Characteristics of Rock Avalanche Deposit in Wangjiapo, Ludian Based on UAV Aerial Image Recognition

Abstract

Rock avalanche disasters in alpine and gorge regions are frequent and large in scale and cause severe damage. The movement of a rock avalanche is complex and has not been fully studied. The deposits of a rock avalanche can provide valuable insights into its movement process, which is crucial in understanding the rock fragmentation mechanism and predicting disaster-affected areas. Taking the Wangjiapo rock avalanche in Yunnan Province of China as an example, the size, shape and distribution characteristics of the deposit were analyzed based on field surveys, unmanned aerial vehicle (UAV) photography and image recognition technology. Initially, 3062 deposited rock blocks were manually measured in the field. Subsequently, the Particles/Pores and Cracks Analysis System (PCAS) was employed to identify 11,357 rock blocks with an area greater than 0.1 m² from UAV orthophotos. By comparing the characteristics of the rock blocks obtained through image recognition and manual measurement, the statistical analysis of UAV aerial imagery combined with PACS proved feasible in studying the Wangjiapo rock avalanche. The results showed that the rock block movement was accompanied by fragmentation and sorting processes; furthermore, the roundness increased with the migration distance. Small blocks were more prevalent at the foot of the slope, while irregularly shaped, large blocks dominated in source areas. The movement of huge blocks was characterized by significant potential energy-driven features and inertia advantages, allowing them to travel farther than smaller blocks, and they tended to be concentrated in the central area of the deposit. Additionally, affected by the cementation degree of breccia and the topography, the blocks in the eastern and western deposit areas exhibited different fragmentation and deposition characteristics.

1. Introduction

Rockfall is a geomorphic process characterized by the rapid descent of soil or rock masses on steep slopes due to gravity and external forces [1,2]. A rock avalanche is a specific type of rockfall triggered primarily by seismic loading, rainfall or freeze–thaw processes [3]. In terms of geological disasters and scientific research, catastrophic rock avalanches with volumes exceeding 1 × 10⁶ m³ are of interest due to their large volumes, rapid instability, high mobility and serious consequences [4,5,6]. For instance, an earthquake-induced rock avalanche in the Andes of Peru in 1970 resulted in the deaths of at least 18,000 residents [7]. Similarly, in 1980, a large-scale rock avalanche occurred at the Yanchihe Phosphate Mine in Hubei Province, China, destroying all buildings and pithead facilities of the mining bureau within 16 s and killing 284 people. Another instance occurred during the Wenchuan earthquake in 2008, when the Jingjiashan rock avalanche buried Beichuan Middle School, leading to over 700 fatalities [8]. Following a rock avalanche event, a large volume of rock or debris accumulates at the slope foot to form deposition, which provides critical historical data for the study of the dynamic fragmentation processes during rock avalanches [6]. Therefore, the extraction and analysis of the deposit characteristics are particularly important for disaster mechanism research and prevention.

To date, extensive studies have been carried out on rock avalanches. Some scholars have paid attention to the sources of rock avalanches and studied their formation mechanisms and movement processes [9,10,11,12,13]. There are also studies focusing on rock avalanche deposits, examining their size distribution characteristics, stability and the possibility of secondary instability [14,15,16,17]. Rock avalanches behave as granular flows in the process of movement and deposition, which are complicated and difficult to monitor [18,19]. Knowledge of the characteristics of the rock blocks generated in the rock avalanche is helpful to analyze the trajectory, migration distance and impact energy of the rock block, which is also useful for the quantitative risk assessment of rock avalanche disasters [20,21]. The existing literature primarily focuses on deposit characteristics such as the size distribution, shape fractals and fragmentation. The most commonly used methods include theoretical research, field surveys, field tests, model tests and numerical simulations [15,22,23,24,25,26].

Field surveys are the most direct way to obtain the characteristics of rock blocks. However, conducting traditional geological surveys in mountainous areas is challenging due to the complex geographical conditions and difficulties in access to deposits on high and steep slopes. Consequently, it is arduous to comprehensively document the characteristics of entire deposits through manual field surveys given the extensive areas of rock avalanches and the number of blocks involved. Ruiz-Carulla et al. [21] selected some representative sampling plots at a rockfall site to count and measure the blocks, finding that both the number and size of the fragments could be well fitted by a power law distribution, indicating scale invariance in the fragmentation process of rock blocks. Field and model tests can replicate the block movement and deposition process [27,28,29]. However, field tests are costly and impractical for large-scale rock avalanches, while model tests are scaled and struggle to accurately represent the geometric similarities of blocks. Therefore, both test methods have inherent limitations. Volkwein et al. [30] carried out a series of field tests on a single rockfall at Oberalppass, Switzerland and analyzed the trajectory of the rockfall. Lin et al. [31] synthesized a breakable block material with different structural forms for physical model tests aimed at analyzing the impact and fragmentation characteristics of blocks. Numerical simulations are convenient in application and low in cost, capable of simulating the block fragmentation and deposition process and quantitatively analyzing the spatial distribution of deposited blocks. Salciarini et al. [32] employed a discrete element model to simulate the fragmentation effects of collapsed rocks, showing that rock fragmentation significantly impacts the location and extent of deposits. Gao et al. [33] also used a discrete element model to analyze the fragmentation characteristics of the rock mass during migration under seismic action. However, numerical simulations also face challenges in accurately replicating the disintegration, diverse gradation and three-dimensional movement behavior of the rock block, resulting in substantial disparities between the simulated and actual block characteristics.

In summary, field measurement is considered the most reliable and accurate method in obtaining the deposit characteristics among the various research approaches. The need to identify and measure the deposit characteristics accurately and efficiently, without being restricted by the terrain conditions, has been increasingly highlighted. With the development of image recognition technology, particularly the maturity of UAV technology, rapid batch and non-contact measurements of block parameters have been developed rapidly [34,35]. Crosta et al. [36] studied the block size distribution in landslide deposits through photography, tracing and image processing techniques. UAV photogrammetry extends the traditional local block statistical method relying on measurement window photography to the recognition of the entire deposit [11,37]. Using photography with centimeter-level precision, enabled by UAV and efficient recognition technology, spatial location and geometric data for thousands of blocks within the deposit area can be quickly obtained. Traditional image recognition algorithms, such as the watershed, region growing and edge detection algorithms, have been widely used for particle boundary extraction and recognition [38,39,40]. Various tools are available for particle feature recognition and analysis, such as MATLAB (R2024a), Split-Desktop (v4.0.0.42), BASEGRAIN (v2.2), ImageJ (v1.8.0.345) and the Particles/Pores and Cracks Analysis System (PCAS, v2.324) [41,42,43,44,45]. Peng et al. [46] analyzed the particle size distribution in landslide deposits and their relationship with house damage using the PCAS based on UAV aerial images. Sun et al. [16] developed a block photo-analytical procedure utilizing Split-Desktop, ImageJ and ArcGIS to successfully identify rock avalanche deposits. In recent years, significant progress has been made in image recognition through deep learning technologies, with models such as convolutional neural networks (CNNs) and recurrent neural networks (ANNs) being extensively applied to particle recognition tasks [47,48,49,50].

On 3 August 2014, a strong earthquake with a magnitude of 6.5 struck Ludian, Yunnan Province, China. A large number of rockfalls and landslides were triggered by this earthquake [51], including the significant Wangjiapo rock avalanche, which resulted in six deaths and the destruction of nearly 20 houses. The Wangjiapo site is highly typical, with abundant and well-documented historical data, making it an excellent case study for quantitative research on rock avalanche deposits. The source and deposit areas of the rock blocks are well preserved, enabling comparative analysis. This study focused on the Wangjiapo rock avalanche and utilized field surveys, UAV photography and image recognition technology to analyze the characteristics of the deposited blocks. Initially, the size of some blocks was manually measured in the field, and a high-resolution image was generated by UAV aerial photography. Subsequently, the PCAS was employed to obtain surface information from the deposits, and the feasibility of the PCAS statistics was analyzed by comparing the image recognition results with the manual measurements. Finally, the size, shape and fragmentation of the deposits were analyzed. This method provides a more efficient means for the investigation of rock avalanche deposits.

2. Study Area

2.1. Geological and Environmental Background

The study area is located in Yunnan Province, China, situated in the upper reaches of the Niulan River, a tributary of the Jinsha River on the southeastern margin of the Qinghai–Tibet Plateau (Figure 1). The region features high mountain valleys and river erosion deposition landforms, predominantly influenced by structural erosion and karst dissolution. The Niulan River within this area generally flows from southeast to northwest, with a riverbed width of approximately 100 m. The valley is deep and steep, and the incision depth exceeds 1000 m. The area has a distinct plateau monsoon climate with significant vertical climatic variation due to the terrain. The temperature difference between seasons is not large, but the dry and wet seasons are clear. The average annual temperature is 15 °C, and the annual precipitation ranges from 780 to 1010 mm, which is concentrated between May and October.

The strata in the study area exhibit well-developed geological formations ranging from the Sinian to Quaternary periods, with the Paleozoic being the most widespread, followed by the Mesozoic. The bedrock in the valley is predominantly exposed, and prominent outcrops consist of Cambrian, Ordovician, Devonian, Permian and Quaternary strata. The fault structure is highly developed within this region; among them, the Baogunao–Xiaohe fault exerts a significant influence on the regional structural stability, and it was also the seismogenic fault of the Ludian earthquake. This fault is of the left-lateral strike-slip type and directly cuts through the Wangjiapo slope (Figure 1).

2.2. Basic Characteristics of Wangjiapo Rock Avalanche

The Wangjiapo slope is situated on the right bank of the Niulan River, where the river changes direction from north–south to east–west, and the slope is at the arc of the river bend (Figure 1). During the Ludian earthquake, Wangjiapo was subjected to intense seismic shaking, triggering a rock avalanche and a secondary landslide at the lower right side of the slope toe (Figure 2). The northwestern region of the rock avalanche top is a strong deformation area, characterized by dense surface fractures, and is a significant threat to the slope’s stability. The elevation of the Wangjiapo rock avalanche’s top was approximately 1870 m, with the slope foot at around 1250 m, resulting in an elevation difference of about 620 m. The slope is relatively steep, with an overall gradient ranging from 40° to 55°. Comparing Figure 2a,b, the rock avalanche buried the houses and roads below and severely damaged the vegetation on the slope.

According to the digital elevation model before and after the rock avalanche, it can be calculated that approximately 10⁶ m³ of fragmented breccia detached from the source area. The deposits were classified into three regions based on the slope terrain and the source material of the rock avalanche (Figure 2b). A representative rectangular range with a side length of 100 m was selected in each area, and the distributions of the rock blocks are shown in Figure 3. The upper part of the slope served as both the primary source area for the Wangjiapo rock avalanche and a partial deposit area. Following the occurrence of the rock avalanche, a smooth back wall with a slope of 60° to 70° formed at the top of the slope, and nearly a third of the material was deposited on the platform below this back wall to form deposit I. The deposit material was mainly massive karst breccia composed of Permian limestone, which corresponded to exposed strata. The breccia cement was primarily composed of carbonate, with calcite as the main component. In deposit I, the bedrock near the surface was subjected to strong tectonic and karstification, and they were cut into huge blocks and subsequently deposited nearby. Therefore, the migration distance of these rock blocks in this area was relatively short. The block presented different colors, affected by the degree of weathering. As shown in Figure 3a, the gray–white blocks are less weathered and better cemented, while the red blocks are severely weathered and poorly cemented. Under the gentle platform of deposit I was a steep cliff, mainly exposing Ordovician gray quartz sandstone. In addition to the breccia at the slope top, substantial amounts of quartz sandstone also collapsed under the action of the earthquake. The slope surface was divided into deposits II and III by two curved gullies on either side, which restricted the lateral dispersion of the blocks. Deposit II was distributed on the eastern slopes, and deposit III occupied western slopes, where the block migration distances were the greatest. The main materials in deposits II and III were karst breccia and quartz sandstone. Although the orthographic image shown in Figure 3 only allows for the observation of the surface rock blocks, two distinct sedimentary layers could be discerned vertically during the field surveys in deposit III. The upper layer was predominantly composed of large limestone breccia blocks, while the lower layer was characterized by a fine matrix with the minor presence of coarse blocks.

3. Methods

3.1. Manual Field Measurement of Rock Blocks

The size distribution characteristics of the rock blocks directly influence the migration and stability of deposits, rendering them a pivotal attribute of rock avalanche formations. From March to April 2021, a field survey was conducted on the Wangjiapo rock avalanche, during which the block sizes were manually measured. Due to the large number of deposited blocks, only those with a long axis exceeding 1 m were measured, and the lengths of the three axes of these blocks were recorded. For the deposits in the middle and upper parts of the slope, particularly in deposit I, the steep terrain made it difficult for the investigators to access these areas, so measurements were only conducted for the blocks in the lower parts of deposits II and III. For the blocks in the upper slope and those with a size smaller than 1 m, UAV aerial photography combined with image recognition technology could be employed to acquire essential data.

3.2. UAV Aerial Photogrammetry

UAV aerial photogrammetry is a remote sensing method that involves using a UAV as a flight platform, equipped with sensor devices to obtain ground information (Pajares, 2015) [52]. The DJI Matrice 200 UAV (DJ-Innovations, Shenzhen, China) was employed for aerial photography in this study. The process primarily encompassed flight path planning, aerial photography and control point measurement. The flight paths were designed considering the survey area, terrain, camera performance and required scale. To meet the data resolution requirements, the UAV was flown with a forward overlap ratio greater than 80% and a side overlap ratio greater than 60%. To obtain more comprehensive image data, both vertical and oblique photogrammetry techniques were employed. After capturing the aerial images, the Agisoft Photoscan software (v1.7.4) was used to perform aerotriangulation, generate point clouds, establish TIN models, map textures and create a 3D realistic model. Subsequently, processing the model resulted in the generation of a high-resolution digital surface model (DSM) and a high-resolution digital orthophoto map (DOM) at a resolution of 0.05 m. The DSM reflects surface undulations and can be used to generate a terrain profile, while the DOM accurately represents surface cover features, facilitating the extraction of the deposited blocks’ characteristics.

3.3. Image Recognition of Rock Blocks

Given that the field measurement data did not encompass all deposited blocks encountered in this study area, we utilized image recognition technology for a comprehensive statistical analysis of these blocks. The image processing software used was the PCAS, developed by Nanjing University, which is commonly used in the recognition and analysis of particles, pores and cracks. The PCAS processes images by binarizing and vectorizing them to calculate statistical parameters such as the number, length, width, orientation and shape coefficients of particles, pores and cracks [53]. In this study, the PCAS processing workflow included the following steps: (1) DOM preprocessing; (2) image binarization; (3) binary image processing; (4) the identification of rock blocks; (5) statistics of the block size. The detailed process was as follows.

(1)

DOM Preprocessing: Due to the extensive coverage of the Wangjiapo rock avalanche and the high resolution of the orthophotos, the direct processing of the DOM in the PCAS was not feasible due to their large size. Therefore, the DOM were divided into smaller sampling windows. When the block boundaries were unclear, the USM sharpening function of the Agisoft Photoshop software (v1.7.4) was used for boundary processing. The window size used in this study was 50 m × 50 m. Taking the orthophoto of a window in Wangjiapo rock avalanche deposit III as an example (Figure 4a), the identification methods of rock blocks were introduced.

(2)

Image Binarization: The DOM is a true-color image, and it was converted to a grayscale image using the ImageJ software (v1.8.0.345). In binarization processing, a grayscale threshold needs to be identified to distinguish the blocks from the deposition environment. The color of the Wangjiapo rock avalanche deposit was mainly grayish white, the surrounding vegetation was green or red and the voids between blocks were black. Since there was a significant color difference between the deposits and the surrounding environment, the gray threshold could be relatively easily identified. When the grayscale values exceed this threshold, the image pixels are recognized as blocks. This threshold can be automatically identified by ImageJ or manually determined through trial and error. The binarized image is shown in Figure 4b, where white represents blocks and black represents vegetation or voids.

(3)

Binary Image Processing: Since the blocks in the deposit were not all the same color, and due to the interference of environmental factors such as vegetation, the binary images often exhibited incomplete rock recognition, resulting in voids or unclear boundary recognition. Additionally, small blocks or overlapping blocks can also lead to blurred boundaries. Therefore, the manual processing of the binary image was necessary, involving the filling of block voids by deleting black pixels or enhancing the block boundaries by adding black pixels (Figure 4c). In this study, this process was performed using the Photoshop tool.

(4)

Identification of Rock Blocks: The processed binary image was imported into the PCAS for block identification and size statistics. For the local areas where blocks were still overlapping, an erosion operation was performed by setting the element radius to remove small connections between blocks. In this study, employing a minimum recognition area threshold of 50 pixels along with a closure radius of 2 pixels allowed the PCAS to accurately identify blocks with a minimum surface area equivalent to approximately 0.008 m². The identification results for the rock block are shown in Figure 4d.

(5)

Statistics of Block Size: Following the identification of independent rock blocks, statistical parameters such as the number, area, perimeter, form factor, length, width and direction of these blocks were obtained (Figure 4f). These parameters could be imported into the Excel software (v16.0.18025.20030) or other data processing software for further statistical analysis, with a focus on the area parameter in this study. As can be seen from Figure 4d, the blocks at the window’s edge are incomplete due to image segmentation. To address large blocks that were clearly segmented, we initially identified the corresponding area datum for the same block in adjacent windows based on the block ID and then obtained the complete block area through additional processing.

In the orthographic image presented in Figure 4a, a total of 7227 blocks were identified, with a significant proportion being small blocks. It should be noted that when the block size is small (as depicted in Figure 4f), there may be issues related to inaccurate edge detection. To ensure relative data accuracy, only blocks exceeding an area of 0.1 m² were selected for further analysis, with a total of 926 blocks. Several examples of blocks with areas close to 0.1 m² are illustrated in Figure 4f. The statistical analysis included the four blocks marked in red and those larger than them, while excluding the two blocks marked in blue and those smaller than them.

4. Results

4.1. Characteristics of Blocks from Manual Measurement

Field measurements were conducted on blocks with a long axis exceeding 1 m in the lower parts of deposits II and III. A total of 3062 blocks were measured, with data collected on the length, width and height of each block. The maximum value among these three dimensions was designated as the long axis (Dl), the middle value as the intermediate axis (Dm) and the smallest value as the short axis (Ds). As depicted in Figure 5, based on the ratios Dm/Dl and Ds/Dm, which represent the proportions between different axis lengths, four shape categories were identified to classify the Wangjiapo rock avalanche deposits. When both ratios exceed 0.6, the block is nearly spherical, accounting for 69% of the measured blocks, which is the dominant shape. Blocks with Dm/Dl > 0.6 and Ds/Dm < 0.6 exhibited disc-like shapes, while those with Dm/Dl < 0.6 and Ds/Dm > 0.6 resembled cuboids. Disc-like and cuboid-like blocks accounted for 16% and 14%, respectively. Blocks where both ratios were less than 0.6 displayed irregular flaky shapes, which constituted only a minor proportion at approximately 1%.

The volume of the blocks was simplified for calculation based on their respective shapes. The volume of the spherical-like block was determined using the sphere volume formula, with the diameter calculated as the average of its three dimensions. For the disc-like block, its volume was calculated using the cylinder volume formula, with the diameter taken as the average between its intermediate and long axes. As for cuboid-like and irregular flaky blocks, their volumes were simplified according to the cuboid formula, calculated as the product of the lengths in three directions. The cumulative frequency curve of the block volumes measured in the field is shown in Figure 6. The field measurements indicated that the volume of the blocks primarily ranged between 0.1 and 0.5 m³, accounting for 70% of the total. The largest block had a volume of approximately 400 m³; additionally, around 35 blocks exceeded 100 m³ in size and were mainly concentrated in the central regions of deposits II and III.

4.2. Reliability Verification of PCAS Image Recognition Method

Since the field measurement data did not cover all of the deposited blocks, image recognition technology was utilized to extract data for all blocks in the three deposit areas. As shown in Figure 7a, each deposit area was divided into rectangular image sampling windows, with a total of 63 windows. Deposit I included windows numbered 1 to 14, deposit II included windows numbered 15 to 31, and windows numbered 32 to 63 were in deposit III. To validate the reliability of the PCAS image recognition technology, the image-recognized blocks were compared with the field measurement results. A total of 4129 blocks with a longest axis greater than 1 m were identified in deposits II and III. As illustrated in Figure 7b, the number of blocks measured manually was lower than that identified through image recognition because the upper and right boundaries of deposit II were not fully reached during the field survey. When further restricting the image-recognized blocks within the surveyed areas only, 3492 blocks larger than 1 m were identified, still being 430 more than the number obtained from manual surveys alone. This suggests that, compared to manual statistics, image recognition methods can capture more comprehensive data without missing entries.

The area distribution proportions of the blocks obtained by field measurement and image recognition are presented in Figure 8. Since the image recognition method cannot obtain block volumes, their size was represented by their surface area in the orthophoto projection. The surface areas of the blocks measured in the field were calculated based on their length and width. According to Figure 5, blocks with Dm/Dl greater than 0.6 were predominantly spherical or disc-shaped, and their area was calculated as that of a circle. Blocks with Dm/Dl less than 0.6 were generally cuboid-like or irregularly flaky, and their area was calculated as that of a rectangle. Figure 8 demonstrates that blocks with an area between 0.5 and 2 m² were dominant, accounting for 65% of the statistical blocks. Overall, the distribution characteristics and trends of the block areas determined by the PCAS statistics were consistent with those obtained from the manual statistics, indicating that the image recognition method employed herein is reliable when the block size exceeds 1 m. As previously mentioned, the minimum identifiable block in this study was 0.008 m²; however, to ensure data accuracy, only blocks with an area greater than 0.1 m² (equivalent to a block size of approximately 0.5 m) were subjected to analysis. In the Wangjiapo deposits, a substantial number of blocks ranging in size from 0.5 m to 1 m were observed. However, due to the absence of field investigation, the validity of the image recognition method at this scale cannot be directly verified through comparison with manually counted results. According to Figure 4e, when the block area is close to 0.1 m², both the contour and area of the identified block exhibit a high degree of similarity with those of the original block. Consequently, it can be inferred that the image recognition method remains reliable even at a scale as small as 0.1 m². The Wangjiapo rock avalanche deposit contained a large number of smaller blocks. Measuring these manually in the field would be time-consuming and prone to omissions, errors and duplicate records. Thus, compared to manual statistics alone, image recognition offers an efficient and accurate approach for the analysis of blocks across a wider range of sizes.

4.3. Characteristics of Blocks from Image Recognition

4.3.1. The Size Distribution of Blocks in Different Deposit Areas

In the PCAS, the accurate recognition of boundaries may be compromised when the blocks are small and densely packed. To ensure data accuracy, only blocks larger than 0.1 m² were analyzed, totaling 11,357 blocks. Deposit I comprised 2661 blocks, deposit II consisted of 2654 blocks and deposit III contained 6042 blocks. The average block area in deposit I was 4.2 m², with the largest block measuring 560 m². The majority of the blocks in deposit I had an area ranging from 0.1 to 10 m², accounting for 92% of the total. In this deposit, ultra-large blocks over 10 m², fragmented along existing fractures, covered nearly 60% of the surface area. The average block area in deposit II was 2.5 m², with the largest block measuring 500 m², and 96% of the blocks had an area between 0.1 and 10 m². Deposit III was the main deposit area, encompassing 53% of all blocks. The average block area in this region was 1.8 m², with the largest block measuring 320 m². Blocks within the range of 0.1 and 10 m² accounted for 98% of the total. Overall, there was an increase in smaller blocks from deposit I to III, while larger blocks experienced a significant decrease in number. As the distance of block transportation increased, the fragmentation intensified, leading to a reduction in the average block size.

The field surveys revealed that the blocks in deposits I and II exhibited rougher textures and looser structures compared to the denser structure of deposit III, where the matrix and large blocks were more thoroughly mixed. The cumulative frequency curves of the block areas in the three deposit areas are shown in Figure 9. The curves correspond with the field observations, indicating that deposit I displayed a higher frequency of coarse blocks with poor sorting, followed by deposit II, while the blocks in deposit III demonstrated a more uniformly sized distribution.

4.3.2. The Shape Characteristics of Blocks in Different Deposit Areas

The analysis also revealed differences in the block shapes across the different deposit areas. The block shape is closely related to the composition of the source rock, weathering processes and dynamic effects during transportation. Due to the irregularity of block outlines, it is necessary to appropriately characterize the block shape for the analysis of variations. In this study, the overall contour coefficient (α) was used to quantify the block shapes; it can effectively describe the morphological characteristics of particles. The overall contour coefficient is defined as the ratio of the perimeter of a circle with an equivalent area to that of the block and its actual perimeter. For a perfectly circular block, the coefficient α reaches its maximum value of 1. For a given area, the more the shape of the block deviates from a circle, encompassing prominent edges and rough surfaces, the larger its perimeter and the smaller the α value.

The calculations indicated that the α values for the Wangjiapo rock avalanche blocks ranged from 0.27 to 0.98, with the minimum α values in the three deposit areas being 0.27, 0.53 and 0.62, respectively. To visually represent the distribution of the block shapes in each deposit area, a histogram illustrating the block shape distribution is presented in Figure 10. In deposits II and III, approximately 97% of the blocks exhibited an α value greater than 0.8, whereas, in deposit I, this proportion was around 82%, indicating a generally higher degree of roundness for the Wangjiapo rock avalanche blocks. Unlike deposits II and III, deposit I contained a significant number of blocks with lower α values, nearly 400, and they were irregularly shaped and the size was large. This difference is likely due to the fact that, during long-distance transport, blocks undergo collisions with the ground or other blocks, which enhances their roundness. In contrast, the large blocks in deposit I were primarily fragmented along existing fractures and deposited locally, without undergoing the long-distance transport and rounding process.

4.3.3. Longitudinal Variation in Block Size

To analyze the longitudinal distribution characteristics of the blocks in each deposit area, the deposit areas were divided into statistical zones based on the sampling windows shown in Figure 7a, with each zone covering a vertical interval of 50 m. Deposit I was divided into five statistical zones, labeled A1 through A5. The sampling windows in the same horizontal direction constituted a statistical zone. For instance, window 1 represented zone A1, windows 2 and 3 represented zone A2, and windows 4 through 7 represented zone A3. Deposit II was divided into six statistical zones, labeled B1 through B6, and deposit III was divided into ten statistical zones, labeled C1 through C10, following the same division rules as for deposit I.

Since the trajectory of falling rocks is influenced by the slope surface conditions, the DSM data obtained from UAV aerial photography were utilized to generate cross-sectional profiles along the main axes of all three deposit areas, as shown in Figure 11. Overall, there was a decrease in the slope gradient across all three deposit areas, with noticeable reductions observed in the central part of deposit I and lower part of deposit III. Figure 11 also shows the longitudinal zoning and the number of blocks in each zone. In terms of the block longitudinal distribution within these deposits, no clear pattern can be discerned for deposit I, whereas, in deposits II and III, the number of blocks exhibits an initial increase followed by a subsequent decrease vertically. Particularly in deposit III, in the area where the slope flattens (C6 to C9), there is a significant increase in the block count.

The longitudinal variation in the block size was visually displayed in an area percentage histogram, as depicted in Figure 12. The longitudinal variation in the block size differs among the zones. In deposit I, zones A1 and A5 are dominated by small blocks, while the proportion of large blocks is higher in the central zone, with blocks larger than 1 m² comprising up to 81%. In deposit II, the upper zones B1 through B3 are mainly composed of small blocks, whereas the lower zones B4 and B5 have a higher proportion of large blocks. In zone B5, blocks larger than 1 m² reach up to 60%. Small blocks dominate in deposit III. In the upper zone C1, blocks larger than 1 m² account for only 10%; however, this proportion gradually increases longitudinally to reach 38% in zone C8, before decreasing again to 20% in zone C10. Combining Figure 11 and Figure 12, it can be observed that the terrain significantly influences the deposition of large blocks, which tend to deposit more easily on gentle slopes, such as in zones A3 and C8.

4.3.4. Transversal Variation in Block Size

To analyze the horizontal distribution characteristics of the blocks in each deposit area, the areas were divided into horizontal zones based on the longitudinal zones. Longitudinal zones A3, B4 and C7 were selected as representatives of each deposit area, and each was further divided into horizontal statistical zones at every 50 m. Deposit I was subdivided into four horizontal statistical zones, labeled A3-1 through A3-4, corresponding to sampling windows 4 through 7 (Figure 7a). Similarly, deposit II was divided into four horizontal statistical zones, labeled B4-1 through B4-4, corresponding to sampling windows 22 through 25. Deposit III was divided into five statistical zones, labeled C7-1 through C7-5, corresponding to sampling windows 43 through 47. The area percentage histograms in each horizontal zone are presented in Figure 13.

The horizontal variation in the block sizes also differs. In deposit I, there is no clear pattern in the horizontal distribution of the block sizes, with the majority being larger than 1 m². In deposits II and III, the proportion of large blocks is higher in the central regions, while smaller particles less than 1 m² are more prevalent on the periphery. This spatial distribution can be attributed to the lateral dispersion of blocks away from the center during downslope movement. The inertia of blocks decreases with the decrease in size, making smaller blocks more prone to deviation from the central area. Furthermore, as the travel distance increases, dynamic fragmentation becomes increasingly effective, gradually transforming large blocks into smaller ones.

5. Discussion

5.1. The Fragmentation and Sorting of the Blocks during the Movement Process

The movement of blocks during a rock avalanche is an extremely complex process, influenced by the characteristics of the blocks, the slope features and the triggering conditions. During movement, blocks also undergo fragmentation and sorting. The sources of the Wangjiapo rock avalanche were karst breccia and quartz sandstone, which were disintegrated into large and unstable rock masses by fault fracturing. Based on the characteristics of the block deposits, three distinct deposit areas were determined. Deposit I served as both a block deposit and primary source area, where numerous large blocks were deposited locally. By comparing the characteristics of the blocks between deposit I and those in other areas, the fragmentation and sorting characteristics of the blocks during their downward movement from the source area could be analyzed. The PCAS statistical results indicate that the number of blocks in deposit III was significantly higher than that in deposit I, especially smaller ones. The average block area decreased by approximately 55%, and the standard deviation of the area converged by about 50%. These findings suggest that significant fragmentation occurred as relatively intact rock masses fell from the source area, increasing the number of blocks and improving the sorting of block areas.

Since the transport distance of the rock blocks in deposit I was relatively short, this limited our ability to capture the changes in the block characteristics during transportation. Therefore, this discussion primarily focuses on the characteristics of the blocks in deposits II and III. The statistical results show that there are significant differences between the blocks in these two areas. Specifically, there is a significantly smaller number of blocks in deposit II compared to deposit III; however, the proportion of large blocks is higher, resulting in a higher average block area. Additionally, the standard deviations of the block areas are 12.2 and 9.2 for deposits II and III, respectively. Combined with the cumulative frequency curve of the block areas (Figure 9), it becomes evident that the blocks in deposit III are better sorted. These differences in block characteristics between the two deposit areas are mainly influenced by the source rock quality and the topographic conditions along their respective transport paths.

The breccia above deposit II exhibits weak cementation, and long-term weathering has produced a large amount of soil-like fine-grained material in the cemented parts. The buffering effect of this fine material reduces the impact force when the blocks collide with the ground or other blocks. In contrast, the breccia above deposit III is more strongly cemented, characterized by a tightly interlocking fragmental mosaic structure that limits the formation of fine-grained soil-like material between the blocks. Furthermore, the transport path of the blocks in deposit II is short, with a distance of about 500 m from the center of the source area to the bottom of this deposit area. Additionally, the slope of the terrain in this area is relatively gentle, with a gradient of about 50° along its main deposition. On the other hand, the transport path of deposit III is longer, with a distance of about 750 m from the center of the source area to the bottom of the deposit area, and the slope above the deposit area is steeper, reaching 65°. Consequently, the block velocity and energy are higher in deposit III during the descent process, leading to stronger collisions and extensive fragmentation. Moreover, the gullies on the slope also have a significant effect on the movement of rock blocks. Although both deposit areas have gullies, the gully above deposit III is located centrally and narrower in width, with lateral constraints that may limit energy dissipation during rock fragmentation, leading to stronger and longer-lasting interactions between blocks during movement. These factors result in higher fragmentation and the better sorting of the blocks within deposit III compared to deposit II. Notably evident within deposit III is the increased occurrence of scraping and plowing phenomena, which have caused significant vegetation destruction, further indicating heightened destructive energies associated with block movement. Additionally, a secondary landslide developed in the lower part of deposit III (Figure 2b). This landslide occurred subsequent to the Wangjiapo rock avalanche, thus causing the block on the landslide to also advance with the secondary landslide, which was another contributing factor to the extensive migration distances of the blocks within deposit III. However, due to both the obstructive influence of bedrock and the Niulan River at the leading edge of the landslide, there was limited displacement observed in terms of the landslide migration distance.

Longitudinally, the distribution of the blocks in deposits II and III shows an increasing trend in the proportion of large blocks from top to bottom, with the highest proportion observed on gentle slopes. Horizontally, both areas exhibit a higher concentration of large blocks in the central regions, while smaller blocks are more prevalent on the sides. This spatial pattern is likely due to the significant potential energy-driven characteristics of large blocks. In comparison to smaller blocks, large blocks convert more potential energy into kinetic energy during movement, resulting in their deposition at farther distances due to this sorting effect caused by the potential energy differences. Moreover, large blocks have a significant inertia advantage, making it difficult to change their direction of movement, even when colliding with surrounding rocks, whereas smaller blocks with lower inertia are more likely to deviate from their paths upon collision. This sorting effect caused by inertia differences leads to the concentration of large blocks in the central areas of slopes and the greater presence of smaller blocks along the sides. Furthermore, the karst breccia in the source area of Wangjiapo is highly irregular but tends to decompose or fragment into nearly spherical blocks during movement, regardless of the block size.

This study shows that, for rock avalanches in deeply incised valleys in Southwest China, the source material, its cementation degree and weathering degree, and the topography are key factors influencing the range of block movement. These factors should be given special consideration in future predictions of the impact ranges of rock avalanches. Additionally, large blocks possess significant kinetic energy after falling, often posing a considerable threat to downstream residential areas. Therefore, in rock avalanche protection efforts, careful consideration should be given to the movement characteristics of large blocks and effective risk management strategies should be implemented for gentle areas at the base of slopes.

5.2. The Advantages and Limitations of the Image Recognition Method

In the field work for the Wangjiapo rock avalanche, a total of 3062 blocks with a long axis exceeding 1 m were manually measured. Utilizing the PCAS, 4129 blocks with a long axis exceeding 1 m were identified from UAV orthophotos in deposits II and III. Because the right boundary of deposit II was not fully reached during the field measurements, the number of blocks manually measured was lower than that identified through image recognition. The analysis reveals that the cumulative characteristics, trends and distribution patterns of the block sizes identified by the PCAS are generally consistent with those measured in the field, indicating that the PCAS statistics can reflect the basic characteristics of the deposit blocks. In another study by Sun et al. [16], the boundaries of blocks with a long axis exceeding 1 m were manually delineated using Google Earth images of the 2014 Wangjiapo rock avalanche, resulting in the identification of a total of 4298 blocks. The number of blocks identified through this image recognition method is close to the results of manual measurements obtained by Sun et al. [16], further validating the reliability of the image recognition method employed in this study. Due to the fragmentation of blocks during movement, most of the blocks at the Wangjiapo site are small in size. The minimum block size identifiable by the PCAS primarily depends on the UAV image resolution and software-defined minimum recognition area threshold. Higher-resolution images and lower area thresholds enable the identification of smaller-sized blocks. Therefore, compared to manual statistics, which require fieldwork, the PCAS can identify blocks over a wider size range and avoids issues such as omissions, errors and duplicate records, offering advantages in terms of high precision, efficiency and safety.

However, there are certain limitations associated with the utilization of the PCAS for block recognition. This method relies on aerial images, resulting in two-dimensional statistical data for the blocks. In this study, the area was used to represent the relative size of the blocks; however, when the height of a block significantly exceeds its length and width, the PCAS results may underestimate its size. Field surveys can obtain three-dimensional data on blocks, which are crucial in analyzing the block characteristics and validating the image recognition results. To simplify the block shapes, in this study, field measurement data were initially used and revealed that most blocks in the Wangjiapo rock avalanche exhibited a nearly spherical shape. Consequently, even though two-dimensional area data can accurately represent the relative size of these blocks, the main sources of error in the PCAS method in this study were vegetation interference and block overlapping. The essence of PCAS recognition is in identifying differences in color between blocks and their surroundings. When there is little color difference between the blocks and the environment, the recognition effect is poor. In environments with significant vegetation interference, binary images often exhibit incomplete rock recognition, leading to voids or unclear boundaries. In this study, deposit II had significant vegetation, requiring manual processing to reduce the errors. Additionally, block overlap can distort the recognition of local block sizes and also necessitates manual processing interventions. When small blocks overlap with large ones, the recognized area of the large blocks may be underestimated. When multiple blocks overlap and the overlapping area exceeds the closure radius used in erosion operations, these blocks may be incorrectly identified as one. The Wangjiapo rock avalanche exhibits numerous instances of small block accumulation. Although manual processing was applied during image processing, only blocks with an area greater than 0.1 m² were extracted for analysis to improve the data accuracy. This study provides a more effective means for the investigation of the characteristics of rock avalanche deposits.

6. Conclusions

The movement of blocks during a rock avalanche is influenced by multiple factors and exhibits significant variability. By analyzing the characteristics of deposited blocks, it is possible to infer the movement mechanisms of the rock avalanche. The well-preserved deposited blocks from the Wangjiapo rock avalanche provide an excellent example for the performance of quantitative studies on such deposits. Based on the combination of field manual measurements, UAV photogrammetry and PCAS image processing, this study analyzed the characteristics of the deposited blocks. The main conclusions are as follows.

(1) Validation of Methodology: The PCAS recognition results showed a similar distribution pattern in terms of size and shape compared to the field-measured blocks, confirming that the statistical method combining UAV aerial imagery with the PCAS accurately reflects the basic characteristics of the Wangjiapo rock avalanche blocks. This method is convenient, efficient, precise and cost-effective.

(2) Fragmentation and Sorting: In deposit I, at the top of the slope, the blocks were large and irregularly shaped. With an increasing transport distance, relatively intact rock masses collide, fragment and disintegrate along cementation surfaces. In the lower deposit areas, the proportion of smaller blocks and the degree of rounding increase, along with improved sorting.

(3) Influence of Cementation and Topography: The characteristics of the blocks in deposits II and III were also different, being influenced by the degree of rock mass cementation, topography and transport distance. In deposit II, the poorly cemented breccia, coupled with the infill of fine-grained soil-like material and the relatively gentle terrain, limited block fragmentation. In contrast, the strongly cemented breccia and steeper channels resulted in more energetic collisions in deposit III, leading to more significant fragmentation and an increased number of blocks.

(4) Potential Energy and Inertia: Large blocks exhibit distinct movement characteristics driven by the potential energy. Longitudinally, the sorting effect caused by the potential energy difference between large and small blocks enables large blocks to travel further than smaller ones. Horizontally, large blocks have a significant inertia advantage, making it difficult to alter their movement direction, even when colliding with surrounding rocks. Consequently, large blocks tend to concentrate within the central regions of the deposit area.