Identification and Assessment of Geological Hazards in Highly Vegetated Areas Based on Multi-Source Radar Remote Sensing Data: Supporting Sustainable Disaster Risk Management

Abstract

Xiahe County, in the northwestern Gannan Tibetan Autonomous Prefecture of Gansu Province, faces recurrent geological hazards—including landslides and debris flows. Geological hazards in highly vegetated regions pose severe threats to ecological balance, human settlements, and socio-economic sustainability, hindering the achievement of sustainable development goals (SDGs). Due to the significant topographic relief and high vegetation coverage in this region, traditional manual ground-based surveys face substantial challenges in the investigation and identification of geological hazards, necessitating the adoption of advanced monitoring and identification techniques. This study employs a comprehensive approach integrating optical remote sensing, interferometric synthetic aperture radar (InSAR), and unmanned aerial vehicle (UAV) photogrammetry to investigate and identify geological hazards in the eastern part of Xiahe County, exploring the application capabilities and effectiveness of multisource remote sensing techniques in hazard identification. The results indicate that this study has shortened the time required for on-site investigations by improving the efficiency of disaster identification while also providing comprehensive, multi-angle, and high-precision remote sensing outcomes. These achievements offer robust support for sustainable disaster management and land use planning in ecologically fragile regions. Optical remote sensing, InSAR, and UAV photogrammetry each possess unique advantages and application scopes, but single-technique approaches are insufficient to fully address potential hazard identification. Developing a comprehensive investigation and identification framework that integrates and complements the strengths of multisource technologies has proven to be an effective pathway for the rapid investigation, identification, and evaluation of geological hazards. These results contribute to regional sustainability by enabling targeted risk mitigation, minimizing disaster-induced ecological and economic losses, and enhancing the resilience of vulnerable communities.

1. Introduction

Geological hazard prevention is a cornerstone of sustainability: frequent disasters disrupt ecosystem stability, displace communities, and divert resources from sustainable development initiatives. Geological hazards are characterized by great abruptness, wide spatial distribution, and severe societal impacts. Rapid acquisition of information on their distribution, morphological features, and hazard severity is of critical importance for minimizing disaster losses, preventing secondary hazards, and informing government emergency responses and post-disaster reconstruction [1,2]. However, traditional geological survey methods face substantial challenges in achieving large-scale coverage for landslide hazard mapping within short timeframes, particularly in regions with vast monitoring areas, steep mountains, deep valleys, and inaccessible terrain due to road obstructions. To address these limitations, this study employs a combined approach of optical remote sensing interpretation, InSAR deformation analysis, and UAV photogrammetry to investigate and identify potential geological hazards, while exploring the applicability and effectiveness of multisource remote sensing techniques [3]. Thus, integrating multisource remote sensing to achieve accurate and efficient geohazard identification has emerged as a frontier in current research, particularly vital for early-warning systems in complex geomorphic settings prone to abrupt failures.

With the rapid advancement of remote sensing technologies, spaceborne high-resolution optical remote sensing, synthetic aperture radar (SAR) remote sensing, airborne/terrestrial LiDAR, and UAV photogrammetry have seen increasing applications in detecting and identifying geological hazard potentials [4,5]. Currently, numerous scholars worldwide have explored the integrated use of multisource remote sensing techniques for geohazard recognition [6]. Xu et al. [7] systematically summarized the state-of-the-art SAR, optical remote sensing, and LiDAR technologies and their platforms (spaceborne, airborne, and ground-based) in China, highlighting that multi-temporal optical imagery and time-series SAR are frequently employed to detect landslide activities at annual and monthly timescales. By synthesizing technical advantages and limitations, they proposed an air-space-ground integrated collaborative investigation strategy for early-stage landslide identification and warning [8,9]. Leveraging optical remote sensing imagery and InSAR deformation monitoring, Zhong et al. [10] identified 47 landslide-prone sites in Longde County, Ningxia, where optical imagery clearly revealed geomorphic features of landslides. At the same time, InSAR technology monitored subtle surface deformations—providing critical evidence for early-stage geohazard warnings. Additionally, the synergistic application of UAV technology and InSAR overcomes the limitations of traditional ground surveys [11,12], such as low efficiency and inadequate coverage of high-risk zones. In particular, InSAR’s all-weather, wide-area, high-precision deformation monitoring capabilities enable detection of millimeter-scale slow surface deformation trends, effectively identifying potential hazards including landslides, land subsidence, and active faults. The integrated “air-space” monitoring framework established through this combination optimizes the full workflow of geohazard potential hazard “identification-evaluation-warning,” as demonstrated by Meng et al. [13], offering a robust solution for comprehensive hazard assessment in complex terrains.

Optical remote sensing, InSAR, and UAV aerial surveying each possess unique advantages and application scopes in geohazard analysis. However, sole reliance on single-technique approaches proves insufficient for effective identification of disaster-prone sites due to inherent limitations in data coverage, environmental adaptability, and spatial-temporal resolution [14]. In terms of technical application scenarios, international studies have mostly focused on areas with low vegetation coverage or medium-to-low slopes [15]. In contrast, this study targets the complex environment in the eastern margin of the Qinghai–Tibet Plateau, which features high altitude, high slope (>50°), and a vegetation coverage of 98.1%. It verifies the effectiveness of the “spaceborne InSAR large-scale screening—UAV fine verification” system, thus filling the gap in the collaborative application of multi-source remote sensing in highly extreme environments. Xiahe County is located in the northwestern part of Gannan Tibetan Autonomous Prefecture, southwestern Gansu Province, characterized by complex topography, landforms, and geological structures, with frequent occurrences of geological hazards such as landslides and debris flows. Due to the steep terrain and high vegetation coverage in this region, traditional manual on-site surveys face great challenges in the investigation and identification of geological hazards, featuring low efficiency and a large number of inaccessible areas. For regions with high vegetation coverage and complex terrain, existing single-source remote sensing technologies suffer from low identification accuracy, with optical remote sensing being affected by vegetation occlusion and InSAR being prone to decorrelation, thus requiring advanced geological hazard monitoring and identification technologies. In terms of multi-source remote sensing collaborative monitoring, regarding the challenging issue of geological hazard identification in high-vegetation-coverage areas, existing international studies have mostly focused on single-band InSAR or the simple combination of optical remote sensing and UAVs [16,17]. In this study, the fusion of L-band InSAR (with strong vegetation penetration) and C-band InSAR (with high temporal resolution) can complement the deformation monitoring capability in vegetated areas. The synergistic advantages of L + C-band InSAR: In areas such as Qu’ao Township (with vegetation coverage > 95%), where traditional C-band InSAR is completely decorrelated, L-band ALOS-2 data successfully identified seven hidden landslide dangers (which were not detected by previous single-band technologies), verifying its unique value in penetrating high-density vegetation to obtain surface deformation. The 12-day short revisit cycle of C-band Sentinel-1 captured the accelerated deformation process of the Bola Township landslide during the heavy rain in July 2023 (with a monthly deformation of −45 mm), reflecting the advantage of “high temporal resolution for monitoring dynamic changes”. The combination of the two realizes full coverage of static hidden danger identification and dynamic process tracking. For high and steep terrain areas (such as the 58° slope in Madang Town), UAVs have solved the problem of missed detection in areas inaccessible to manual surveys, improving the interpretation accuracy of the boundary of hidden danger points from the “hundred-meter level” of satellite images to the “meter level”. Field investigations using UAVs can effectively verify the preliminary remote sensing screening results and reduce the missed detection rate in high and steep terrain areas. A combination of multiple technical means, including optical remote sensing, synthetic aperture radar interferometry, and UAV aerial surveying, was comprehensively adopted to carry out the identification of geological hazard hidden dangers in the eastern part of Xiahe County, with verification and analysis through field investigations. The results show that multi-source remote sensing technology can significantly reduce on-site working time, provide multi-angle, all-round high-precision visual results, and offer an effective approach for the rapid investigation, identification, and evaluation of geological hazards—key to long-term sustainability in highly vegetated, ecologically sensitive regions.

2. Overview of the Study Area and Data Sources

2.1. Overview of the Study Area

Xiahe County is situated in the northwestern part of Gannan Tibetan Autonomous Prefecture, southwestern Gansu Province, and serves as a primary region prone to geological hazards. The eastern portion of Xiahe County, highlighted as the key hazard-prone area in Figure 1, encompasses Damai Township, Madang Town, Qu’ao Township, Tangga’ang Township, Wangge’ertang Town, Bola Township, and Zayou Township. This region lies at the eastern margin of the Qinghai–Tibet Plateau, with elevations ranging from 2100 to 4700 m. It is characterized by frequent winds in winter and spring, abundant rainfall in summer and autumn, and an average annual precipitation of 516 mm. The mean vegetation coverage reaches 98.1%, with a forest coverage rate of 12.28%, dominated by alpine meadows and forest ecosystems. Geographically, the study area features overlapping mountain ranges and crisscrossing valleys, exhibiting a distinct plateau climate with extensive alpine dissection. Influenced by seismic activities, including the Wenchuan Earthquake, Minxian–Zhangxian Earthquake, and Jiuzhaigou Earthquake, loosening of rock and soil has occurred, making it difficult for the geological environment to recover in the short term. This has lowered the threshold of geological hazard-inducing factors, increasing the probability of hazard occurrences.

2.2. SAR Data

The Sentinel-1A satellite, launched by the European Space Agency (ESA) in April 2014 as the first mission in the Copernicus Sentinel series for Earth observation, provided descending orbit data used in this study, which were acquired from the ASF Data Search. The VV co-polarization mode was selected, as it generally demonstrates superior performance for surface deformation monitoring compared to other polarization configurations. The sensor offers a range resolution of 5 m and an azimuth resolution of 20 m. A total of 27 descending orbit image scenes were collected between January and December 2023, spanning the study period for geological hazard analysis using InSAR techniques.

The Advanced Land Observing Satellite-2 (ALOS-2), launched by the Japan Aerospace Exploration Agency (JAXA) in May 2014, carries the Phased Array L-band Synthetic Aperture Radar-2 (PALSAR-2) sensor. Operating in the L-band, this sensor achieves a maximum resolution of 3 m (range) × 1 m (azimuth), enabling the satellite to operate under all-weather conditions with a maximum swath width of 490 km × 350 km. Compared to short-wavelength SAR systems, ALOS-2 effectively mitigates vegetation-induced decorrelation effects on slopes, thereby facilitating more accurate surface information acquisition and enabling precise monitoring of geological hazards [18]. In this study, co-polarized (HH polarization) images acquired on 22 May 2022 and 26 February 2023 were selected (

Table 1. Basic parameters of SAR.

Key Parameters	Data Source Parameter Values
	ALOS-2	Sentinel-1A
Resolution	3 × 1	5 × 20
Wavelength	24.6 cm	5.6 cm
Band	L band	C band
Orbit Direction	Descending Orbit	Descending Orbit
Imaging Mode	Ultra-Fine	IW
Polarization Mode	HH	VV
Image Count/Scene	2	27

).

2.3. Ancillary Data

The Digital Elevation Model (DEM) used in this study, with a spatial resolution of 30 m, was obtained from the Resource and Environmental Science and Data Center (RESDC) (https://www.resdc.cn/Default.aspx accessed on 21 August 2024). Road and river network data were sourced from OpenStreetMap (https://openmaptiles.org/languages/zh/#0.58/0/0 accessed on 23 August 2024). Geological hazard inventory data were provided by the Third Geological and Mineral Exploration Institute of the Gansu Provincial Bureau of Geology and Mineral Resources. Precipitation data were acquired from the National Centers for Environmental Information (NCEI) of the United States (https://www.ncei.noaa.gov/data/global-summary-of-the-day/archive/ accessed on 26 August 2024).

The coordinate system of the data is uniformly adopted as WGS84, and specific information regarding the data sources is presented in

Table 2. The data sources, types, and timeframe of ancillary data.

Data Name	Data Source	Type	Timeframe
DEM	RESDC	Raster	2023
Precipitation	NCEI	Vector	2023
Road and river network	OpenStreetMap	Vector	2023
Geological hazard	The Third Geological	Vector	2024

.

3. Research Methodology

The continuous advancement of remote sensing technology has driven the continuous improvement of the geological hazard detection technical system, integrating air-space-ground integrated remote sensing observation and on-site verification. Multisource remote sensing technology that combines multi-platform observation methods is revolutionizing traditional geological hazard investigation approaches and has become a critical technical support for hazard identification. Aiming at the complex geological hazard prevention and control needs of the study area, this research systematically integrates high-resolution optical remote sensing, InSAR, and UAV photogrammetry to construct a multisource remote sensing collaborative investigation framework [18]. This technical combination fully leverages the advantages of each method: optical remote sensing provides high-precision surface morphology data, InSAR technology achieves millimeter-level deformation monitoring, and UAV photogrammetry ensures detailed exploration of high-risk areas. Through the fusion of multisource data, the positioning accuracy of geological hazards and the efficiency of feature recognition have been significantly improved. The specific technical process is shown in Figure 2.

In the spatial dimension, InSAR with L/C bands enables deformation screening at large regional scales, facilitating the delineation of potential hazard zones. UAVs are deployed for detailed surveys in high-risk areas to verify deformation boundaries. Optical remote sensing provides background information on surface morphology, aiding in the interpretation of geomorphic features related to hazards. The priority sequence follows: large-scale initial screening via InSAR, feature identification using optical remote sensing, and detailed verification by UAV.

In the temporal dimension, Sentinel-1 (C band, 12-day revisit cycle) is utilized for monitoring long-term slow deformation, while ALOS-2 (L band) captures sudden large deformations such as short-term acceleration of landslides. UAVs conduct intensive aerial surveys during the active period of disasters in the rainy season, forming a temporal collaborative framework characterized by long-term sequence monitoring and short-term detailed tracking.

3.1. InSAR Deformation Monitoring

While optical remote sensing interpretation remains one of the primary methods for large-scale early identification of landslide hazards, it is significantly constrained by cloudy and foggy conditions and typically provides only qualitative interpretation of macro-geomorphic features [19]. In contrast, InSAR technology offers distinct advantages, including all-weather/all-day operation, cloud-penetrating capability, wide coverage, high spatial resolution, non-contact measurement, and cost-effectiveness, making it suitable for large-scale geological hazard surveys, surface deformation analysis, and long-term continuous monitoring. These attributes provide critical data support for early hazard identification and analysis, particularly their unique ability to detect deformation zones through continuous tracking of millimeter-scale surface displacements over extensive areas [20].

To address the challenges of high vegetation coverage and rapid slope deformations, this study employs L-band ALOS-2 satellite data and C-band Sentinel-1A satellite data, utilizing D-InSAR and SBAS-InSAR techniques, respectively, for surface deformation monitoring in the study area. The L-band (23 cm wavelength) of ALOS-2 penetrates dense vegetation canopies to reach the ground surface, significantly reducing vegetation-induced decorrelation and enabling effective capture of abrupt large-gradient deformation events such as landslides and collapses. Conversely, the C-band (5.6 cm wavelength) of Sentinel-1A leverages its short revisit period (6 days) to support SBAS-InSAR in retrieving long-term slow deformation processes and patterns through time-series data analysis. This synergistic observation system formed by the two datasets balances the monitoring requirements for both abrupt and gradual deformations.

3.1.1. SBAS-InSAR Data Processing

The fundamental principle of the Small Baseline Subset (SBAS) method involves setting predefined temporal and spatial baseline thresholds to select interferometric image pairs that meet the criteria. These image pairs are then freely combined to generate multiple differential interferograms. Subsequently, based on the baseline characteristics of each interferometric pair, they are partitioned into several subsets. Within each subset, the singular value decomposition (SVD) method is employed to solve the least-squares problem, thereby deriving the surface deformation time series for the entire observation period. The SBAS technique not only enhances the spatiotemporal resolution of monitoring but also effectively addresses the issue of temporal discontinuities caused by excessively long baselines. Notably, in regions with high vegetation coverage and low coherence, its deformation monitoring performance is particularly prominent [21]. Therefore, due to its favorable adaptability to vegetated areas, the SBAS-InSAR technology is well-suited to the characteristics of high vegetation coverage in the study area, making it a robust approach in the field of InSAR applications for geological hazard monitoring.

3.1.2. The Principle of SBAS-InSAR

Based on the SARscape 6.1 platform, a total of 27 scenes of descending orbit SAR data acquired from January to December 2023 were processed through a systematic workflow. The data processing pipeline was broadly divided into seven sequential steps: SLC data import and cropping, image registration, interferometric processing, orbital refinement and reflattening, phase unwrapping, deformation inversion, and geocoding (Figure 3).

To enhance satellite orbital accuracy and mitigate systematic errors arising from orbital uncertainties, this study employed Precise Orbit Determination (POD) data for orbital correction. To eliminate topographic phase contributions, a 30 m resolution DEM from the Geospatial Data Cloud was utilized. A temporal baseline threshold of 108 days and a spatial baseline threshold of 5% were set, with only connections featuring the minimum spatial baseline retained to optimize interferometric coherence.

To improve coherence in the study area, the Delaunay minimum cost flow (MCF) algorithm was applied for phase unwrapping, configured with a decomposition level of 2, an unwrapping correlation threshold of 0.2, and the Goldstein filter for adaptive phase noise reduction. Subsequently, interferogram pairs with suboptimal unwrapping performance were excluded to ensure reliability. The remaining data underwent deformation inversion and geocoding to convert line-of-sight displacement measurements into georeferenced deformation products. This rigorous processing chain balances computational efficiency and measurement accuracy, making it suitable for retrieving subtle surface deformations in complex geological settings prone to hazards.

3.1.3. The Principle of D-InSAR

D-InSAR isolates topographic and deformation information through the interferometric phase difference derived from two or more SAR images. The core principle involves eliminating the flat-earth phase, topographic phase, and atmospheric delay through differential processing, such that the residual phase directly reflects millimeter-scale surface deformations [22]. Subsequent phase unwrapping and geometric inversion then enable the retrieval of precise deformation fields. Although D-InSAR cannot capture the time-series evolution of surface deformations across the study area, it excels in monitoring relative displacements occurring between two acquisition epochs. This makes it particularly suitable for detecting abrupt deformations over short temporal intervals. The technique has been widely adopted in emergency investigations of geological hazards such as landslides and debris flows, where it has demonstrated remarkable effectiveness in practical applications by providing high-precision deformation measurements critical for rapid hazard assessment and response planning [23].

3.1.4. D-InSAR Data Processing

In response to the study area’s characteristics of complex topography, high vegetation coverage, and broad potential deformation zones, this study employed D-InSAR for surface deformation monitoring. The specific workflow was as follows: L-band synthetic aperture radar interferometric data with 3 m resolution acquired by the ALOS-2 satellite were selected, focusing on two image pairs from 22 May 2022 and 26 February 2023. Data processing was conducted using the SARscape 6.1 platform, comprising the following key steps: First, image co-registration and resampling were performed to achieve sub-pixel alignment, followed by interferometric processing to generate the interferometric phase. A 30 m resolution SRTM DEM was utilized as reference data to remove the flat-earth phase effect. Subsequently, noise filtering was applied to enhance the signal-to-noise ratio, and a phase unwrapping algorithm was implemented to retrieve deformation-related phase information. Finally, geocoding transformed the radar-coordinate results into a geographic coordinate system, yielding the surface deformation monitoring products. Experimental validation demonstrated that this method achieves centimeter-to-millimeter scale accuracy, effectively adapting to deformation monitoring requirements in regions with complex topography and dense vegetation (Figure 4). This approach highlights the robustness of D-InSAR for detecting subtle displacements in challenging environments prone to geological hazards [24].

3.2. UAV Photogrammetric Methodology

Although optical remote sensing and InSAR techniques have been widely applied in the early identification of geological hazard potential, both exhibit distinct applicability scopes and limitations. Relying solely on traditional manual ground surveys not only imposes heavy workloads and high risks but also significantly exacerbates investigation difficulties and time consumption in inaccessible areas such as steep terrains or dangerous zones, where human access is restricted or perilous. In recent years, advancements in UAV technology have enabled rapid generation of geospatial datasets through high-precision vertical and oblique photogrammetry. These include digital topographic maps, digital orthophotography maps (DOM), digital surface models (DSM), and realistic 3D scene models of the surveyed area. Such products offer multidimensional representations of hazard body characteristics and their surrounding environments, clearly illustrating hazard types, scales, and topographical features and precisely delineating the spatial extent of hazard bodies and their affected zones. This capability addresses the limitations of conventional methods by providing detailed, multisource geospatial information essential for comprehensive hazard characterization, risk assessment, and emergency planning in complex or inaccessible terrains [25].

The primary workflow of UAV-based aerial surveying comprises sequential stages: site selection, flight path planning, UAV remote sensing image acquisition, aerial triangulation, generation of visualized digital imagery products (DOM and DEM), and application of resultant data. The UAV utilized in this study is the DJI Matrice 300 RTK, equipped with a Hasselblad L1D-20c camera (20-megapixel resolution). The flight operations were conducted at a relative altitude of 120 m above the ground surface. The imaging parameters were configured with an 85% forward overlap and a 75% lateral overlap to ensure sufficient redundancy for photogrammetric processing.

A combined imaging strategy was employed, incorporating both vertical photography (90° camera angle) and oblique photography (45° camera angle) with five different orientations. The flight path followed a strip pattern, covering not only the geological hazard-prone area but also an additional 0.5 km buffer zone around its perimeter to capture comprehensive topographic and structural information. This configuration of UAV-based data acquisition is designed to provide high-resolution, multi-perspective imagery that can support detailed geospatial analysis and hazard assessment in the study area.

3.3. Construction of Geological Disaster Interpretation Markers

Based on the collection and review of geological environmental data in the study area, and in accordance with the geological environmental conditions, potential geological hazard points were identified through the integration of field reconnaissance and remote sensing imagery. Remote sensing interpretation markers were established according to their spectral and spatial characteristics in the imagery, encompassing two key components: optical remote sensing interpretation markers and InSAR deformation interpretation markers. By comparing the geomorphic changes in multi-temporal images and combining the texture, tone anomalies, and shadow features of surface coverings, the boundaries and evolutionary trends of disaster bodies such as landslides, collapses, and debris flows can be accurately identified.

Landslide bodies typically exhibit morphological characteristics such as winnowing basket-like, horseshoe-shaped, and tongue-like features. Tensile fractures at the rear edge of landslides often appear as linear dark-tone belts in remote sensing images. InSAR monitoring reveals significant differences in deformation magnitude between the landslide deformation zone and the peripheral area. In the case of rapid landslides, large deformations at the rear edge often lead to decorrelation phenomena, manifested as image fractures or void pixel areas in the interferograms (Figure 5).

In collapse areas, rock outcrop zones often exhibit significant color contrast with surrounding vegetation. High-resolution images reveal rock blocks, collapses, and accumulation bodies with distinct shadows, meanwhile presenting complex radar wave reflection characteristics. Such areas typically feature steeper slopes and smaller areas, demonstrating deformation characteristics of overall subsidence displacement and out-of-slope tilting (Figure 6).

Debris flows and their deposits typically exhibit a light tone with subtle flow-like texture features in remote sensing images. The main body is distributed in an elongated shape along valleys, and fan-shaped deposits are often formed at the valley exits. The deformation characteristics of debris flows show a close correlation with precipitation cycles and intensity in the temporal dimension (Figure 7).

4. Results

4.1. Deformation Results from InSAR

Surface deformation monitoring in the study area was conducted using SBAS-InSAR and D-InSAR techniques, yielding vertical deformation rate distributions. SBAS-InSAR results indicate that the average surface deformation rates in the study area range from −106.96 to 60.14 mm/year (Figure 8), where negative values denote surface subsidence (moving away from the reference plane) and positive values indicate uplift (approaching the reference plane). The identification thresholds for InSAR deformation rates were determined by integrating statistical data of historical hazards in the study area with previous InSAR-based geological hazard studies: annual average deformation of landslides < −50 mm; collapses < −30 mm; and a correlation coefficient > 0.7 between debris flows and precipitation (p < 0.05) [26,27].

In relatively flat residential areas, the surface exhibits good stability, with deformation rates concentrated between −15 and 15 mm/year, primarily in regions such as Madang Town. These areas are characterized by sparse population, distance from rivers, and gentle topography, leading to low geological hazard risks due to their stable geomorphological conditions.

Negative deformation rate zones are predominantly distributed along rivers and roads, focusing on Zayou Township, Bolai Township, and Qu’ao Township. These regions are marked by severe topographic relief, developed fault structures, frequent anthropogenic mining activities, and intense fluvial erosion, which collectively degrade surface stability and elevate geological hazard susceptibility. Therefore, they should be prioritized for intensive monitoring and management.

As shown in Figure 8, regions such as Qu’ao Township and Tangga’ang Township exhibit dense vegetation cover, which results in the inability of Sentinel-1’s C-band radar to effectively penetrate the canopy and acquire surface deformation information. By contrast, the L-band radar of ALOS-2 possesses stronger vegetation penetration capability. Its D-InSAR technique can efficiently penetrate the canopy to obtain surface deformation information (Figure 9), making it particularly suitable for rapidly identifying significant deformation areas in single observations.

The D-InSAR monitoring results indicate that in areas with severe C-band decorrelation, such as Qu’ao Township and Tangga’ang Township, dense vegetation cover enables root systems to stabilize soil, reduce water and soil loss, enhance surface stability, and effectively prevent landslides and debris flows. However, negative surface deformation areas are concentrated along both sides of river channels and transportation corridors, with major distributions in townships such as Zha’you Township, Bola Township, and Qu’ao Township. These regions are characterized by steep terrain slopes, frequent human engineering activities, and intense fluvial erosion, which reduce the stability of surface rock and soil masses and pose a higher risk of geological hazards such as landslides and collapses.

The delineation of geological hazard-prone areas follows the steps below: First, the InSAR deformation threshold criteria are established as follows: the annual average deformation of landslides is < −50 mm, the annual average deformation of collapses is < −30 mm, and the correlation coefficient between debris flows and precipitation is > 0.7 (p < 0.05). Furthermore, the initially delineated ranges were revised by incorporating geomorphic features extracted from optical remote sensing (e.g., landslide scarps, debris flow fans) and field verification via UAV (

Table 3. Summary table of geological hazards and potential hazard identification.

Hazard Category	Number of Identified Locations	Area/km2
Landslide	34	11.962
Collapse	16	5.118
Debris flow	9	9.253

).

4.2. Analysis of Geological Hazards in Typical Areas

Taking the deformation characteristics of potential hazard sites, spectral and spatial features of remote sensing imagery, results of on-site verification, and the threat level to villages and buildings as comprehensive judgment criteria, this study conducts a systematic analysis of typical geological hazard-prone sites.

4.2.1. Analysis of Collapsing Hazard Points

As depicted in Figure 10, the collapsing hazard site is situated on the sunny slope, with a slope length of approximately 177 m and width of 356 m, and a slope angle of 58°. The imagery exhibits an overall brown tone with clear structural textures. The lower area of the slope appears gray, showing distinct artificial mining traces, and the underlying gray-black stripe is inferred to be a highway, suggesting that the slope was formed during road construction, posing a primary threat to the underlying road and passing vehicles. On-site verification reveals that the hazard site features a steep terrain with significant topographic relief, and the cut slope surface is predominantly composed of earth-rock layers without retaining walls, further exacerbating the threat to the road and vehicles. Currently, the overall stability is basically stable, but under adverse conditions such as rainfall or earthquakes, deformation of the collapsing mass may be intensified.

4.2.2. Analysis of Landslide-Prone Sites

As shown in Figure 11, this hazard site is characterized as a landslide located on the sun-facing slope of a hillside. The landslide body has a longitudinal length of approximately 315 m and a transverse width of 213 m, with a slope gradient of 77°. The landslide area exhibits distinct imaging features, presenting a soil-brown hue on the sun-facing slope, which contrasts significantly with the tonal characteristics of surrounding land cover. Obvious landslide traces are visible, accompanied by sparse vegetation. The landslide’s front edge has impacted residential buildings and roads. With a clear morphological outline and loose soil texture, the landslide is primarily influenced by rainfall, vegetation cover, topographic factors, and anthropogenic activities. Interpretation of InSAR results indicates an annual average deformation rate of −70 mm/year at this site, with pronounced deformation manifested as red and orange zones in the interferometric map.

Field investigations confirmed that this landslide was triggered by sand and gravel mining activities. A residual mining platform persists in the mid-section of the landslide body, around which two distinct landslide masses have developed. Prominent landform features, including landslide steps, slip surfaces, and flow marks, are clearly discernible, indicative of progressive failure processes. Sparse vegetation covers the surrounding area of the landslide mass, while the region below the platform consists of a slag accumulation zone and a waste dump. The loose surface materials in these areas are highly susceptible to secondary sliding induced by rainfall. Construction facilities located at the base of the landslide are directly threatened, highlighting the urgent need for risk mitigation measures.

4.2.3. Debris Flow Hazard Analysis

As shown in Figure 12, the watershed morphology is conducive to water convergence, and various unfavorable geological phenomena develop within the channel. Comprehensive assessment indicates the presence of debris flow hazards in this watershed. The overall tone of the regional imagery is yellowish-brown, with distinct boundaries of debris flow disasters. The image texture is generally smooth, and the mountain slope gradient is large, forming steep slopes. The influencing factors include precipitation, topography, and anthropogenic factors, among which the precipitation amount reaching 140 mm in July 2023 is the primary influencing factor. The interpretation results of InSAR show that the region experienced significant deformation in July.

Field investigations reveal that the debris flow exhibits a band-like planar morphology in this area. The rugged topography here is characterized by a special torrent carrying massive sediments and stones, triggered by natural disasters such as rainstorms and landslides. In the source area, the steep terrain is coupled with sparse vegetation and severe local surface exposure, accompanied by accumulations of fallen rocks; abandoned houses are distributed within the region, with farmland dominating the land use. The transportation area is developed with multiple gullies and soft soil; the deposition area is located at the gully mouth, forming an alluvial fan landform. Sediment deposition is observable in the diversion channels, and distinct step-like scarps at the front are formed by soil layer sliding. Current damage primarily manifests as sediment burial, and the potential loss area in the future is concentrated in the deposition zone at the gully mouth.

5. Discussion

5.1. Suitability Comparison of ALOS-2 and Sentinel-1 Data in Vegetated Areas

The ALOS-2 satellite demonstrates superior penetration capability in densely vegetated areas due to its L-band SAR system operating at a wavelength of approximately 23.6 cm. This longer wavelength enables effective penetration through vegetation canopies, thereby acquiring essential ground surface scattering information that proves particularly advantageous for monitoring potential geological hazards under vegetative cover. For instance, in heavily vegetated regions such as Qu’ao and Tanggaan Townships, where C-band SAR systems suffer severe decorrelation, ALOS-2’s L-band maintains coherent penetration capacity to retrieve surface deformation signals. However, practical applications face limitations due to the system’s relatively low temporal resolution, relatively high cost of data acquisition, and potential challenges in capturing rapid geological hazard evolution processes.

The Sentinel mission provides C-band SAR data (wavelength ~5.6 cm) characterized by high temporal resolution. Whereas C-band demonstrates inferior vegetation penetration capability compared to L-band systems, it can still retrieve measurable surface deformation signals in areas with sparse vegetation coverage or low canopy height. The frequent temporal resolution enables frequent monitoring during geologically active periods, facilitating the timely detection of anomalous surface displacements and supporting real-time early warning systems. Furthermore, the operational advantages of Sentinel data, including extensive global coverage and cost-effectiveness, make it particularly suitable for large-scale geological hazard inventories. However, in densely vegetated regions, C-band signals are significantly affected by vegetation scattering effects, resulting in degraded coherence that complicates phase unwrapping processes and reduces measurement reliability.

The synergistic integration of ALOS-2 L-band and Sentinel C-band datasets establishes complementary advantages through their distinct electromagnetic and temporal characteristics. The L-band system ensures vegetation penetration reliability and initial deformation identification in densely vegetated areas, while the C-band’s intensive temporal sampling enhances deformation process continuity and enhances micro-displacement characterization. By synergistically combining DInSAR and time-series analysis techniques, this dual-band fusion approach spatially improves deformation interpretation reliability in vegetated terrain through L-band’s canopy penetration, while temporally achieving comprehensive monitoring capabilities spanning transient to gradual deformation dynamics. Such a multi-sensor fusion strategy not only optimizes the spatiotemporal continuity and measurement precision of vegetation-covered deformation monitoring but also demonstrates multi-scale analytical strengths in deciphering genetic correlations of compound geological hazards. This integrated framework ultimately delivers enhanced resolution, temporally continuous, and physically robust solutions for surface deformation monitoring in high vegetation-coverage environments.

5.2. The Synergistic Enhancement Mechanism of Multisource Remote Sensing Data

Optical remote sensing, with its high spatial resolution (<1 m), can clearly present macroscopic features such as landslides’ dustpan-shaped landforms and debris flow fan deposits. However, it is susceptible to cloud, fog, and vegetation occlusion, resulting in a high data missing rate during continuous rainy periods. Nevertheless, when overlaid with InSAR deformation data, it can significantly improve the accuracy of geological disaster interpretation. L-band InSAR (ALOS-2), with its long wavelength (24.6 cm), has outstanding penetration ability for high-density vegetation (coverage > 90%). However, limited by its long revisit cycle, it is difficult to capture sudden deformation. In the future, this shortcoming can be compensated for by fusing with C-band data. C-band InSAR (Sentinel-1), with a short revisit cycle of 12 days, is good at monitoring long-term gradual deformation processes (such as landslide annual deformation of −50~−100 mm). But in densely vegetated areas, it is prone to reduced accuracy due to decorrelation. If combined with machine learning to optimize the phase unwrapping algorithm, it is expected to reduce error accumulation in complex terrains. UAV aerial surveys can accurately depict details such as collapse cracks and landslide steps on high and steep slopes, but they are limited by the single-mission operation range (<5 km²) and weather conditions. It is suitable as a microscopic supplement to satellite remote sensing. The linkage with spaceborne data can construct a hierarchical monitoring system of macroscopic screening–microscopic verification.

These technologies demonstrate unique advantages throughout the lifecycle of geological disasters: InSAR excels in pre-disaster creep detection; optical remote sensing supports dynamic monitoring via multi-temporal change detection, and UAV systems enable on-site investigation and validation to achieve post-disaster damage quantification. Their integration establishes a hierarchical monitoring framework: “satellite-based wide-area screening UAV-focused detailed investigation; ground truth validation.” This multi-platform collaborative system effectively bridges the scale gap between satellite observations (kilometer-scale) and in situ measurements (centimeter-scale), significantly enhancing the identification rate of disaster hazards in complex terrain areas. Compared to traditional surveys, multi-source remote sensing enables proactive rather than reactive disaster management, aligning with the principle of sustainable development to meet present needs without compromising future generations.

5.3. Limitations and Improvements in Future

This study employed multisource remote sensing techniques to identify and investigate geological hazards in the eastern region of Xiahe County. Although preliminary achievements have been attained, the research still contains several limitations to be further improved, primarily due to the complexity of geological hazard-inducing factors and the regional geological environment. The remote sensing monitoring and interpretation of geological hazards such as landslides, collapses, and debris flows represent a core task in the field of disaster prevention and mitigation. Traditional manual visual interpretation methods relying on optical remote sensing images exhibit significant drawbacks: interpretation efficiency is highly dependent on operators’ experience, requiring substantial time and labor for processing massive remote sensing datasets, while subjective judgment may lead to omissions or misinterpretations. In recent years, breakthroughs in deep learning technology have provided a new paradigm for intelligent interpretation of geological hazards. For instance, landslide boundary extraction is a multi-feature fusion scheme based on the U-Net model will be proposed, integrating three categories of key data: RGB bands and NDVI vegetation indices from optical remote sensing (to distinguish boundaries between vegetation and bare soil); slope and aspect data derived from UAV-based DSM (to identify topographic scarps as indicators of landslide boundaries); and InSAR deformation rate data (with a threshold of −25 mm/year for defining inner boundaries). This scheme will be particularly suitable for landslides in medium-to-high vegetation-covered areas with 70–95% vegetation coverage, effectively addressing the core issue of blurred boundaries caused by vegetation occlusion. Deformation hotspot dynamic monitoring: A combined scheme of time-series InSAR and deep learning will be developed, employing a “bidirectional LSTM + Mask R-CNN” hybrid model. By inputting time-series deformation rate data from Sentinel-1 (with a temporal resolution of 6 days), the model will automatically identify anomalous regions where the monthly average rate abruptly increases from −20 mm to over −50 mm (e.g., during the acceleration phase of landslides). The model will be able to output the time nodes and spatial distribution of such mutations. It will be embedded into the geological hazard monitoring system of Xiahe County, with a planned trial operation during the 2024 rainy season (June-September). Upon detecting deformation hotspots, the system will automatically push early warning information to the County Bureau of Natural Resources, assisting in formulating emergency inspection routes.

6. Conclusions

Surface deformation monitoring was carried out in the study area using SBAS-InSAR based on Sentinel data and D-InSAR based on ALOS-2 data, and the distribution characteristics of vertical deformation rates were obtained. Areas with negative deformation rates are mainly distributed along both sides of rivers and roads, concentrated in townships such as Zhayou, Boluo, and Qu’ao. These regions exhibit significant topographic undulations, frequent human mining activities, and severe river erosion, leading to poor surface stability and proneness to geological disasters. In areas with severe decorrelation in Sentinel data, such as Qu’ao and Tangga’ang townships, ALOS-2 data demonstrates better penetration capability, acquiring surface deformation information. These areas have dense vegetation, whose root systems can stabilize soil, reduce water and soil loss, enhance surface stability, and effectively prevent landslides and debris flows.

In this study, the fusion of L-band InSAR (featuring strong vegetation penetration) and C-band InSAR (with high temporal resolution) enables complementary deformation monitoring capabilities in vegetated areas. Field investigations using UAV can effectively verify the preliminary screening results from remote sensing, thereby reducing the false negative rate in high and steep terrain regions. A comprehensive integration of multiple technical approaches, including optical remote sensing, InSAR, and UAV aerial surveying, was employed to conduct the identification of geological hazards in the eastern part of Xiahe County, with subsequent verification and analysis through field surveys. Multi-source remote sensing technology can significantly reduce on-site working time, provide multi-angle, all-round high-precision visual results, and offer an effective approach for the rapid investigation, identification, and evaluation of geological hazards. The research methodology presented herein can serve as a reference for geological hazard prevention and control.