temporal and spatial baseline decorrelation and atmospheric delays to perform phase modeling and deformation calculation [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. For N\u0026thinsp;+\u0026thinsp;1 SAR images of the study area, the primary image is selected based on image quality and the distribution of spatial and temporal baselines, and differential interferometric processing is performed on the registered images to generate N interferograms. By analyzing the differences in Permanent Scatterer (PS) points across different image pairs, the deformation rate and elevation error of each PS point relative to the main reference point can be calculated, accounting for factors such as atmospheric effects, orbital errors, and surface deformation. After obtaining the differential interferometric phase, phase unwrapping is necessary to extract the actual physical displacement information. The unwrapped linear phase residuals can be used to calculate the actual surface deformation. However, this technique may result in poor coherence for some interferometric pairs due to long spatial and temporal baselines when selecting one image as the common primary image and using the remaining images as secondary images. In mountainous regions with significant relief degree land surface and low coherence, supplementary techniques are required to validate the deformation monitoring results of PS-InSAR.\u003c/p\u003e \u003cp\u003eSBAS-InSAR is an InSAR time series method based on multiple master images. Its fundamental principle involves calculating the spatial-temporal baselines of images from different times within the region using the short baseline principle, selecting appropriate spatial-temporal thresholds to form interferograms, and performing multi-look processing to reduce noise. Using the singular value decomposition (SVD) method, the spatial small baseline subset data are combined into a time series to compute the least squares solution within the subset [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]. Finally, the residual phase is used to invert the atmospheric phase and non-linear phase to derive the deformation time series for the given time phase. Compared to PS-InSAR, SBAS-InSAR can further mitigate the effects of temporal and spatial decorrelation, thus obtaining higher precision deformation information. If the study area has N\u0026thinsp;+\u0026thinsp;1 SAR images, the registered images are arranged in chronological order and divided into several sets based on spatial-temporal baselines for differential interferometric processing, resulting in M interferograms. The resulting M interferograms must satisfy the constraint:\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Equ1\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e\n$$\\frac{\\text{N+1}}{\\text{2}}\\text{\u0026le;M\u0026le;}\\frac{\\text{N(N+1)}}{\\text{2}}$$\u003c/div\u003e \u003cdiv class=\"EquationNumber\"\u003e1\u003c/div\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe expression for the interferometric phase of the i-th interferogram pair is:\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Equ2\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ2\" name=\"EquationSource\"\u003e\n$${\\text{\u0026phi;}}_{\\text{i}}\\text{=}{\\text{\u0026phi;}}_{\\text{topo}}\\text{+}{\\text{\u0026phi;}}_{\\text{flat}}\\text{+}{\\text{\u0026phi;}}_{\\text{orb}}\\text{+}{\\text{\u0026phi;}}_{\\text{def}}\\text{+}{\\text{\u0026phi;}}_{\\text{atm}}\\text{+}{\\text{\u0026phi;}}_{\\text{scat}}\\text{+}{\\text{\u0026phi;}}_{\\text{noise}}$$\u003c/div\u003e \u003cdiv class=\"EquationNumber\"\u003e2\u003c/div\u003e\u003c/div\u003e \u003cdiv id=\"Equ3\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ3\" name=\"EquationSource\"\u003e\n$${\\text{\u0026phi;}}_{\\text{topo}}\\text{=-}\\frac{\\text{4\u0026pi;}{\\text{B}}_{\\text{┴}}\\text{h}}{\\text{\u0026lambda;Rsin\u0026theta;}}$$\u003c/div\u003e \u003cdiv class=\"EquationNumber\"\u003e3\u003c/div\u003e\u003c/div\u003e \u003cdiv id=\"Equ4\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ4\" name=\"EquationSource\"\u003e\n$${\\text{\u0026phi;}}_{\\text{flat}}\\text{=-}\\frac{\\text{4\u0026pi;}{\\text{B}}_{\\text{┴}}}{\\text{\u0026lambda;}}$$\u003c/div\u003e \u003cdiv class=\"EquationNumber\"\u003e4\u003c/div\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eIn the equation, φ\u003csub\u003etopo\u003c/sub\u003e represents the topographic phase; φ\u003csub\u003eflat\u003c/sub\u003e represents the flat-earth phase; φ\u003csub\u003eorb\u003c/sub\u003e represents the orbital error phase; φ\u003csub\u003edef\u003c/sub\u003e represents the deformation phase, which includes both deformation and non-deformation components; φ\u003csub\u003eatm\u003c/sub\u003e represents the atmospheric error phase; φ\u003csub\u003escat\u003c/sub\u003e represents the phase due to changes in point target scattering characteristics; and φ\u003csub\u003enoise\u003c/sub\u003e represents the noise phase. B\u003csub\u003e┴\u003c/sub\u003e represents the perpendicular baseline length, h represents the elevation error, λ represents the radar wavelength, R represents the slant range, and θ represents the incidence angle.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section3\"\u003e \u003ch2\u003e2.3.2. Landslide recognition based on deep learning\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThis study utilizes a deep learning framework based on the open-source Python machine learning library PyTorch to build and train a ResNet50 network model for learning landslide features. After extracting classification information, the trained network is validated using test data. Landslide classification divides the sample set into landslide and non-landslide targets, i.e., a binary classification, and learns to delineate landslide boundaries. During the adjustment of learning rate and loss weight, the optimal parameter combination is iteratively updated to obtain the best-trained network. Finally, the landslide recognition model is evaluated using binary classification metrics: Precision, Accuracy, and Recall. Precision represents the accuracy of the positive part predicted by the model; Accuracy is the accuracy of the model. The higher the accuracy, the better the model effect; Recall rate is the correct proportion predicted by the model. The calculation formulas are as follows:\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Equ5\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ5\" name=\"EquationSource\"\u003e\n$$\\text{Precision=TP/(TP+FP)}$$\u003c/div\u003e \u003cdiv class=\"EquationNumber\"\u003e5\u003c/div\u003e\u003c/div\u003e \u003cdiv id=\"Equ6\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ6\" name=\"EquationSource\"\u003e\n$$\\text{Accuracy=(TP+TN)/(TP+TN+FP+FN)}$$\u003c/div\u003e \u003cdiv class=\"EquationNumber\"\u003e6\u003c/div\u003e\u003c/div\u003e \u003cdiv id=\"Equ7\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ7\" name=\"EquationSource\"\u003e\n$$\\text{Recall=TP/(TP+FN)}$$\u003c/div\u003e \u003cdiv class=\"EquationNumber\"\u003e7\u003c/div\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eWhere True Positive (TP) represents the number of samples correctly predicted as positive by the model; False Positive (FP) represents the number of samples incorrectly predicted as positive by the model; True Negative (TN) represents the number of samples correctly predicted as negative by the model; and False Negative (FN) represents the number of samples incorrectly predicted as negative by the model.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Visibility analysis\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eSpaceborne radar satellites, due to their side-looking imaging mode, often result in images with layover, shadow, and foreshortening effects when observing the ground using radar beams. Based on elevation data of the study area, the county exhibits significant elevation differences, with variations reaching over 2000 meters. The dramatic topographical relief makes the area highly susceptible to tropospheric atmospheric delays when monitored using InSAR technology. In this region, the satellite incidence angle is 37.13\u0026deg; [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. The visibility classification standards are based on Sentinel-1A's incidence angle and topographical factors, dividing the area into three categories: non-visibility, low sensitivity, and high visibility. Non-visibility areas are primarily located on slopes facing east, southeast, and northeast. When the slope angle is less than 37.13 \u0026deg;, perspective shrinkage will occur, and when the slope angle is greater than 37.13 \u0026deg;, overlap will occur. Visibility areas include high visibility and low sensitivity regions. High visibility areas are mainly on slopes facing west, southwest, and northwest, where shadowing occurs when the slope angle exceeds 52.87\u0026deg;. Low sensitivity areas are primarily on the south and north slopes, which are less sensitive to surface deformation. The classification of visibility types and the geometric distortion areas of the study region are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. From the mapping results, visibility areas cover 1252.42 km\u0026sup2;, accounting for 73.07% of the area. Layover and shadow areas are sparse, covering only 6.14 km\u0026sup2; (0.36%). Foreshortening areas are mostly on southeast-facing back slopes, covering 455.41 km\u0026sup2; (26.57%). Historical landslide point distributions reveal that landslides frequently occur in valleys and low mountain hills near the Yellow River. These areas are minimally affected by layover, shadow, and foreshortening, indicating that the selected ascending radar imagery has good visibility and high reliability for landslide monitoring.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.2 PS-InSAR deformation monitoring results\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe image from January 4, 2020, was selected as the primary image, with the remaining 59 images serving as secondary images. These were used to generate SAR data pairs and connection graphs for subsequent differential interferometry processing. The temporal baseline between the primary image and all secondary images ranges from \u0026minus;\u0026thinsp;133\u0026ndash;131d, and the spatial baseline ranges from \u0026minus;\u0026thinsp;92\u0026ndash;131m, both below the critical baseline. The longest temporal baseline is 133 days, corresponding to the image from June 10, 2022. PS-InSAR involves processing each interferometric pair individually for registration and interferometry, resulting in unwrapped phase maps of the residual phase. During registration, the secondary images are aligned with the primary image, with a range-to-azimuth ratio set at 4:1. After completion, quick-look images are examined to verify the registration and interferometry results of all pairs. Interferometric pairs with poor unwrapping or coherence are excluded, ensuring that only correctly processed pairs are used. The first inversion method automatically selects reference points with minimal deformation and high coherence, and then analyzes the phase changes over the time series to obtain displacement rates and residual elevation. The second inversion converts the phase shifts from the first inversion results into deformation information in the geographic coordinate system, yielding the final deformation rates. The deformation data from the first and second inversions are geocoded to produce a PS point vector file, which is then interpolated to generate a point target deformation rate map (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe line-of-sight (LOS) deformation rate in the study area, as measured by PS-InSAR technology, ranges from \u0026minus;\u0026thinsp;68\u0026thinsp;\u0026minus;\u0026thinsp;28 mm/a. From the deformation rate distribution map, it is observed that the PS-InSAR deformation monitoring area is primarily located in relatively flat valley areas near the Yellow River channel. Fewer valid PS points are extracted from densely vegetated areas, valleys with foreshortening effects, and mountainous regions, making it challenging to assess surface deformation. Significant subsidence is observed in the northern and eastern directions in towns such as Kanbula, Kangyang, Dangshun, and Maketang, while localized uplift is evident in Cuozhou Township. These contrasting phenomena arise due to the concentrated rainfall in the study area, leading to extensive overall subsidence caused by water erosion. Additionally, the expansion of human activities prompts the construction of production-friendly structures near residential areas, resulting in scattered local uplift.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.3 SBAS-InSAR deformation monitoring results\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eTo ensure the highest quality of interferograms and improve result accuracy, the maximum temporal baseline threshold was set to 120d, and the maximum spatial baseline threshold to 2%. The image from August 7, 2020, was selected as the super master image, with the remaining 59 images serving as secondary images, resulting in 228 freely combined interferometric pairs. The temporal baseline between the super master image and all secondary images ranges from \u0026minus;\u0026thinsp;116\u0026ndash;106d, and the spatial baseline ranges from \u0026minus;\u0026thinsp;110\u0026ndash;152m. Image pairs with short temporal and spatial baselines undergo SBAS-InSAR inversion after interferometric processing. Statistical analysis of the connected pairs reveals that the maximum number of pairs for a single image is 11, while the minimum is 1, which is sufficient to meet the landslide identification requirements for Jianzha County. The SBAS-InSAR interferometric processing can flatten, filter, and unwrap the phase of the image pairs. Similarly, set the ratio of distance direction to azimuth direction to 4:1. Subsequently, the topographic phase is removed based on DEM data, and a polynomial model (Goldstein) is employed for filtering. Because the study area is a mountainous area with low coherence, 3D unwrapping is not carried out, and the minimum cost flow is selected. Finally, interference pairs with poor unwrapping effect and coherence are eliminated. The results are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eUsing the selected control points as a reference, residual phase and phase ramps remaining after unwrapping are removed, and finally, the unwrapped phase is converted into elevation or deformation values. The introduction of precise ground control points can effectively remove residual constant phase and flat-earth effect, thereby enhancing the validity of the results. The selection of ground control points should first ensure uniform coverage over a wide area with high coherence, good unwrapping results, and stability; residual topographic areas, phase jump regions, and deformation stripe areas should be avoided as much as possible [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. Since the GCP files obtained from PS-InSAR cannot be directly used for SBAS-InSAR processing, automatically selected and geocoded GCP points from the PS-InSAR technique are chosen here. This reduces subjective errors in selecting control points manually in the SBAS-InSAR method while enhancing the accuracy of deformation estimation. The distribution of GCP points is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e(a). The distribution of GCP points indicates that the 266 selected GCP targets are uniformly distributed within the study area and are mostly located in high-coherence urban regions, which aligns with the actual conditions of the study area. The deformation data obtained from SBAS-InSAR is geocoded to produce a deformation rate map (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e(b)).\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eApart from the mountainous regions with significant topographic relief in the southwest and northwest of Jianzha County, the coverage extracted using this monitoring method shows a significant improvement compared to the deformation rate results of PS-InSAR. Deformation information can be extracted even in some mountainous areas. The LOS deformation rate of SBAS-InSAR ranges from \u0026minus;\u0026thinsp;62\u0026thinsp;\u0026minus;\u0026thinsp;37 mm/a. Using SBAS-InSAR technology, it is found that, in addition to the slight uplift observed in Cuozhou Township, local uplift also occurs in Maketang Township and Angla Township. The reason is that the number of permanent scatterers extracted by the PS-InSAR technology is limited by the environment, resulting in a limited amount of deformation data obtained [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e], making it difficult to meet the requirements for extracting large-area continuous surface deformation. In contrast, SBAS-InSAR compensates for this deficiency, making it more effective for extracting surface deformation information within the study area. Therefore, subsequent surface deformation information will primarily utilize SBAS-InSAR technology, supplemented by surface subsidence data extracted using PS-InSAR, to better identify potential landslide hazards.\u003c/p\u003e \u003cp\u003eTo evaluate the reliability of the combined results from PS-InSAR and SBAS-InSAR technologies, 400 identical points were selected from the annual average deformation rates obtained using Sentinel-1A data for both technologies. A statistical analysis of the linear relationship between the annual average deformation rates (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e) reveals that the deformation rate distribution of identical points in both technologies is essentially consistent, with an R\u0026sup2; greater than 0.91. This indicates a high correlation between the two technologies, demonstrating the validity of combining the results from PS-InSAR and SBAS-InSAR technologies.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.4 Landslide recognition results\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eDue to the limited number of samples in the landslide dataset, and considering that convolutional neural networks require a large number of samples to extract feature information. To improve the accuracy of landslide sample recognition and achieve the best recognition results, data augmentation of the landslide samples is necessary. The samples in the data set were randomly rotated 90\u0026deg;, 180\u0026deg; and 270\u0026deg;, horizontally mirrored and vertically mirrored to increase the quality of the landslide data set. The image data after data amplification is 3850 pieces, and then the training set and the verification set are randomly selected according to the ratio of 7:3. Select any landslide sample to show the effect of data expansion process as shown in the figure. The Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e shows the processing examples of original sample, rotation 270\u0026deg;, rotation 270\u0026deg;+ horizontal flip, vertical mirroring, and 90\u0026deg; rotation. The landslide prediction samples are created using Google satellite images with a spatial resolution of 3 meters. The Google images of Jianzha County are preprocessed by cropping them into 256\u0026times;256 pixels. The landslide hazard areas in the prediction samples are then identified and annotated sequentially, and finally, the annotated samples are merged to obtain the landslide prediction distribution map of the study area.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eInfluenced by the topography, the surface deformation is mostly in the negative direction of LOS and has a large deformation rate, which is very easy to cause landslide disasters. Based on the surface deformation information extracted using the aforementioned PS-InSAR and SBAS-InSAR technologies, areas with significant deformation, reaching the threshold of 20 mm/a in the LOS direction, are delineated as landslide hazard zones. By integrating Google Earth images, the areas with significant deformation from 2020 to 2022 are summarized to obtain the distribution of landslide hazard zones. The model training is set to 1500 iterations, with a batch size of 32 and an initial learning rate of 0.0001. For the prediction part of the model, the potential landslide areas delineated in the Google images are cropped into 256\u0026times;256 pixel files. The landslide boundaries are then mapped on the subdivided local images.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eCombining the augmented dataset, the test accuracy curve and training loss curve (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e) reveal that the test accuracy of the model rapidly increases from 43.6% at the beginning of training, reaching 85% for the first time after 80 iterations. The fluctuation amplitude decreases after 800 iterations, stabilizing around 91% after 1500 iterations. Training loss decreases from 0.52 at the beginning of training and stabilizes at 0.0557 after 1500 iterations. Considering both the test accuracy and iteration number as evaluation metrics for the weight files, the 770th weight file (test accuracy 94.4%, test precision 98.7%, recall 84.2%) is selected for subsequent landslide hazard monitoring. A total of 56 potential landslide areas are identified. As of 2020, 10 of the 13 historical landslide disaster points are located at the identified high hidden danger landslide points, and 3 landslides are missed, with the monitoring accuracy of 76.92%. For easier statistical analysis, the potential landslide hazard points are represented as dots. The distribution of historical and potential landslide monitoring points is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e. The topographic features of historical landslides are extracted to assess the 56 identified landslide hazard points, revealing 39 high-risk and 17 medium-risk landslide points.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003e \u003cb\u003eInfluence of topography on landslide distribution\u003c/b\u003e \u003c/p\u003e \u003cp\u003eThe frequent occurrence of landslides is closely related to geological structures. By integrating slope topography factors, the environmental elements for landslide development are extracted, including slope, aspect, relief degree, and lithology, to analyze the control conditions of the landslide development environment (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e). Since steep slopes increase the risk of soil or rock instability, combined with the differences in slope topography, humidity, and vegetation cover due to different slope aspects, slope angle is often considered an important factor in landslide prevention and mitigation planning. Using the equal interval method, the slope is divided into six categories to count the number of landslides: 0\u0026ndash;10\u0026deg;, 10\u0026ndash;20\u0026deg;, 20\u0026ndash;30\u0026deg;, 30\u0026ndash;40\u0026deg;, 40\u0026ndash;50\u0026deg;, and above 50\u0026deg;. Aspect determines the wind direction, precipitation, surface water flow direction, and vegetation growth, which all affect slope stability. To visually display the relationship between landslides and aspect, the slope aspects are categorized into eight directions for statistical analysis. Relief degree refers to the vertical distance between the highest and lowest points within a specific area and is typically used to measure the unevenness of the terrain, reflecting the slope characteristics that foster landslide occurrences. Statistics reveal that the relief degree of landslide disaster points in this area is concentrated between 0 and 90 meters. Using the equal interval method, the relief degree is divided into six levels: 0-15m, 15-30m, 30-45m, 45-60m, 60-75m, and 75-90m. Research indicates that the distribution of landslides is inseparable from the characteristics of rock and soil masses. The rock or soil types and structures in landslide areas differ significantly from those in surrounding non-sliding slopes. Previously sliding rock layers or soils usually have a looser structure, making areas with frequent fault activities prone to landslides. Therefore, lithology is selected as one of the topographic factors for analyzing landslide development.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eAnalyzing the development and distribution characteristics of potential landslide points using slope terrain factors. As illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig12\" class=\"InternalRef\"\u003e12\u003c/span\u003e(a), potential landslides are predominantly distributed within the 10\u0026ndash;40\u0026deg; slope range, comprising 76.8% of the total. This indicates that there is no significant positive correlation between landslide distribution and slope steepness. Landslides are more likely to occur on natural slopes where environmental factors such as wind direction, precipitation, lithology, and soil quality are favorable. The relief degree statistics indicate that the relief degree in this area is significantly influenced by lithology. Potential landslide points are predominantly distributed at elevations of 15-45m, primarily in areas where rainfall and unsound engineering activities have caused environmental damage. 98.21% of landslide points exhibit an relief degree of less than 60m, and as relief degree increases above 60m, there is no observed trend of worsening landslide severity. Lithological statistics indicate that landslide points are predominantly found in areas composed of sandstone, mudstone, feldspathic sandstone, and aeolian deposits. These lithological formations belong to a group prone to sliding and, upon contact with water, form impermeable layers within the slope, thereby creating potential sliding surfaces. The statistical results of slope aspect are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig12\" class=\"InternalRef\"\u003e12\u003c/span\u003e (b). It is found that potential landslides are concentrated on the northeast and Southeast slopes, and the probability of occurrence of the facing slope is significantly higher than that of the back slope. Combining the analysis of slope aspect characteristics, it is highly likely that the terrain causes windward slopes to receive more rainfall than leeward slopes, thus reducing slope stability and loosening the soil, making precipitation a dominant factor in landslide occurrence. Additionally, the study area's topography, characterized by higher elevations in the west and lower elevations in the east, provides natural east-facing slopes with effective free faces, thereby making landslides more likely to occur.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eFrom the observed landslide points, the aforementioned four types of terrain factors were extracted as potential landslide development elements. The thresholds for landslide-prone factors were identified as follows: elevation between 2066-2839m for river terraces and low hills, slope gradient of 5.19\u0026ndash;39.78\u0026deg;, aspect of 39.69-358.53\u0026deg;, surface relief degree of 5-54m, with no significant differentiation in lithological factors. Therefore, it can be concluded that landslides are more likely to occur when the aforementioned slope terrain factors fall within the appropriate threshold ranges, combined with environmental triggers such as rainfall. Not all potential landslide points are equally prone to landslide disasters.\u003c/p\u003e \u003cp\u003e \u003cb\u003eImpact of precipitationand temperature\u003c/b\u003e \u003c/p\u003e \u003cp\u003ePrecipitation is a critical factor influencing landslide deformation [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]. Due to the influence of terrain and the southeast monsoon, precipitation is higher in the southwestern part of Jianzha County, characterized by mid- and high-mountain areas. In 2018, Jianzha County received an annual precipitation of 570.9mm, with summer precipitation averaging 323.7mm, comprising 56.7% of the annual total. The precipitation in spring and autumn were 101.5mm and 138.7mm respectively. The precipitation in winter is only 7mm. Synthesis of the monitoring data from the past five years indicates that the average downward deformation rates at the landslide points are consistently 0.17mm\u0026bull;(30d)\u003csup\u003e-1\u003c/sup\u003e, 0.79mm\u0026bull;(30d)\u003csup\u003e-1\u003c/sup\u003e, 3.71mm\u0026bull;(30d)\u003csup\u003e-1\u003c/sup\u003e, 3.04mm\u0026bull;(30d)\u003csup\u003e-1\u003c/sup\u003e, -7.05mm\u0026bull;(30d)\u003csup\u003e-1\u003c/sup\u003e, 6.09mm\u0026bull;(30d)\u003csup\u003e-1\u003c/sup\u003e, 6.52mm\u0026bull;(30d)\u003csup\u003e-1\u003c/sup\u003e, 4.04mm\u0026bull;(30d)\u003csup\u003e-1\u003c/sup\u003e, and 5.86mm\u0026bull;(30d)\u003csup\u003e-1\u003c/sup\u003e respectively. From July to September 2018, the deformation rates at P2, P3, P4, and P7 reached 6-10mm\u0026bull;(30d)\u003csup\u003e-1\u003c/sup\u003e, which was much higher than the average deformation rate. From June to September 2019, the deformation rates at P4, P6, P8, and P9 reached 3-5mm\u0026bull;(30d)\u003csup\u003e-1\u003c/sup\u003e. From August to September 2020, the deformation rates at P6, P7, and P9 reached 3-5mm\u0026bull;(30d)\u003csup\u003e-1\u003c/sup\u003e. From August to September 2021, the deformation rate at P4 reached 3mm\u0026bull;(30d)\u003csup\u003e-1\u003c/sup\u003e. From June to September 2022, P1, P3, P4, and P5 experienced sudden deformation changes, reaching 4-7mm\u0026bull;(30d)\u003csup\u003e-1\u003c/sup\u003e (Fig.\u0026nbsp;\u003cspan refid=\"Fig13\" class=\"InternalRef\"\u003e13\u003c/span\u003e). These peak deformation periods coincided with times of higher annual precipitation, indicating a correlation between landslides and rainfall, and suggesting that areas with past landslides are at risk of recurring events.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eJianzha County is characterized by a plateau cool temperate semi-arid climate, with frequent spring droughts, hail, and frost events, large diurnal temperature variations, long sunshine duration, and intense solar radiation. The annual average temperature in the region is 7.9\u0026deg;C, with an extreme annual maximum temperature of 34.1\u0026deg;C, and annual sunshine duration totaling 4432.3 hours. The relationship between monitoring points and temperature indicates that from October 2018 to January 2019, the deformation rates at P1, P3, P4, and P6 reached 3-6mm\u0026bull;(30d)\u003csup\u003e-1\u003c/sup\u003e. From October 2019 to January 2020, the deformation rates at P4 and P7 reached 4-7mm\u0026bull;(30d)\u003csup\u003e-1\u003c/sup\u003e. From October 2020 to January 2021, the deformation rates at P1 and P5 reached 2-7mm\u0026bull;(30d)\u003csup\u003e-1\u003c/sup\u003e, significantly higher than the average deformation rates at these points (Fig.\u0026nbsp;\u003cspan refid=\"Fig14\" class=\"InternalRef\"\u003e14\u003c/span\u003e). These periods of significant deformation coincide with the winter months, characterized by low precipitation and large temperature variations, indicating that, in addition to rainfall, temperature changes also affect the deformation rates of landslides.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"5. Conclusions","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThis paper combines InSAR technology with deep learning methods, utilizing InSAR technology to monitor surface deformation and convolutional neural networks to identify potential landslides in Jianzha County based on known landslide characteristics. The main conclusions are as follows:\u003c/p\u003e \u003cp\u003e(1) PS-InSAR and SBAS-InSAR technologies exhibit a high correlation. Time series analysis of extracted feature points confirms that the surface deformation trends detected by both InSAR technologies are consistent. In this study, SBAS-InSAR technology is primarily used for surface deformation information, supplemented by surface subsidence data extracted by PS-InSAR, allowing for better identification of potential landslide hazards.\u003c/p\u003e \u003cp\u003e(2) Based on the deep learning model and combined with existing landslide data, potential landslide points can be quickly and accurately identified, particularly the 39 high-risk landslide points, which have a high probability of experiencing actual landslide events in a short period.\u003c/p\u003e \u003cp\u003e(3) Landslide points in Jianzha County are concentrated on the northeast and southeast slopes with angles ranging from 10\u0026deg; to 40\u0026deg;, and the relief degree is concentrated in river terraces and low hilly areas with elevations ranging from 15 to 30m. The left bank of the Yellow River is a key area for landslide occurrences. The formation of landslides is also related to precipitation and temperature.\u003c/p\u003e \u003cp\u003eThese studies demonstrate that combining InSAR technology with deep learning methods can quickly and accurately identify early potential landslide points, providing scientific evidence for landslide monitoring and management. Future research should focus on improving the landslide dataset, particularly by incorporating historical landslide data from the study area, to make the landslide identification results more targeted.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets generated and analyzed during the current study are not publicly available due to the funding responsibility but are available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe research was sponsored by the National Natural Science Foundation of China (No. 41701449 and No. 41930102), Nanhu Scholars Program for Young Scholars of XYNU. Postgraduate Education Reform and Quality Improvement Project of Henan Province (HNYJS2020JD14).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eConceptualization, X.Y. and D.C.; methodology, X.Y. and D.C.; software, Y.D.; validation, Y.X.; formal analysis, K.Q.; investigation, Y.D.; resources, Y.X. and K.Q; data curation, D.C.; writing\u0026mdash;original draft preparation, D.C.; writing\u0026mdash;review and editing, X.Y.; visualization, D.C.; project administration, X.Y. All authors have read and agreed to the published version of the manuscript. All authors agree to submit this manuscript and affirm that it has not been published or submitted to any other journal.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflicts of Interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no conflict of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAdditional information\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCorrespondence and requests for materials should be addressed to X.Y.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eFlentje P, Chowdhury R. Resilience and sustainability in the management of landslides[C]//Proceedings of the institution of civil engineers-engineering sustainability. Thomas Telford Ltd, 171(1): 3\u0026ndash;14(2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eConstantin, M.; Bednarik, M.; Jurchescu, M.C.; Vlaicu, M. Landslide susceptibility assessment using the bivariate statistical analysis and the index of entropy in the Sibiciu Basin (Romania). Environ. Earth Sci, 63, 397\u0026ndash;406(2011). [CrossRef]\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen Siming; Huo Aidi; Zhang Jia et al. Identification of potential landslides in the loess hilly area (Xiji County) of Ningxia with InSAR technology. Science Technology and Engineering, 22(12): 4721\u0026ndash;4728(2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJin, Y.; Li, X.; Zhu, S.; Tong, B., Chen, F.; Cui, R.; Huang, J. Accurate landslide identification by multisource data fusion analysis with improved feature extraction backbone network. Geomatics, Natural Hazards and Risk, 13(1), 2313\u0026ndash;2332(2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePang, D.; Liu, G.; He, J.; Li, W.; Fu, R. Automatic Remote Sensing Identification of Co-Seismic Landslides Using Deep Learning Methods. Forests, 13, 1213(2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKov\u0026aacute;cs, I.P.; Czig\u0026aacute;ny, S.; Dobre, B. et al. A field survey\u0026ndash;based method to characterise landslide development: a case study at the high bluff of the Danube, south-central Hungary. Landslides, 16, 1567\u0026ndash;1581(2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXin W.; Xuanmei F.; Qiang X.; Peijun D. Change detection-based co-seismic landslide mapping through extended morphological profiles and ensemble strategy, ISPRS Journal of Photogrammetry and Remote Sensing, 187, 225\u0026ndash;239(2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, T.; Zhang, W.; Cao, D.; Yi, Y.; Wu, X. A New Deep Learning Neural Network Model for the Identification of InSAR Anomalous Deformation Areas. Remote Sens., 14, 2690(2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJia, H.; Wang, Y.; Ge, D.; Deng, Y.; Wang, R. InSAR Study of Landslides: Early Detection, Three-Dimensional, and Long-Term Surface Displacement Estimation\u0026mdash;A Case of Xiaojiang River Basin, China. Remote Sens., 14, 1759(2022). [CrossRef]\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, R.; Zhao, X.; Dong, X.; Dai, K.; Deng, J.; Zhuo, G.; Yu, B.; Wu, T.; Xiang, J. Potential Landslide Identification in Baihetan Reservoir Area Based on C-/L-Band Synthetic Aperture Radar Data and Applicability Analysis. Remote Sens., 16, 1591(2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDong, J., Niu, R., Li, B., Xu, H., \u0026amp; Wang, S. (2022). Potential landslides identification based on temporal and spatial filtering of SBAS-InSAR results. Geomatics, Natural Hazards and Risk, 14(1), 52\u0026ndash;75(2022). [CrossRef]\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYao, 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., 14, 4728(2022). [CrossRef]\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang, H.; Ju, S.; Duan, W.; Jiang, D.; Gao, Z.; Liu, H. Landslide Monitoring along the Dadu River in Sichuan Based on Sentinel-1 Multi-Temporal InSAR. Sensors, 23, 3383(2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, J.; Gong, Y.; Huang, W.; Wang, X.; Ke, Z.; Liu, Y.; Huo, A.; Adnan, A.; Abuarab, M.E.-S. Identification of Potential Landslide Hazards Using Time-Series InSAR in Xiji County, Ningxia. Water, 15, 300(2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYi, Y.; Xu, X.; Xu, G.; Gao, H. Rapid Mapping of Slow-Moving Landslides Using an Automated SAR Processing Platform (HyP3) and Stacking-InSAR Method. Remote Sens., 15, 1611(2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHussain, S., Pan, B., Afzal, Z. et al. Landslide detection and inventory updating using the time-series InSAR approach along the Karakoram Highway, Northern Pakistan. Sci Rep, 13, 7485(2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLian, B.; Wang, D.; Wang, X.; Tan, W. Early Identification and Dynamic Stability Evaluation of High-Locality Landslides in Yezhi Site Area, China by the InSAR Method. Land, 13, 569(2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKalavrezou, I.-E.; Castro-Melgar, I.; Nika, D.; Gatsios, T.; Lalechos, S.; Parcharidis, I. Application of Time Series INSAR (SBAS) Method Using Sentinel-1 for Monitoring Ground Deformation of the Aegina Island (Western Edge of Hellenic Volcanic Arc). Land, 13, 485(2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuo, H.; Mart\u0026iacute;nez-Gra\u0026ntilde;a, A.M. Susceptibility of Landslide Debris Flow in Yanghe Township Based on Multi-Source Remote Sensing Information Extraction Technology (Sichuan, China). Land, 13, 206(2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlbanwan, H.; Qin, R.; Liu, J.-K. Remote Sensing-Based 3D Assessment of Landslides: A Review of the Data, Methods, and Applications. Remote Sens., 16, 455(2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu, L.; Wang, J.; Fu, Y. Early identifying and monitoring landslides in Guizhou province with InSAR and optical remote sensing. Bulletin of Surveying and Mapping, (07), 98\u0026ndash;102(2021). [CrossRef]\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePiroton V, Schl\u0026ouml;gel R, Barbier C, et al. Monitoring the recent activity of landslides in the Mailuu-Suu Valley (Kyrgyzstan) using radar and optical remote sensing techniques, Geosciences, 10(5): 164(2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCasagli N, Intrieri E, Tofani V, et al. Landslide detection, monitoring and prediction with remote-sensing techniques, Nature Reviews Earth \u0026amp; Environment, 4(1): 51\u0026ndash;64(2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBouali E H, Oommen T, Escobar-Wolf R. Evidence of Instability in Previously-Mapped Landslides as Measured Using GPS, Optical, and SAR Data between 2007 and 2017: A Case Study in the Portuguese Bend Landslide Complex, California, Remote Sensing, 11(8): 937\u0026ndash;956(2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, C.; Shen, Z.; Weng, Y.; You, S.; Lin, J.; Li, S.; Wang, K. Modeling Landslide Susceptibility in Forest-Covered Areas in Lin\u0026rsquo;an, China, Using Logistical Regression, a Decision Tree, and Random Forests. Remote Sens, 15, 4378(2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSheng, Y.; Xu, G.; Jin, B.; Zhou, C.; Li, Y.; Chen, W. Data-Driven Landslide Spatial Prediction and Deformation Monitoring: A Case Study of Shiyan City, China. Remote Sens, 15, 5256(2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXiao L, Zhang Y, Peng G. Landslide susceptibility assessment using integrated deep learning algorithm along the China-Nepal highway, Sensors, 18(12): 4436(2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXiong, K.; Adhikari, B.R.; Stamatopoulos, C.A.; Zhan, Y.; Wu, S.; Dong, Z.; Di, B. Comparison of Different Machine Learning Methods for Debris Flow Susceptibility Mapping: A Case Study in the Sichuan Province, China. Remote Sens., 12, 295(2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang F, Zhang J, Zhou C, et al. A deep learning algorithm using a fully connected sparse autoencoder neural network for landslide susceptibility prediction, Landslides, 17: 217\u0026ndash;229(2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVan Dao D, Jaafari A, Bayat M, et al. A spatially explicit deep learning neural network model for the prediction of landslide susceptibility, Catena, 188: 104451(2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShahabi, H.; Ahmadi, R.; Alizadeh, M.; Hashim, M.; Al-Ansari, N.; Shirzadi, A.; Wolf, I.D.; Ariffin, E.H. Landslide Susceptibility Mapping in a Mountainous Area Using Machine Learning Algorithms. Remote Sens., 15, 3112(2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang, W., Ding, M., Li, Z. et al. Landslide susceptibility mapping and dynamic response along the Sichuan-Tibet transportation corridor using deep learning algorithms, Catena, 222: 106866(2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAli, N.; Chen, J.; Fu, X.; Ali, R.; Hussain, M.A.; Daud, H.; Hussain, J.; Altalbe, A. Integrating Machine Learning Ensembles for Landslide Susceptibility Mapping in Northern Pakistan. Remote Sens., 16, 988(2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, Q.; Wang, T. Deep Learning for Exploring Landslides with Remote Sensing and Geo-Environmental Data: Frameworks, Progress, Challenges, and Opportunities. Remote Sens., 16, 1344(2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, X.; Yang, L,; Song, X. Runoff and sediment load changes in the upper Yellow River and their influencing factors in recent 60 years. Journal of Lake Sciences, 36(2), 602\u0026ndash;619(2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRichard, W.; Daniel, A.; Mark, F. Volumetric interferometry for sparse 3D synthetic aperture radar with bistatic geometries. Electronics Letters, 59(12), 12851(2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen Y, Xia J, Yu C and Chen B. Editorial: InSAR crustal deformation monitoring, modeling and error analysis. Front. Environ. Sci., 10:1009492(2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang, X.; Li, X.; Li, H.; Yang, Y.; Duan, S.; Xiao, W.; Du, H.; Liu, H. Study on Rock Strata Movement Deformation and Surface Subsidence in Mining Area Based on PS-InSAR Technology. Preprints,06,2212(2023). [CrossRef]\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKaraca, Ş. O., Erten, G., Ergintav, S., Khan, S. D. Anthropogenic problems threatening major cities: Largest surface deformations observed in Hatay, T\u0026uuml;rkiye based on SBAS-InSAR. Bulletin of the Mineral Research and Exploration, 173(173), 235\u0026ndash;252(2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuo R.; Li S.; Chen Y.; Yuan,L. A method based on SBAS-InSAR for comprehensive identification of potential landslide. Journal of Geo-information Science, 21(7):1109\u0026ndash;1120(2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShe, X.; Li, D.; Yang, S.; Xie, X.; Sun, Y.; Zhao, W. Landslide Hazard Assessment for Wanzhou Considering the Correlation of Rainfall and Surface Deformation. Remote Sens., 16, 1587(2024).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"InSAR, Landslide identification, Visual analysis, Deep learning","lastPublishedDoi":"10.21203/rs.3.rs-4642799/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4642799/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eLandslide disasters have characteristics of frequent occurrence, widespread impact, and high destructiveness, posing serious threats to human lives, property, and the ecological environment. Timely and accurate early identification of landslides remains an urgent issue within the disaster prevention field. This study focuses on Jianzha County, Qinghai Province, integrating PS-InSAR、SBAS-InSAR and optical remote sensing techniques to delineate potential landslide-prone areas. Utilizing Google Earth imagery and existing landslide datasets, potential landslide points were identified through a deep learning model. The results indicate that: (1) In Jianzha County, the variation trend of the average surface velocity monitored by PS-InSAR and SBAS-InSAR technology is consistent, and the deformation monitoring results are reliable. (2) Utilizing the deep learning model, 56 potential landslide points were identified, comprising 39 high-risk points and 17 medium-risk points. By integrating the spatial distribution data of historical geological disaster points, it was found that 10 out of 13 previously occurred landslide disaster points were located at the identified high-risk landslide points, achieving a detection accuracy of 76.92%. (3) The spatial distribution of landslide points exhibits clustering, with slopes ranging from 10\u0026ndash;40\u0026deg;, elevations between 15\u0026ndash;30 m, and slope orientations predominantly towards the northeast. (4) Landslide formation is correlated with seasonal precipitation concentrations and temperature fluctuations. This method can provide a crucial basis for large-scale surface deformation monitoring and early identification of landslide risks.\u003c/p\u003e","manuscriptTitle":"Identification of Potential Landslide in Jianzha Counctry Based on InSAR and Deep Learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-07-20 05:40:26","doi":"10.21203/rs.3.rs-4642799/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-07-26T18:23:05+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-07-25T14:30:47+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-07-23T14:04:39+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"841529042291880232349668111789694126","date":"2024-07-16T22:17:11+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"179424216806128277793573705510227421721","date":"2024-07-16T15:32:09+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-07-16T15:28:05+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-07-16T15:27:04+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2024-07-02T16:10:51+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-06-28T06:24:03+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2024-06-26T12:40:05+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"c501c9c4-2318-40e7-934b-7b8a359d5363","owner":[],"postedDate":"July 20th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":34767883,"name":"Earth and environmental sciences/Natural hazards"},{"id":34767884,"name":"Earth and environmental sciences/Environmental sciences/Environmental impact"}],"tags":[],"updatedAt":"2024-09-06T06:01:35+00:00","versionOfRecord":[],"versionCreatedAt":"2024-07-20 05:40:26","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4642799","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4642799","identity":"rs-4642799","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}