GIS-Based Sliding Surface Reconstruction for Rapid Landslide Volume Estimation

Abstract

Landslides cause significant casualties and economic losses worldwide each year, creating an urgent demand for rapid and reliable volume estimation during emergency response. Conventional approaches often involve trade-offs among accuracy, efficiency, and data availability, particularly when pre-event topographic data are unavailable. This study proposes a novel GIS-based method for rapid landslide volume estimation through sliding surface reconstruction. By integrating open-source geospatial data (post-landslide Digital Elevation Model and landslide boundary KML) with spline interpolation and spatial analysis, the method reconstructs the subsurface sliding surface and calculates volume by comparing this surface with the post-landslide DEM. Applied to the 2019 Shuicheng landslide (Guizhou Province, China), the method yielded a volume estimate of 1.58 × 10⁶ m³, which deviates by only ~5% from official survey data. The entire workflow can be completed within approximately one hour, demonstrating high efficiency, low operational cost, and acceptable accuracy for rapid post-disaster assessment during the critical “golden 72 h”.

1. Introduction

Landslides, defined as the gravity-driven downslope movement of rock, soil, and debris, are among the most destructive natural hazards globally, claiming thousands of lives and causing extensive socioeconomic damage annually [1,2]. China, with its complex topography and geology, is a global hotspot for landslide disasters [3]. The urgent need for rapid and accurate post-disaster volume estimation is universally recognized, as this parameter is critical for assessing destructive potential, modeling runout behavior, and optimizing emergency resource allocation during the critical “golden 72 h” [4].

Conventional methods for landslide volume estimation, however, present a persistent trilemma, forcing a trade-off among accuracy, efficiency, and data availability. Field surveys involving seismic reflection and drilling can provide detailed subsurface data but are labor-intensive, time-consuming, costly, and pose significant safety risks in unstable terrain [5]. Empirical models, which establish statistical power-law relationships between landslide area (A) and volume (V), are efficient but primarily suited for regional-scale studies. Their application to individual events often produces substantial errors due to the oversimplification of site-specific geomorphic complexities and sliding mechanics. Finally, geometric methods that approximate the landslide body using simple shapes (e.g., a semi-ellipsoid) provide high computational efficiency but cannot adequately represent complex three-dimensional morphologies, leading to potentially large and unquantifiable errors.

The advancement of remote sensing has popularized the use of multi-temporal Digital Elevation Models (DEMs) obtained from drone photogrammetry and LiDAR [6,7,8,9,10]. Unmanned Aerial Vehicle (UAV) photogrammetry estimates landslide volume by acquiring high-resolution multi-view images and reconstructing three-dimensional topography using Structure-from-Motion (SfM) and Multi-View Stereo (MVS) techniques [11]. The derived dense point clouds are used to generate digital surface models (DSMs), from which landslide volume is calculated through pre- and post-event DEM differencing.

Although DEM differencing is widely used for landslide volume estimation, its application is often limited by the unavailability of high-quality pre-event DEMs. In addition, this method may underestimate the true depletion volume when displaced materials remain partially within the source area after failure, which is common in coherent landslides (Figure 1) [12]. The reliability of the method can also be affected by environmental conditions (such as strong winds and reflections) during data acquisition, while the associated data-processing workflow is often computationally intensive. LiDAR-based landslide volume estimation method derives high-resolution terrain models from dense laser point clouds and calculates volume through pre- and post-event DEM differencing. Despite high accuracy and vegetation penetration capability, it is limited by high operational costs, complex data processing, and potential errors from point classification and multi-temporal data inconsistency.

A promising alternative approach that avoids dependence on pre-event topographic data is to leverage the geometric information implicit in the landslide itself to reconstruct the sliding surface. This surface constitutes the fundamental boundary of the landslide mass [13,14], and its reconstruction allows for direct calculation of the displaced volume by comparing it with the post-sliding ground surface. However, the detailed subsurface data required for such an approach are traditionally obtainable only through extensive and costly field investigations [15].

Despite recent advances in remote sensing and DEM-based analysis [16], a practical method capable of rapidly estimating the volume of a single landslide using only readily available post-disaster data remains lacking. Existing approaches either depend heavily on pre-event topographic datasets, extensive field investigations, or simplified geometric assumptions, thereby limiting their applicability during emergency response.

To address this gap, this paper proposes a novel, GIS-based methodology for rapid landslide volume estimation. Our approach is grounded in the principle that the geometry of the landslide’s depletion area and its boundary, as captured by post-event DEMs and optical imagery, contain critical constraints on the morphology of the underlying sliding surface. The core of the proposed method is to integrate open-source geospatial data—specifically a post-landslide DEM and a manually delineated landslide boundary—with spline interpolation techniques under geologically constrained inclination models. This data-driven integration allows for the simulation of the most probable sliding surface geometry, from which the landslide volume is directly calculated.

To address these limitations, this study proposes a GIS-based framework for rapid landslide volume estimation through the reconstruction of sliding surface [17,18]. The proposed method integrates post-event DEM data, manually delineated landslide boundaries, and spline-based interpolation to reconstruct a geometrically plausible subsurface sliding surface under limited-data conditions. Unlike conventional DEM differencing approaches, the method does not require pre-event topographic datasets.

The feasibility of the proposed workflow is evaluated using the 2019 Shuicheng landslide in Guizhou Province, China. The results demonstrate that the method can provide reliable volume estimates with relatively low computational cost and short processing time, making it suitable for rapid post-disaster emergency assessment.

2. Materials and Methods

2.1. Overview of the Landslide

The 2019 Shuicheng landslide occurred at 21:20 on July 23 in Pingdi Village, Jichang Town, Shuicheng County, Guizhou Province. A sudden rainstorm was the primary trigger of this disaster (Figure 2). In the six days before the disaster, the precipitation even reached 189 mm. Triggered by intensive rainfall [19], the event buried 21 houses, caused 43 fatalities and 9 missing persons, and resulted in 190 million yuan in direct economic losses [20]. The local stratigraphy consists of Permian Emeishan basalt interbedded with tuff and claystone, with weathered, water-sensitive layers forming potential sliding surfaces. The “steep-gentle-steep” topography and inadequate drainage further exacerbated slope instability [21].

The Shuicheng landslide was listed among China’s top ten natural disasters for 2019 by the Ministry of Emergency Management, highlighting the significant risk of geological disasters in the mountainous regions of southwestern China. The landslide caused severe damage to infrastructure and local communities. As shown in Figure 3 (the original image is taken from Google Earth pro 7.3), the remote sensing images before and after the landslide are very different, reflecting the destructiveness of this disaster. The lessons learned from this event have accelerated the adoption of early warning technologies and improvements in disaster prevention systems, establishing it as a critical case study for geological disaster mitigation in China.

2.2. Overall Framework

The core of this study is a GIS-based method for rapid landslide volume estimation through the reconstruction of the sliding surface. The methodology integrates Geographic Information System (GIS) spatial analysis, digital terrain modeling, and numerical computation principles to overcome the timeliness, safety, and cost limitations of traditional field surveys. The workflow (Figure 4) consists of four main phases: (1) Data Collection, (2) Data Preprocessing, (3) Sliding Surface Reconstruction, and (4) Volume Calculation. Key procedures include the acquisition and refinement of geospatial data (DEM and landslide boundary), estimation of the sliding surface inclination, fitting of spline curves to model the subsurface geometry, and final volume computation via spatial integration. Among these steps, sliding surface reconstruction constitutes the central component and primary methodological contribution of the proposed workflow.

2.3. Data Collection and Processing

The method relies on two primary types of open-source geospatial data:

Digital Elevation Model (DEM): A DEM is a raster representation of surface topography, structured as a matrix of elevation values. The post-landslide DEM data used in this study were obtained from the Geospatial Data Cloud platform of the Chinese Academy of Sciences (https://www.gscloud.cn/). The DEM dataset has a spatial resolution of 30 m and was used to represent the post-event terrain morphology of the Shuicheng landslide area. This resolution range is considered sufficient for capturing the overall morphology of medium- to large-scale landslides while maintaining efficient processing performance during emergency response. These data form the foundation for constructing the 3D model and subsequent volume calculations.

Landslide Boundary: The areal extent of the landslide deposit is manually delineated using high-resolution optical imagery from Google Earth and exported as a Keyhole Markup Language (KML) file. This file defines a vector polygon enclosing the landslide. According to previous studies, the accuracy of Google Earth’s high-resolution image files in the horizontal position is enough to evaluate medium-resolution remote sensing products across most of the world’s peri-urban areas [22].

Data preprocessing is performed in ArcMap (version: 10.8; release year: 2019) to integrate these datasets and focus the analysis. The KML file is first converted to a Shapefile (SHP) format. The core preprocessing step involves using this landslide boundary polygon as a “mask” to clip the original, larger DEM raster. This spatial overlay operation, mathematically represented as extracting values where (X(i, j), Y(i, j)) ∈ Polygon, significantly reduces data volume, eliminates irrelevant terrain, and enhances the computational efficiency [23,24]. The output is a cropped DEM that perfectly aligns with the landslide boundary.

2.4. Sliding Surface Modeling

As mentioned, the core objective of this study is to reconstruct the subsurface sliding surface using processed geospatial data. However, different surface models may produce substantially different fitting performances and volume estimates. Therefore, it is necessary to test which surface type best simulates real sliding surfaces.

To determine which geometric surfaces better resemble real sliding surfaces, the entire sliding surface point cloud is fitted to different geometric surfaces, and the results are compared. Three geometric surfaces are selected: a plane, a quadratic surface and an irregular surface, represented in three-dimensional space by specific equations.

The mathematical expressions of the planar and quadratic surfaces are straightforward and can be derived directly from standard geometric formulations, while the construction of irregular surfaces is relatively indirect. In this study, we use the cubic spline curve as a control line and perform spatial interpolation to construct an irregular surface. The cubic spline curve is a segmented cubic polynomial with second-order continuity at the junctions. Dividing the sliding surface data into countless intervals according to the x or y direction, the cubic spline function in each interval can be defined.

To evaluate the fitting performance of different geometric surface models under controlled conditions, a series of indoor simulation experiments were conducted in the laboratory. An adjustable inclined platform was constructed to reproduce varying slope conditions, with inclination angles systematically set to 15°, 20°, 25°, 30°, 35°, and 40°. Three representative sliding surface geometries—including a planar surface, a quadratic curved surface, and an irregular concave surface—were designed and fabricated using 3D printing technology. These base models were sequentially mounted onto the inclined platform to simulate different types of potential sliding surface morphologies commonly observed in natural landslides. A homogeneous sandy soil layer with uniform thickness and moisture content was then evenly distributed above each base structure to establish a simplified pre-failure slope model. Landslide initiation was induced through a combination of controlled mechanical vibration, artificial rainfall simulation, and gradual surcharge loading. These disturbances were applied until visible deformation and sliding occurred. During the experiment, the deformation process and sliding evolution were continuously monitored and recorded using a high-speed camera system. After slope failure, the overlying displaced material was carefully removed to expose the basal sliding surface. A three-dimensional laser scanner was subsequently employed to acquire dense spatial point-cloud data from the exposed surface, from which 50–100 uniformly distributed representative measurement points were extracted as reference datasets for geometric fitting analysis. For each inclination scenario, the experimentally measured sliding surface coordinates were also fitted using three different surface reconstruction approaches: plane fitting, quadratic surface fitting, and spline-based irregular surface fitting (according to their respective equation expressions).

The fitting performance of each method was quantitatively evaluated using the coefficient of determination (R²) and root mean square error (RMSE). To ensure the robustness and repeatability of the experimental results, each test condition was repeated multiple times under identical settings, and the corresponding datasets were statistically analyzed.

Root mean square error (RMSE) and R² are metrics used to evaluate the quality of a fit. RMSE is a common measure for assessing the difference between the actual data and the results of the fit, while R² serves as another standard indicator for determining the quality of the fit. The formula of the two is as follows:

$$\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}=\sqrt{{\displaystyle \frac{{\displaystyle {\sum }_{\mathrm{i}=1}^{\mathrm{n}}{(\stackrel{\wedge }{{\mathrm{z}}_{\mathrm{i}}}-{\mathrm{z}}_{\mathrm{i}})}^2}}{\mathrm{n}}}}$$

(1)

$${\mathrm{R}}^2=1-{\displaystyle \frac{{\displaystyle {\sum }_{\mathrm{i}=1}^{\mathrm{n}}{(\stackrel{\wedge }{{\mathrm{z}}_{\mathrm{i}}}-{\mathrm{z}}_{\mathrm{i}})}^2}}{{\displaystyle {\sum }_{\mathrm{i}=1}^{\mathrm{n}}{(\stackrel{\wedge }{{\mathrm{z}}_{\mathrm{i}}}-\stackrel{-}{{\mathrm{z}}_{\mathrm{i}}})}^2}}}$$

(2)

where $\stackrel{\wedge }{{\mathrm{z}}_{\mathrm{i}}}$ is the Z coordinate in the input data, and ${\mathrm{z}}_{\mathrm{i}}$ is the Z coordinate corresponding to the fitting result point. Here, ${\mathrm{z}}_{\mathrm{i}}$ represents the measured elevation value of the experimental sliding surface obtained from the 3D laser-scanned point-cloud dataset, while $\stackrel{\wedge }{{\mathrm{z}}_{\mathrm{i}}}$ denotes the corresponding elevation predicted by the fitted geometric surface model at the same spatial location (xi, yi).

By comparing RMSE and R², the optimal geometric surface for fitting the sliding surface can be selected.

Overall, irregular surfaces exhibited superior fitting performance compared with the other two models. However, it should be emphasized that the reconstructed sliding surface is not generated through unconstrained mathematical interpolation alone. Instead, the reconstruction process is jointly constrained by multiple geomorphological factors, including landslide boundary geometry, topographic continuity, depletion-area morphology, and DEM-derived slope characteristics. These constraints reduce the likelihood of physically unrealistic local surface oscillations and ensure that the reconstructed geometry remains consistent with the overall landslide evolution pattern. Therefore, the spline-based surface should be interpreted as a geomorphologically constrained approximation for rapid engineering-scale volume estimation under limited-data conditions, rather than a uniquely determined geological sliding interface.

After selecting the best-fitting surface, the processed data from ArcMap are imported into MATLAB (version: R2024a; release year: 2024) for sliding surface construction. The main principles and steps are as follows:

Data input and preprocessing: The processed regional Digital Elevation Model (DEM) in TIFF format and the landslide boundary defined in KML format are imported into MATLAB and converted to a consistent projected coordinate system (e.g., WGS 1984 UTM) to maintain spatial consistency. The DEM provides a topographic representation of the landslide area, while the KML file delineates its spatial extent.

Generating grid points: Grid points located within the landslide boundary are extracted to serve as computational units. The objective of this step is to estimate the depth of the sliding surface beneath each grid point.

Building a vertical profile: The highest elevation point (typically the landslide crown) and the lowest elevation point (typically the landslide toe) within the boundary are automatically identified. The line connecting these points is projected onto the horizontal plane to define the main sliding direction. For each grid point inside the boundary, a vertical profile perpendicular to this direction is constructed. This profile intersects the landslide boundary at two locations—left and right endpoints. The elevation values at these intersections are derived by interpolating the original DEM data using the gridfit function.

Spline curve fitting: To reconstruct the sliding surface geometry, the tangential inclination at both ends of each profile must be specified. Three potential strategies can be adopted depending on data availability and geological constraints: Method 1: Apply fixed average slope angles for the left and right profile sections. Method 2: Use inverse distance weighting to assign nonlinearly varying inclinations from the crest to the toe, based on four user-defined slope angles. Method 3: Assume a circular arc geometry, resulting in a linear variation in inclination, also based on four input slope angles. In this study, only Method 1 was implemented and validated, while Methods 2 and 3 are proposed as theoretical extensions for future application under data-rich conditions. Then, the slope angles are converted into tangential slopes for spline construction. Combined with the endpoint coordinates, these parameters define a cubic spline curve (Figure 5). The horizontal position of each grid point along the profile is substituted into the spline equation to compute the corresponding sliding surface elevation.

3D sliding surface generation: By iterating over all grid points within the landslide area and applying the spline fitting procedure, a discrete 3D point set representing the sliding surface is obtained. This point set is then spatially interpolated using the gridfit function to generate a continuous DEM of the reconstructed sliding surface.

Following these steps, the hidden sliding surface beneath the terrain is reconstructed, resulting in a sliding surface DEM, while the original ground surface DEM is already available. But it should be noted that the post-failure terrain surface does not uniquely determine the exact geometry of the original sliding surface. Therefore, the proposed method does not attempt to reconstruct a deterministic geological interface, but rather to generate a geometrically and physically plausible approximation constrained by the landslide boundary, terrain morphology, and slope continuity. The spline-based interpolation strategy is adopted to produce a smooth and continuous subsurface surface consistent with the overall geomorphic characteristics of the landslide depletion zone, thereby providing a practical solution for rapid post-disaster volume estimation under limited-data conditions.

2.5. Volume Calculation

Once the sliding surface DEM (Z_surf_fail) and the post-landslide terrain DEM (Z_topo) are obtained, calculating the landslide volume becomes a straightforward process. The volume of the landslide is computed by subtracting the sliding surface DEM from the post-landslide DEM and summing the volume contributions over all grid points within the landslide boundary. The total volume V is calculated as:

$${\mathrm{V}}_{\mathrm{t}\mathrm{o}\mathrm{t}\mathrm{a}\mathrm{l}}=\mathrm{\Delta }\mathrm{x}\times \mathrm{\Delta }\mathrm{y}{\displaystyle {\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\displaystyle {\sum }_{\mathrm{j}=1}^{\mathrm{m}}\left(\mathrm{z}\_\mathrm{t}\mathrm{o}\mathrm{p}\mathrm{o}\left(\mathrm{i},\mathrm{j}\right)-\mathrm{z}\_\mathrm{s}\mathrm{u}\mathrm{r}\mathrm{f}\_\mathrm{f}\mathrm{a}\mathrm{i}\mathrm{l}\left(\mathrm{i},\mathrm{j}\right)\right)}}$$

(3)

where Δx and Δy are the cell sizes of the grid in the X and Y directions, respectively. Z_topo(i, j) is the elevation of the original terrain, and Z_surf_fail(i, j) is the elevation of the sliding surface computed via spline curve interpolation. This operation effectively calculates the content between the two surfaces, providing an estimate of the main body volume.

3. Results

This study primarily utilizes Google Earth and the Geospatial Data Cloud for data acquisition. KML data supply boundary information, while DEM data provide elevation details [25]. The gathered data are analyzed and processed using ArcMap. Raster cropping is employed to refine the data and extract relevant features (Figure 6).

Because the present study relied primarily on post-event remote sensing data and lacked detailed in-situ geological measurements, the dynamic inclination models (Methods 2 and 3) could not be reliably implemented. Methods 2 and 3 require spatially variable inclination constraints or geometric assumptions (e.g., circular arc parameters), which rely heavily on in-situ measurements or subsurface geological information. However, in this study, only post-event DEM data are available, which primarily reflect surface morphology rather than the geometry of the underlying sliding surface. As a result, the key parameters required for Methods 2 and 3 cannot be reliably determined, leading to significant uncertainty.

Method 1 only needs an average inclination, which can be estimated by a slope diagram based on DEM data. Therefore, the constant inclination approach (Method 1) is adopted as a more robust and practical solution under data-limited conditions. At the same time, for the 2019 Shuicheng landslide, field photographs and post-event analyses indicate that the landslide occurred along a single, continuous sliding surface with minimal spatial variation in inclination. Consequently, the constant average slope angle strategy (Method 1) is adopted.

To determine the inclination constraints for Method 1, slope analysis was performed on the cropped DEM in ArcMap using the standard “Slope analysis tool”. Representative average inclination angles for the left and right profile sections were then estimated through visual interpretation of the generated slope map, yielding values of 27° and 30°, respectively.

Based on indoor experiments, irregular surfaces are found to provide a better simulation for sliding surfaces. As shown in

Table 1. Statistical results of fitting experiments.

Landslide Inclination Angle (°)	Plane Fitting		Quadratic Surface Fitting		Irregular Surface Fitting
	R2	RMSE	R2	RMSE	R2	RMSE
15	0.90	0.19	0.91	0.18	0.92	0.16
20	0.85	0.25	0.92	0.11	0.95	0.11
25	0.72	0.38	0.95	0.10	0.98	0.08
30	0.58	0.52	0.97	0.09	0.99	0.06
35	0.45	0.68	0.98	0.08	0.99	0.06
40	0.35	0.82	0.99	0.10	0.99	0.05

, irregular surfaces show smaller RMSE values, indicating lower fitting errors, and their R² values are closer to 1, demonstrating a stronger explanation of slip surface morphology. Therefore, irregular surfaces have been chosen as the optimal fitting surfaces for the proposed method. In the next step of the sliding surface reconstruction, spline curve fitting is utilized and the 3D model of this landslide is then created (Figure 7).

The indoor experiments (Table 1) function as a methodological validation step for selecting the most appropriate surface fitting strategy, whereas Figure 7 demonstrates the application of the selected reconstruction method to an actual landslide case. The laboratory experiments therefore provide the theoretical and experimental justification for the spline-based reconstruction approach employed in the real-world volume estimation workflow.

Finally, the estimated landslide volume is approximately 1.58 × 10⁶ m³, covering an area of 4.0 × 10⁵ m².

4. Discussion

4.1. Error Analysis

The proposed method estimated the volume of the 2019 Shuicheng landslide to be approximately 1.58 × 10⁶ m³. Compared to the previous survey data of 1.50 × 10⁶ m³ [26] reported by the China Earthquake Administration (an authoritative official institution in China), the relative error is approximately 5%. The approximately 5% relative error indicates good agreement between the proposed method and the authoritative field-survey result. These results demonstrate the feasibility and practical applicability of the proposed method. Compared with conventional geometric method or DEM differencing method, the official survey incorporates substantial in-situ geological evidence and engineering investigation, making it the most credible reference currently available for this event.

The observed error originates from multiple sources, which are analyzed as follows:

1. DEM Resolution Limitations: The spatial resolution of the DEM is one of the key factors affecting the accuracy of the reconstructed sliding surface and the final volume estimation. Higher-resolution DEMs generally provide a more detailed representation of local terrain variations, particularly near landslide scarps, boundaries, and depletion zones, thereby improving the precision of spline interpolation and surface reconstruction. In contrast, coarse-resolution DEMs may smooth critical geomorphological features and reduce the reliability of inclination estimation and volume calculation. However, the relationship between DEM resolution and estimation accuracy is not strictly linear. Extremely high-resolution DEMs may introduce local noise, vegetation-related artifacts, and increased computational burden without proportionally improving the final estimation accuracy, especially for medium- to large-scale landslides [27]. Since the primary objective of this study is rapid post-disaster assessment under emergency conditions, a balance between data accessibility, computational efficiency, and acceptable accuracy is required.

In this study, open-source DEM data with a spatial resolution of 30 m were adopted because they are rapidly obtainable and sufficiently capable of representing the overall morphology of the Shuicheng landslide. Although the DEM resolution is generally adequate for rapid assessment purposes, it may still introduce minor uncertainties in local terrain representation.

2. Subjectivity in Boundary Delineation: The landslide boundary was manually interpreted from optical imagery. An underestimation of the accumulation area directly leads to a proportional underestimation of the calculated volume [28]. Conversely, an overestimated boundary would incorporate stable terrain, inflating the volume.

3. Uncertainty in Inclination Estimation: The slope angles (27° and 30°) for the spline constraints were derived empirically from the post-event slope map. A sensitivity analysis was conducted by varying these angles by ±5°. The results indicated that this variation could propagate to a volume error of up to ±8%, highlighting the critical influence of this parameter and the need for more objective constraint methods in future applications.

4. Algorithmic Limitations: The slope algorithm in ArcMap, based on a finite difference method, is sensitive to DEM noise. Furthermore, the volume integration assumes a planar surface within each grid cell, potentially overlooking intra-cell topographic variations and introducing minor systematic errors.

4.2. Method Advantages

One notable aspect of the proposed method is that it reconstructs the sliding surface directly from post-event geomorphic information, thereby reducing the dependence on pre-event topographic datasets. Unlike empirical area-volume scaling laws that treat the landslide as a black-box entity, the proposed method leverages the boundary points as ‘geomorphic anchors’ that constrain the three-dimensional spline interpolation. This approach essentially treats the landslide boundary not merely as a spatial limit, but as a critical transition zone where the pre-sliding topography and the basal sliding surface converge, providing a physically constrained basis for subsurface reconstruction in the absence of pre-event data.

Cubic spline interpolation provides greater flexibility for representing natural slope heterogeneity than simplified geometric surfaces such as planar or quadratic models. The second-order continuity inherent in cubic splines ensures a smooth, mathematically rigorous transition between the depletion and accumulation zones, effectively simulating the realistic, non-linear curvature of sliding surfaces in weathered geological units like basalt. This flexibility allows the model to adapt to localized topographic variations.

The principal advantage of this method lies in its effective balance between computational efficiency and acceptable accuracy, making it particularly suitable for rapid volume estimation during emergency response operations. The entire workflow, from data acquisition to final estimation, was completed within one hour on a standard desktop computer, demonstrating substantial efficiency advantages over conventional field investigations.

Compared with existing methods, the proposed method addresses a major limitation frequently reported in previous studies: the dependence on pre-event topographic data and oversimplified geometric assumptions. Multi-temporal DEM differencing methods based on UAV photogrammetry or LiDAR can achieve high accuracy, but they generally require pre-event DEM datasets and substantial data-processing efforts. Empirical area-volume relationships and simplified geometric approximations are computationally efficient, yet they often fail to represent the complex three-dimensional morphology of individual landslides. In contrast, the proposed method utilizes only post-event open-source geospatial data while preserving terrain continuity and sliding-surface morphology, thereby providing a balance between operational efficiency and acceptable estimation reliability. Furthermore, the method demonstrates superior accuracy compared to other rapid estimation techniques when applied to this case. We also used two other estimation methods to calculate the volume of the Shuicheng landslide, and the results can better reflect the superiority of the proposed method. In contrast, a volume of 2.0 × 10⁶ m³ was obtained by approximating the landslide body with a semi-ellipsoid (geometric method), resulting in a 34% error due to oversimplification of the complex morphology. Meanwhile, an idealized reconstruction using multi-temporal DEM principles—simulated for comparison in the absence of pre-event data—yielded 1.05 × 10⁶ m³, a 30% underestimation consistent with documented pitfalls of the method, such as the inability to account for material that did not fully evacuate the source area.

The robustness of the sliding surface reconstruction is strongly supported by the fitting experiments summarized in Table 1. The performance of the plane fit deteriorated significantly with increasing slope angle (R² dropping from 0.90 to 0.35, RMSE rising from 0.19 to 0.82), validating our premise that simple geometric assumptions are inadequate for complex landslides. While the quadratic surface performed well, the irregular surface based on spline interpolation consistently achieved the best fit (R² up to 0.99, RMSE as low as 0.05) across most inclination scenarios, justifying its selection as the most robust model for capturing realistic sliding surface geometries. The 3D visualization in Figure 7 provides an intuitive validation of the reconstructed surface, showing a continuous and geologically plausible concave morphology that aligns with typical sliding patterns in weathered basalt slopes.

The method’s reliance on open-source data (Google Earth, Geospatial Data Cloud) and standard software (ArcMap, MATLAB) ensures high operational feasibility, low cost, and minimal demand for specialized equipment or hazardous fieldwork. These characteristics indicate the potential applicability of the proposed method for rapid post-disaster assessment during the critical emergency-response period.

4.3. Limitations

This method is primarily designed for and most effective against landslides characterized by a single, continuous sliding surface and clearly defined boundaries. Its application to more complex sliding mechanisms is limited.

The key limitation is the assumption of a single, continuous sliding surface. This assumption may not adequately represent progressive or highly complex sliding mechanisms, such as those with multiple, superimposed shear surfaces often observed in creeping landslides or deep-seated compound slidings. Applying the method in such contexts risks oversimplifying the internal deformation and yielding inaccurate volume estimates.

Furthermore, the method also does not explicitly account for structural controls on slope stability, such as faults, joints, or bedding planes, which can dominate sliding mechanics in geologically complex terrains.

4.4. Future Prospects

Future research will focus on enhancing the method’s robustness. Key directions include: introducing automated constraints for terrain feature guidance, enhancing the constraint mechanism of geological structure, and utilizing AI technology to optimise the links that require manual operation.

• Introducing automated constraints for terrain feature guidance: At present, the landslide boundary and slope constraint (Methods 1–3) still have a certain degree of human participation and subjectivity. To minimize the subjective bias inherent in manual slope constraint selection, future iterations of this methodology could integrate a terrain-aware optimization algorithm. By analyzing the longitudinal profiles and cross-sectional curvatures of the stable terrain flanking the landslide boundary, the algorithm can automatically derive the most probable tangential inclinations for the spline interpolation. This transition from user-defined parameters to geomorphometric-driven constraints would significantly enhance the objectivity and repeatability of the volume estimation, particularly in complex terrains where representative slope angles are difficult to ascertain through visual inspection alone.
• Enhancing the constraint mechanism of geological structure: While the current spline-based reconstruction effectively captures the general concave morphology of the sliding surface, its purely geometric nature can be further refined by incorporating deterministic geological constraints. For rock-slope slidings, the sliding surface is frequently governed by pre-existing structural discontinuities such as bedding planes, joints, or faults. By embedding these structural orientations as localized curvature constraints within the spline function, the model could transition from a purely mathematical interpolation to a geologically constrained simulation. This integration would be particularly beneficial for structural landslides where the sliding surface exhibits high anisotropy and does not conform to idealized circular or ellipsoidal geometries.
• Utilizing AI technology to optimise the links that require manual operation: Combined with the current popular AI technologies (such as machine learning [29,30] and deep learning), many links can be optimised and the processing efficiency can be further improved. For instance, developing a Machine Learning (ML) module trained on extensive regional landslide inventories to automate the selection of spline boundary conditions will transform this framework from a semi-automated GIS tool into a fully autonomous, intelligent system for real-time disaster assessment.

5. Conclusions

This study developed and validated a GIS-based method for rapid landslide volume estimation through sliding surface reconstruction. Applied to the 2019 Shuicheng landslide, the method produced a volume estimate of 1.58 × 10⁶ m³ with an error of approximately 5% relative to official survey data. The proposed approach eliminates the dependence on pre-event topographic data by reconstructing the sliding surface using open-source post-event geospatial information and spline interpolation. Compared with conventional methods, the proposed workflow provides a practical balance between computational efficiency and estimation reliability, making it particularly suitable for rapid post-disaster assessment during the “golden 72 h”. Future work will focus on incorporating multi-source geospatial constraints and AI-assisted optimization to improve the applicability of the method to complex landslide scenarios.