Remote Sensing Identification and Information Extraction Method of Glacial Debris Flow Based on Texture Variation Characteristics

Abstract

The formation mechanism of glacial debris flows in alpine gorge mountain areas is complex, with varying characteristics across different regions. Due to the influence of mountain shadows and the accumulation and ablation of ice and snow, accurately identifying and rapidly extracting glacial debris flows using optical images remains challenging. This study utilizes the Random Forest method to develop a multi-feature spatiotemporal information extraction model based on Landsat-8 images and a glacial debris flow gully identification model. These models were applied to the Songzong–Tongmai section of the Sichuan–Tibet Highway to identify glacial debris flows. The results showed that (1) the multi-feature spatiotemporal extraction model effectively eliminated the interference of mountain shadows and ice–snow phase changes, resulting in a higher accuracy for identifying and extracting glacial debris flows in areas with significant information loss due to deep shadows. The total accuracy was 93.6%, which was 8.9% and 4.2% higher than that of the Neural Network and Support Vector Machine methods, respectively. (2) The accuracy of the glacial debris flow gully identification model achieved 92.6%. The proposed method can accurately and rapidly identify glacial debris flows in alpine gorge mountain areas, facilitating remote sensing dynamic monitoring. This approach reduces the damage caused by debris flows to both transportation and the environment, ensuring the safe passage of highways and promoting the sustainable development of the region.

1. Introduction

In high alpine glacier environments, debris flows and ice avalanche deposits are the primary sources of solid material replenishment. Glacial debris flows are the result of the melting of glaciers and ice sheets. They can occur in a number of ways, including the release of meltwater, the collapse of ice shelves, and the breakup of ice lakes [1,2,3]. Glacial debris flows are characterized by a high suddenness, high intensity, short duration, large changes in volume, and high capacity for siltation [4,5,6,7]. Glacial debris flows in China are found in the southeast highlands near Tibet. There are many flows along the Sichuan–Tibet Highway, and the weather is harsh. People and property downstream of glacier basins are at risk from debris flows [8,9,10,11]. Debris flows not only present significant threats to transportation but also exert considerable impacts on the ecological environment. Consequently, monitoring debris flows not only mitigates the direct economic losses associated with such disasters but also safeguards the ecological environment and fosters sustainable development.

Numerous studies have examined how debris flows develop [12,13,14], feature classification [15,16,17,18,19,20], the mechanism of occurrence [21,22,23,24,25], monitoring and early warning [26,27,28,29], and risk assessment [30,31,32,33,34,35]; all of these conducted extensive studies and produced significant findings. The visual interpretation of image morphological features from the early stages to the semi-automatic quantitative interpretation of spectral, texture, and geometric features is the foundation of study on the extraction of remote sensing information from a debris flow. Studies on debris flows in a debris flow fan [36], channel, source [37], provenance, and possibly prone areas [38,39], have specifically improved the accuracy and effectiveness of judgment. A growing number of techniques are being used to automatically extract debris flow data, including multi-source remote sensing [40], multi-phase imaging [36], high-resolution imagery [41], and Ground-Based SAR (GBSAR) [42]. The extraction of debris flow data from remote sensing images has been a major research accomplishment in the area of disasters. There is little research on glacial debris flows. The formation process of glacial debris flows is distinct from that of a typical debris flow [43]. High-resolution images [44] and three-dimensional remote sensing image model technology [45] were used to extract information on single gully debris flows. These high-precision data cost much and have a limited time sequence, which is not good for long-term serial large-scale regional disaster studies. For large-scale areas, there are currently few full identifications and extractions of glacial debris flow formation, circulation, and accumulation areas. According to research analysis, Landsat series optical images are highly accurate, simple to acquire, and long-term [46], which can accommodate a wide range of glacial debris flow information.

An image is vulnerable to mountain shadows in mountainous and undulating landscapes, making it difficult to identify features and resulting in the misidentification and loss of target information [47,48]. Numerous correction models, including the early ideal, Cosine correction, and the subsequently improved cosine correction (C-correction) model, Cosine-C model, Cosine-b model, SCS model, SCS+C model, etc., have been developed both domestically and internationally to eliminate the influence of shadows on information extraction [29,49,50]. In order to image such terrain correction models based on DEM data, it is typically necessary to consider the solar elevation angle and sensor settings. A number of these variables are omitted due to their inherent complexity and the difficulty of obtaining them. The derivation of empirical models is subject to significant constraints [51]. The primary challenge in the recognition and extraction of glacial debris flow information using optical remote sensing is the elimination of shadows on the image and the recovery of image structure information in the shadowed area.

The formation area of glacial debris flows typically includes glaciers and snow in addition to the shadows. However, the seasonal variability of snow and ice coverage, particularly between winter and summer, makes it difficult to precisely determine the extent of the formation area of glacial debris flows [52,53,54]. The remote sensing spectral characteristics of a glacial debris flow, which is composed mainly of glacial debris, soil, rock, and water, are highly similar to the characteristics of a non-glacial debris flow. Furthermore, different types of debris flow channels also exhibit similarities in terms of texture, morphology, and spectral characteristics. Consequently, accurately determining the glacial debris flow channels is another key issue [55,56].

To address the issues of mountain shadow effects, changes in glacier snow accumulation phases, and the confusion between glacial debris flow gullies and other types of debris flow gullies, in this paper we conducted the following study based on three phases of Landsat-8 imagery for the study area: (1) We introduced a spatiotemporal recognition degree to fully utilize the temporal and spatial information from imagery to extract glacial debris flow information. A multi-feature spatiotemporal identification and information extraction model for glacial debris flows is proposed to reduce the interference from mountain shadow effects and snow accumulation phases, allowing for the accurate identification and extraction of the boundary of glacial debris flows. (2) By analyzing the texture characteristics of different types of debris flow gullies, we clarified the texture variation patterns of glacial debris flows and non-glacial debris flows, as well as the typical features within these gullies. A glacial debris flow gully identification model was constructed to achieve the identification and extraction of glacial debris flow gullies. Additionally, spatial overlay analysis was used to extract comprehensive information from the entire basin of glacial debris flows, thereby improving the accuracy of glacial debris flow information extraction.

2. Study Area and Data

2.1. Study Area

The Songzong–Tongmai section of the Sichuan–Tibet Highway (Figure 1) is situated at 30°13′15″ N, 29°36′7″ N, 95°00′5″ E, and 96°7′30″ E in the southeast of Tibet. Due to the area’s steep valleys and sharp cuts, ice, and snow can confluence and move more easily. Fluvial, glacial, mountainous, and glacial landforms all evolve over time. The abundance of different kinds of loose hailstones and valleys left behind by the last glaciers during the Quaternary era make up a rich source of solids for mudslide activity [57]. Residual slope deposits and collapse-accumulated deposits formed at the valley slope. The Zhamu–Maniun strike-slip fault, the Yigong Zangbo–Parlung Zangbo deep fault zone, and the Lhasa–Bomi fold belt are all nearby, which cause significant neotectonic movements with Devonian and Carboniferous rocks. The primary sources of debris movement are sedimentary rocks, metamorphic rocks, and Quaternary deposits [58]. The study area is primarily impacted by the monsoon from the southwest Indian Ocean, which brings heavy rains as well as the growth of both ancient and contemporary marine glaciers. The unique glacial debris flow observed in the study area was the result of a complex interaction of geological and geomorphological processes, climate factors, and glaciers [59].

2.2. Data

2.2.1. Landsat-8 OLI Data

The multi-spectral, long-temporal, and continuously observed medium-to-high-resolution remote sensing images provided by Landsat-8 OLI (Operational Land Imager) contain abundant band information. With its high radiometric resolution and signal-to-noise ratio, it is suitable for extracting information related to large-scale glacial debris flows [60].

Three phases of Landsat-8 imagery at 30-m resolution were selected: the initial phase of the debris flow (March to September), the occurrence phase (June to September), and the later occurrence phase (October to March). The acquisition dates were 20 April 2015, 25 July 2015, and 14 November 2015, respectively. Table 1 contains more details about the images. A total of 7 bands of aquamarine, red, green, blue, near-red, and two short-wave infrared bands from those data were selected. The data were primarily downloaded from the Geospatial Data Cloud website: “www.gscloud.cn (accessed on 7 April 2023)”, and the projection coordinate system is UTM-WGS84 Zone 48N. An initial multiband image was created and subsequently subjected to a series of corrections and enhancements. These included geometric correction, radiation calibration, atmospheric correction, and additional operations. The image was then converted to true surface reflectivity, cropped, enhanced, and subjected to further operations to obtain the initial image of the study area.

2.2.2. Extra Data

• (1). field survey data

The status of disasters, such as mudslides, slippery slopes, collapses, and landslides on both sides of the highway, as well as their precursor disasters, was investigated in June 2015 and June 2016. A total of 194 surveys was conducted, including one that examined the condition of debris flow ditches along the gully. Of these, 124 sites were affected by debris flows of a glacial nature. The present study focuses on the Songzong–Tongmai section of the Sichuan–Tibet Highway. The stretch comprises 58 glacial debris flows and 90 common waterways.

• (2). Watershed boundary data

A detailed analysis of the 16 m Digital Elevation Model (DEM) data was conducted to gain insight into the study area’s hydrology. The minimum flow threshold of the secondary tributary on both sides of the Parlung Zangbo River was calculated using the four-neighborhood flow rate analysis algorithm. The vector boundaries of the watershed units along the road in the study area were accurately extracted through calculations such as catchment analysis, depression filling, and watershed identification.

3. Research Methods

The deep-cut valleys within the gorge are areas prone to glacial debris flows. Both debris flows and glaciers are present on either side of the circulation area. Shadows significantly hinder the extraction of glacial debris flows from optical images [61]. These flows typically manifest in photographs from July to November, coinciding with the development of glaciers and snow-covered mudslides. The formation area of glacial debris flows undergoes regular changes due to seasonal fluctuations in coverage, reaching its peak from June to August and its trough from December to March. This area is often covered in snow and ice, and because the quantities of ice and snow vary over time, determining the origins of these debris flows proves challenging. Identifying the various types of debris flow gullies is critical for extracting reliable glacial debris flow information, as they often share similar image characteristics.

In this paper, the temporal and spatial discrimination of glacial debris flow was used to extract and study the texture change characteristics of glacial debris flow. The corresponding rules and models were then constructed to extract the remote sensing information of glacial debris flow. The issue of false extraction and missed extraction of glacial debris flow was resolved due to shadow, ice and snow interference, gully type confusion, and other factors. The overall research technique methods are shown in Figure 2.

3.1. Extraction of Glacial Debris Flow Range

A multi-feature spatiotemporal recognition and extraction model was proposed to fully leverage the temporal and spatial dimensional information of images while minimizing interference from shadows, ice, and snow.

3.1.1. Extraction and Refining of Features

Extraction of Features

Water bodies, rock/soil, and glacial till constitute the main body of glacier debris movement. The reflectivity of water is relatively small and inversely proportional to the wavelength. Glacial till is a kind of glacier sediment with spectral characteristics resembling those of glaciers. In the first four bands, the reflectivity is high, and the brightness value of the glacial debris flow on the image should be between moraine and water. At 0.75 m (about B5), it is almost completely absorbed by the water body (Figure 3).

The spectral value of glacial debris flow in Figure 3 illustrates the upward trend from B5 to B7, which clearly separates other landmarks. In order to improve its spectral characteristics, this paper constructed the Glacial Debris Flow Index (GDFI) using B5 and B7:

GDFI = ((B5 − B7))/((B5 + B7)),(1)

By enhancing the difference between the near-infrared and red light bands, the normalized vegetation index was extracted. Principal component analysis (PCA) was utilized to eliminate redundant information between bands. Subsequently, the Tasseled Cap transformation was conducted to extract the moisture index, greenness index, and soil brightness index. The application of independent component analysis (ICA) enabled the transformation of mixed signals into mutually independent components, facilitating the differentiation of features characterized by significant interference, such as accumulation areas and river floodplains. The spectral information of exposed rock and soil can be enhanced by applying minimum noise fraction (MNF) for image denoising. The Gray-Level co-occurrence matrix was used to extract image texture features.

Refining of Features

The purpose of feature optimization is to improve the relationship between features and eigenvalues while lowering the dimensionality. Therefore, remote sensing classification and recognition of many features are based on the principle of optimizing key features, but it is highly susceptible to error [62]. The method of Random Forest (RF) is often used as a tool for feature selection to assess the importance of features while training samples [63,64].

Correlation analysis and RF were combined in this paper to optimize features. Initially, a band correlation analysis was conducted on the feature images to eliminate highly correlated features and derive a set of preliminary features. Subsequently, the importance of these preliminary features was assessed using the Random Forest (RF) algorithm. The contribution of each preliminary feature across the trees in the Random Forest (RF) classifier was analyzed, and the features that significantly influenced the results were selected based on their average contribution values.

3.1.2. Multi-Feature Spatiotemporal Extraction Model

Building on the optimized features of multi-temporal images, a multi-feature spatiotemporal identification and extraction model for glacial debris flow was developed. The spatiotemporal identification degree of glacial debris flow was integrated with the Random Forest (RF) algorithm and the density segmentation algorithm.

Spatial Recognition

The percentage of pixels from glacial debris flows to all other pixels in the window at 3 × 3 intra-window single-phase image classification findings is known as the spatial recognition degree.

(2)$E_{nm}^i={\mathrm{r}}_{nm}^i/T,(i=1,2,3)$

where i is the i-th remote sensing image; n is the pixel of the n-th row of the remote sensing image; m is the m-th pixel on the remote sensing image; $E_{nm}^i$ is the spatial recognition degree of the pixel (n, m) in the i-th image; $r_{nm}^i$ is the number of pixels belonging to the glacial debris flow in the window when the pixel (n, m) on the i-th image is the center pixel; and $T$ is the total number of pixels in the window at 3 × 3 intra-window single-phase image.

Spatiotemporal Recognition Degree

The shadowed multi-phase images are averaged first, followed by the non-shaded images, to determine the mean value of the pixel space recognition degree at the same spatial position on the time series images. Figure 4 displays the schematic layout.

(3)$E_{nm}={\displaystyle \frac{{\displaystyle \sum _{t=1}^{N-J}{E}_{nm}^t}+\frac{{\displaystyle \sum _{s=1}^J{E}_{nm}^s}}{J}}{(N-J+1)}}$

In the formula, $E_{nm}$ is spatiotemporal recognition degree; s is the s-th shadowed remote sensing image; t is the t-th non-shadowed remote sensing image; N is the total number of images in which the pixel (n, m) participates in the calculation; and J is the number of shadowed images in which the pixel (n, m) participates in the calculation.

Information Extraction

The glacial debris flow information in images, especially the information under shadow and ice–ice interference, was realized by using the spatiotemporal recognition model. Using the density segmentation method, the threshold range of the glacial debris flow extraction was established. The density slice’s threshold values were chosen based on expert classification. Through the repeated verification of the sample pixels, the information in the glacier debris motion image was extracted to determine the ideal density slice cutoff value.

3.2. Gully Type Determination Method

In the study area, there are many different types of debris flow. The characteristics of glacial debris flow gullies regarding land cover types, spatial distribution, and structural zoning closely resemble those of non-glacial debris flow gullies. As a result, there is a higher likelihood of misidentification or overlooking them when extracting debris flow gullies from satellite imagery. It is necessary to establish a classification model capable of distinguishing their specific types. Glacial debris flow is distinguished from other types of debris flow by its unique formation mechanism and evolutionary process. In regions susceptible to glacial debris flow, notable variations in glacier and moraine structures are particularly evident before and after debris flow events. To minimize errors in identifying the type of debris flow, it is crucial to extract image texture change features, develop a model for categorizing glacial debris flow gullies, and accurately determine the relevant information pertaining to these gullies. These steps collectively contribute to enhancing the precision of debris flow classification.

3.2.1. Texture Change Features Extraction

The texture variation features extracted from two distinct time-phase images within the same band can quantitatively express the local time-phase correlation of multi-temporal images [65]. Figure 5 shows the deep blue band texture variance features (Figure 4d and Figure 5c) and texture change features (Figure 5e) of two single-phase images. In the texture change map, the transition from dark to light in red represents a positive texture change from strong to weak, while the transition from dark to light in blue denotes a negative texture change from strong to weak. The glacial debris flow formation zone glacial till (in the yellow box in the figure), the circulation zone gully, and the supply source (in the white box in the figure) are highlighted in Figure 5e compared to Figure 5c,d. In the glacial debris flow, other landmarks (red circles in the figure) in the vegetation, such as texture changes, can also be seen. The majority of the single-phase texture maps are shown in purple. Therefore, it can be concluded that the texture changes image of two consecutive phases highlights the key characteristics of the glacial debris flows more effectively than that of any single-phase texture features image.

3.2.2. Glacial Debris Flow Gully Determination Model

To avoid incorrectly determining the gully type, we analyzed the landmark texture changes within the glacial debris flow features using the results of the image landmark classification, texture change features images, and basin unit.

Preliminary Determination

A glacial debris flow basin and other types of debris flow basins in the study area were selected as samples to establish a database of debris flow texture change characteristics, and the average texture change values of different partitions and different types of characteristics in the basin were calculated and used as independent variables of glacial debris movement. The dependent variable was the characteristic value of texture change, including glaciers, bare land, vegetation, moraine, and the area where glacial debris flow develops, flows, and accumulates. Then, using the binary logistic regression analysis algorithm, the classification model of various types of glacial debris flow gullies in the study area was established. The equation is as follows:

(4)$\begin{array}{l}logit(P)=9.414 \times glacier+2.857 \times formation area - 7.387 \times circulation area+8.288 \times accumulation area \\ +2.787 \times glacial till - 0.793 \times bare land+11.530 \times vegetation\end{array}$

where the dependent variable P shall only be 0 or 1; when the river basin unit meets the conditions for glacial debris flow after calculation, P = 1 shall be returned; otherwise, P = 0. According to its relationship with the dependent variable and its contribution to the model, several independent variables were selected, such as glacier, formation area, circulation area, accumulation area, glacial till, bare land, vegetation. The value of the independent variable was the mean value of the texture changes of a type of feature in the basin. The regression coefficient before multiplication was estimated by maximizing the likelihood function. An optimization algorithm, such as Newton–Raphson method or quasi-Newton method, was used to find a parameter value that maximized the likelihood function. The likelihood ratio test, Akaike Information Criterion (AIC), or Bayesian Information Criterion (BIC) were used to verify the goodness-of-fit of the model, and finally the optimal regression coefficient values could be obtained. A positive value of the regression coefficient indicates a positive correlation, and a negative value of the regression coefficient indicates a negative correlation. The absolute value of the regression coefficient represents the importance of the independent variable to the dependent variable. The independent variable vegetation was the most important influencing factor.

Secondary Determination

The glacial texture feature value in the glacial debris flow basin was almost zero in the texture change feature. The reason for this is that there were some glacial debris flows at higher altitudes in the study area. The glaciers formed in the basin did not change before or after the occurrence of the debris flow, and the glacier source was still generated and stored by the previous geological activities. In this case, a model that identifies the type of glacial debris flow may be wrong. In order to assist in the identification of glacial debris flow gullies, this paper examined the texture changes of various types of debris flow basins (Figure 6), quantitatively summarized the changes (Table 2), and formulated the rules of glacial debris flow texture changes.

The characteristics of glacier, formation area, circulation area, accumulation area, glacial till, bare land, and vegetation differed significantly from those of other types of debris flow, as shown in Figure 6. Glacial till is the primary source of glacial debris flow. Before and after the occurrence of debris flow, there must be texture changes, such that the texture features of the glacial till in the basin of the glacial debris flow are not zero.

Table 2 shows that while other debris flows lack glacial features, glacial debris flows do contain a few glaciers because of their high altitude and year-round snow cover, which causes them to undergo minor changes quickly and lack glacial change characteristics. According to Figure 6 and Table 2, this study establishes the rule of the texture change of the glacial debris flow primarily based on the change of the texture of the glacial till, the formation area, and the glacier (see Figure 7). This makes it possible to accurately identify and extract a glacial debris flow gully in this case.

The basin with a glacier texture change of 0 adopts the rule of Figure 7 to perform the secondary determination of a glacial debris flow gully, while basins with a glacier texture change of more than 0 are based on the results of the glacial debris flow-type identification model.

3.3. Extraction Method for Glacial Debris Flow Information

The glacial debris flow information extraction and gully type determination under the influence of shadows, ice, and snow are the main focal points of the remote sensing extraction method (Figure 8). The multi-characteristic spatiotemporal extraction model of glacial debris flows and a gully type determination model were established, and the extraction of glacial debris flows was realized using spatial superposition analysis.

4. Results

4.1. Glacial Debris Flow Information Extraction

4.1.1. Shadow and Snow Interference Treatment

Features Extraction and Features Analysis

• (1). Overall features

Figure 9b shows the features image of the Glacial Debris Flow Index (GDFI). Compared to the original image, the overall information of the glacial debris flow is highlighted in the gray-black (within the yellow boundary). This is particularly evident in the formation region where the glacial till is black. (in the red circle). In Figure 10c, the mean, variance, and homogeneity of the 7th band are displayed in RGB. The overall information on the glacial debris flow is exceptional and is displayed in purple overall (in the red box), with a clearly visible yellow boundary of the formation area.

• (2). Features of the formation area

The original image was processed using the MNF transform to generate an MNF image (Figure 10a). The formation area of glacial debris flow was bright white (inside the red circle), while the surrounding ground-based objects were gray and black. After panicle cap transformation (Figure 10d), the high water content of solid loose matter and moraine was reflected by the humidity index, while the vegetation damage characteristics was reflected by the greenness index.

• (3). Features of the circulation area

The vegetation damage in the debris flow gully is severe, and it has unique characteristics that set it apart from other areas in terms of normalized difference vegetation index characteristics. Glacial debris flows have a strong destructive impact on the surrounding vegetation.

• (4). Features of the accumulation area

The independent component analysis (ICA) assigned the image bands 1, 5, and 3 to red, green, and blue, respectively. (Figure 10b). The area affected by the glacial debris flow looks lavender (red circle), tinged with emerald green, and stands out from the surrounding vegetation’s pale green background and the floodplain’s croc-like hue (blue circle). The characteristics of the accumulation area are clear and simple to recognize.

Features Optimization

When the feature image was first created, it had 78 bands and distinguishing characteristics such as spectral, texture, and exponential features. Using correlation analysis, we obtained 20 main selected feature bands (Table 3), including GDFI, Normalized Difference Vegetation Index (NDVI), PCA, ICA, MNF, and texture features, by removing highly correlated (positive and negative correlation coefficients between 0.8 and 1) bands. The importance degree of main feature bands was then determined for each period of 20 April 2015, 25 July 2015, and 14 November 2015 using the Random Forest algorithm. To analyze the three-phase preferred feature images, we chose the best four glacial debris flow feature images of importance in each time phase. In total, nine glacial debris flows, including ICA3, ICA4, ICA5, ICA6, ICA7, PCA1, PCA2, GDFI, and MNFband5, were chosen (Table 4). We combined the best features from each picture to produce the image classification background information.

Extraction of Spatiotemporal Recognition Degree

To build the background dataset, feature extraction and optimization were employed, utilizing remote sensing data from three distinct stages: the initial (20 April 2015), intermediate (25 July 2015), and terminal (14 November 2015) phases of the glacial debris flow event. The Random Forest method was used to categorize landmarks. Information about glacial debris flows was originally identified. Glacial debris flows were divided into three types: glacial debris flow formation areas, glacial debris flow circulation areas, and glacial debris flow accumulation areas. This is because the characteristics of glacial debris flow formation areas, circulation areas, and accumulation areas are very different. Floodplain, water body, habitation, shadow, bare land, glacial till, glacier, cloud, vegetation, glacial debris flow formation area, glacial debris flow circulation area, and glacial debris flow accumulation area made up the general classification scheme. During the processing phase, groups were merged following the classification. To ensure that the classification accuracy of each image exceeded 75%, field survey data were utilized as a sample for both classification and subsequent accuracy evaluation.

The categorization results for the historic sites in Guxiang Gully are shown in Figure 11. The region where the glacial debris flow formed was the smallest in the image taken on 20 April 2015, and glaciers covered the majority of the glacial debris flow formation area on 25 July 2015 (the white in the red circle is the formation area of glacial debris flow and the yellow is glaciers). It demonstrates that the range in glacial debris flow formation area is more affected by periodic variations in ice and snow areas. The images from 25 July 2015 and 14 November 2015 make it obvious that there is a shadow (brown-red) on the right side of the glacial debris flow circulation area (in the blue box), making it hard to determine the extent of the glacial debris flow area in shadow. Therefore, shadow has a larger influence on the size of the circulation area for glacial debris flows.

Spatiotemporal Recognition Degree Extraction Result

Figure 12 shows the spatial recognition degree between each scene in the Guxiang gully basin from 20 April 2015 to 14 November 2015 and the glacial debris flow pixel on the right side of the three-phase image. The color in the images is from dark to bright, indicating that with the increase of the recognition value of glacial debris flow, the pixel is more likely to fall within the range of a glacial debris flow. The right side of the circulation area of the glacial debris flow (in the yellow circle) is dark gray or black in the image space recognition degree map for 14 November 2015 and 25 July 2015, suggesting that it is less likely to be part of the glacial debris flow. The area is bright white in the corresponding spatial recognition degree map, indicating that it is extremely likely to belong to the glacial debris flow, even though the original image from 20 April 2015 indicates that the area is part of the glacial debris flow area. Similarly, there is much less information about the glacial debris flow than the information from the previous two stages in the region where the image was formed on 20 April 2015. The information about the glacial debris movement on each time image was combined in this paper, and, from the multi-phase space recognition degree maps, spatiotemporal recognition degree maps were calculated (Figure 13). Figure 13 shows that the glacial debris flow information is light gray, which indicates that there is a substantially increased likelihood that it belongs to the glacial debris flow area beneath the shadow of the circulation area (in the yellow circle). Values of the map of glacier debris movement spatiotemporal recognition are in the interval (0,1). The findings of range identification are shown as a fuzzy set.

4.1.2. Non-Glacial Gully Elimination

In this paper, the texture feature images of the glacial debris flow occurrence period image (25 July 2015) and the glacial debris flow occurrence period image (14 November 2015) were selected, and the dynamic change of the texture phase was detected and analyzed using the difference method, and then the texture change characteristics of the glacial debris flow basin in the shortest time after the occurrence of the debris flow were obtained. This can be seen in Figure 14. Figure 15 displays the basin units along the line in the study region. The glacial debris flow gullies in the study region were identified by the glacial debris flow gully recognition model, which combined the aforementioned image landmark classification results.

4.1.3. Results of Extraction

Extraction of Glacial Debris Flow Gully Type Information

Using the glacial debris flow gully determination model, the type of glacial debris flow gully in the study region was determined and extracted. The results in Figure 16 demonstrate that glacial debris flows are represented by the blue watershed unit and other debris flows are represented by the purple watershed unit. Figure 16 shows that while other types of gullies are dispersed throughout the study area, the glaciated debris flow basin is the main area along the highway on both sides, making up more than 80% of the total basin units. The glaciated debris flow gully on the left side is larger in area and number than the one on the right side.

Glacial Debris Flow Activity Range Information Extraction

In this paper, the density of the space–time recognition map of glacial debris flow was classified (Figure 17). The activity range information of glacial debris flow was extracted. Using field survey data and a large number of experiments, it was found that the optimal extraction threshold of glacial debris flow is 0.25 (Figure 18). As shown in Figure 18, the activity range of glacial debris flow in the study area is mainly concentrated in the valley range of rivers on both sides of the Parlung Zangbo River, covering a wide range and accounting for about 40% of the total area of the study area; and there are many residential areas in the area, indicating that glacial debris flow has a greater impact on the local area.

Extraction Results of Glacial Debris Flow

By performing a spatial overlap analysis on the extracted information about the range of glacial debris flow activity and the type of glaciation debris flow in the research area, glacial debris flow information was acquired. As seen in Figure 19, a total of 62 glacial debris flows with a combined area of 649.95 km² were recovered.

4.2. Accuracy Evaluation

4.2.1. Spatiotemporal Recognition Degree Extraction Accuracy

A confusion matrix was used to evaluate the accuracy of the image classification results. The field survey data of image on 24 November 2013 was used as a validation set. The precision assessment sample was used to compute the three-stage image classification confusion matrix (error matrix). The overall classification accuracy and Kappa coefficient were obtained (Table 5).

The classification accuracy of single-phase images increased to more than 80%, which satisfies the criteria for the development of the ensuing glacial debris flow remote sensing model. The universality of the optimized glacial debris flow characteristic picture combo was also demonstrated.

4.2.2. Extraction Accuracy of Debris Flow Gully Type

The team made two field visits to the study area in 2015 and 2016 to conduct field research and field surveys there in order to identify the different types of debris flows. The field survey data were used to confirm the correctness of the extracted glacial debris flow gully categories. Table 6 presents the findings.

4.2.3. Extraction Accuracy of Glacial Debris Flow Range

In the study area (three red squares in Figure 20), three glacial debris flows were selected and combined with three-phase image information, a visual interpretation of coverage (Figure 21a and Figure 22a), a calculation of area, and a comparison with single-phase image information. Information is provided on glacial debris flows after Random Forest classification (Figure 21b–d, Figure 22b–d, and Figure 23b–d) and on glacial debris flows extracted using this method (Figure 21e, Figure 22e, and Figure 23e).

Table 7 demonstrates that the area of the glacial debris flow extracted from the image classification for 20 April 2015 had the largest difference when compared to the visual interpretation of the glacial debris flow using the single-phase image extraction method. The area of error was up to 9.607 km², and the glacial debris flow extracted from the image of July 2015 was available. The single-phase image extraction of glacial debris flows was compared with the glacial debris flow extracted using this method, and the results obtained using this method could eliminate the interference of shadows. The area was relatively close, the image was in summer, and the glacial debris flow formation area was the most complete, but there was a shadow interference in the circulation area, and there was still an error in the visual interpretation. The three glacial debris flow regions had an error of less than 0.524 km².

4.3. Analysis of Glacial Debris Flow Extraction in Shadow Coverage Area

Statistics were compiled on the areas covered and the type of ground cover in the shadowy areas along highways and roadways before, during, and after the glacial debris flow. Table 8 displays the extent to which the shadows cast by the peaks within the basin were covered by the single-phase image road.

Table 8 shows that the 14 November 2015 image had the most severe image shadow interference; the mountain shadow area in this image accounted for roughly 16.62% of the total area of the river basin along the road in the study area, while the mountain shadow area in the images from the other two periods accounted for less than 5%. Figure 24a,b show the shaded area vector diagram after selecting the typical sub-region in the 14 November 2015 image, analyzing the impact of extracting the glacial debris flow information from the shadow and extracting the shaded area in the sub-region. On the basis of the classification system in Section 3.1.1, the three categories of Shadowed glaciers, Shadowed glacial debris flows and Shadowed other features were added. This will further improve the classification effect of the neural network and vector machine methods under the shadow. The results extracted using the neural network and support vector machine methods were tailored using the results of shadow vectorization to obtain the glacial debris flow data pertinent to the shadow coverage area and were then contrasted with the method and the visual interpretation method to extract the results. Use was made of the shadow I and shadow II areas (Figure 25 and Figure 26). The blue areas show the ranges of glacial debris flows extracted by different methods; and the yellow areas show the ranges of glacial debris flows extracted by visual interpretation.

The information about the glacial debris flow that was extracted using support vector machines and neural networks in the shaded region I in Figure 25 was contrasted with the outcome of the visual interpretation. The information was mostly inaccurately extracted in the shaded region. The information about the glacial debris flow in the shadow that was extracted using this technique was essentially consistent with the range of visual interpretation in the patchy form. The extraction impact of this method is still the best from Figure 26, specifically in the shadow II area. There are shadows in the image, and the shadow in Area II is shallower than in Area I, so from the image’s point of view, the light red glacial debris flow area and the green vegetation, etc., can be discerned in the shadows of the second area, and the shadow in Area II is deeper than in Area I. At the same time, the support vector machine and the neural network method could also completely extract part of the glacial debris flow information in the shadow coverage area better in Zone I than in Zone II.

Thirty samples were randomly selected from each group to determine the classification accuracy (Table 9). The results show that in the shadow coverage region, the overall accuracy, user accuracy, and cartographic accuracy of the classification results of this method were 93.6%, 92.1%, and 93.2%, respectively. The overall accuracy of the glacial debris flow was 4.2% and 8.9% higher than that of the support vector machine and neural network, respectively. An explanation for this is that the shadows in the image, as well as the spectral heterogeneity, had an impact on the neural network and the support vector machine. This approach avoids the issue of inaccurate classification results brought on by image shadows in single-phase images by making full use of the temporal and spatial dimensional information of pixel categories to accurately forecast the distribution of glacial debris flows in shadow coverage areas.

5. Conclusions and Discussion

5.1. Discussion

When extracting information about glacial debris flow, the multi-featured spatiotemporal extraction model and glacier gully identification model proposed in this paper meets the necessary accuracy requirements and can be semi-automated. Interactive Data Language (IDL) programming can be used to further integrate technology, quickly and accurately extract glacial debris flow and other disaster information, and increase the effectiveness of disaster relief, reduction, and prevention efforts. Similar to shadow coverage images, cloud coverage images result in the loss of land cover information within the affected regions, making it difficult to extract ground feature information. The distinction is that the coverage and distribution of clouds on the image are random, while the distribution of shadows on the image has the regularity of the time phase. This technique can eliminate the impact of shadows on the extraction of ground feature information to a certain extent. The use of multi-phase images for the extraction of ground feature information under variable cloud cover shows potential possibilities for other applications. Formulas (2) and (3) in this paper were locally optimized to reduce cloud interference, though further experimental validation is required.

One of the key variables in disaster studies is the climate. Rainfall and temperature both have a significant impact on glacial debris flow in the research area. Investigating the coupling relationship between disasters and the climate provides guidance for scientific disaster prevention and reduction, along with climate variables like temperature and rainfall in the study area. A further exploration of its dynamic change law is critical for the climate change inversion and prediction based on remote sensing data extraction of glacial debris flows.

5.2. Conclusions

To achieve our research goal, a multi-characteristic spatiotemporal extraction model for glacial debris flows was established, texture changes of glacial debris flows were extracted, a model for determining the gully of a glacial debris flow was built, and shadow and ice–snow interference effects and other types of gully confusion information were removed. This model was based on the optimization and classification of glacial debris flow features. The following conclusions were drawn:

• (1). The Glacial Debris Flow Index (GDFI) features were created to extract the spectral, texture, spatial, and other distinctive features of glacial debris flows. Key features of the glacial debris flow image were optimized using correlation analysis and the Random Forest algorithm. Once the optimal features for extracting glacial debris flows were identified, they included MNF (Band5), ICA (band3, band4, band5, band6, band7), PCA (band1, band2), and GDFI, which validated the efficacy of single-phase images. When feature correlation was removed, the Random Forest algorithm was employed to optimize the image features, assigning a higher priority to the selected features. This method successfully identified glacial debris flows in shadow-covered areas and outperformed neural network and support vector machine methods in terms of overall accuracy, user accuracy, and cartographic precision. Under a thorough classification scheme, neural networks and support vector machines can extract some information about glacial debris flows. However, in deep shadows, extracting glacial debris flows is challenging. Our technique can extract glacial debris flows, even under very dark conditions, fully exploiting the information of images in the time dimension, removing image shadow interference, and identifying the activity range of glacial debris flows using optical imagery.

• (2). The texture change feature of glacial debris flow was extracted based on the texture characteristics of different divisions of glacial debris flow and its various types of features. This feature fully highlights the differences between glacial debris flow gullies and other types of debris flow gullies. A glacial debris flow gully determination model was established based on this feature, and the accuracy of extracting glacial debris flow gully reached 92.6%. The texture change feature can effectively eliminate non-glacial gullies, thus further improving the glacial debris flow extraction accuracy. These findings offer substantial technical support for disaster prevention and mitigation and regional sustainable development in Tibet.

Author Contributions

Conceptualization, J.F., Y.H. and T.L.; methodology, J.F., Y.H. and Z.Y.; formal analysis, J.F. and Z.Y.; data curation, L.L. and. D.C.; validation, J.F. and Y.H.; writing—original draft preparation, J.F. and Z.Q.; writing—review and editing, Y.H. and L.L.; visualization, D.C. and Z.Q.; project administration, T.L. and L.C.; funding acquisition, T.L. and L.C. All authors have read and agreed to the published version of the manuscript.

Data Availability Statement

The data presented in this study are available upon request from the corresponding authors.

Acknowledgments

We would like to thank the anonymous reviewers for their suggestions and comments.

Conflicts of Interest

The authors declare no conflicts of interest.

Footnotes

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Figures and Tables

Figure 1. The location of the research area and its topographic map.

Figure 2. The overall study technical methodology diagram.

Figure 3. Significant spectral structures Landsat-8 image characteristics.

Figure 4. A diagram depicting the spatiotemporal recognition degree of glacial debris flow.

Figure 5. Texture features with a single time-phase and texture change features. (a) 25 July 2015 image; (b) 14 November 2015 image; (c) 25 July 2015 texture; (d) 14 November 2015 texture; (e) texture changes image).

Figure 6. Kind of landmark texture change chart (in Figure 6, the abscissa axis values represent various types of features in Table 2): (a) glacier type; (b) other types.

Figure 7. Rule of glacial debris migration surface modification.

Figure 8. Method for glacial debris flow extraction.

Figure 9. Original photograph and glacial debris migration index features: (a) original image, (b) GDFI.

Figure 10. Graphical characteristics of the glacial debris flows: (a) MNF feature map band4; (b) ICA feature map (R: band1; G: band5; B: band3); (c) texture feature map (R: mean; G: variance; B: homogeneity); (d) K–T transformation.

Figure 10. Graphical characteristics of the glacial debris flows: (a) MNF feature map band4; (b) ICA feature map (R: band1; G: band5; B: band3); (c) texture feature map (R: mean; G: variance; B: homogeneity); (d) K–T transformation.

Figure 11. Results of image classification for Guxiang gully: (a) 20 April 2015; (b) 25 July 2015; (c) 14 November 2015.

Figure 12. Remote sensing image and spatial recognition degree map of Guxiang gully: (a) 20 April 2015; (b) 25 July 2015; (c) 14 November 2015.

Figure 13. Degree of spatiotemporal recognition of glacial debris flow in the investigated area.

Figure 14. Texture change characteristics from 25 July 2015 to 14 November 2015.

Figure 15. Basin unit vector boundary along the route.

Figure 16. Extraction result of glacial debris flow type information.

Figure 17. Glacial debris flow spatiotemporal recognition degree density slice result.

Figure 18. Extraction results of glacial debris flow activity range information.

Figure 19. Glacial debris flow extraction results.

Figure 20. Image overlay of glacial debris flow extraction results.

Figure 21. Extract of glacial debris flow 1: (a) visual interpretation range; (b–d) Phase III single-phase image extraction results in 2015; (e) results of the method extraction.

Figure 22. Extract of glacial debris flow 2: (a) visual interpretation range; (b–d) Phase III single-phase image extraction results in 2015; (e) results of the method extraction.

Figure 23. Extract of glacial debris flow 3: (a) visual interpretation range; (b–d) Phase III single-phase image extraction results in 2015; (e) results of the method extraction.

Figure 24. 14 November 2015 Image typical subarea and its shadow area vector: (a) subarea; (b) shadow area vector boundary.

Figure 25. Extraction of glacial debris flow information in the shadow I area: (a) support vector machine method; (b) neural network method; (c) method of this article; (d) visual interpretation.

Figure 26. Extraction of glacial debris flow information in the shadow II area: (a) support vector machine method; (b) neural network method; (c) method of this article; (d) visual interpretation.

Table 1

Information table for Landsat-8 images.

Sensor	Line Number	Acquisition Time	Image Recognition	Cloud Volume (%)
Landsat-8 OLI(Operational Land Imager)	135/39	20 April 2015	LC81350392015110LGN00	8.89
		25 July 2015	LC81350392015206LGN00	3.66
		14 November 2015	LC81350392015318LGN00	9.77

Table 2

Kinds of landmark texture changes in various glacier basins and other debris movements are governed by law.

	Glacier 1	Formation Area 2	Circulation Area 3	Accumulation Area 4	Glacial Till 5	Bare Land 6	Vegetation 7
glacier type	−43.8~−7.9	−3.0~4.53	−3.4~−1.2	−2.8~0.5	−8.1~2.0	−3.6~2.0	−2.4~−1.9
other type	−	−2.6~−0.7	−2.7~−1.9	−2.5~−1.3	−3.4~−2.1	−3.0~−1.1	−2.4~−1.9

Table 3

Primary characteristic band.

ID	1	2	3	4	5	6	7	8	9	10
band name	GDFI	NDVI	PCA1	PCA2	PCA3	ICA3	ICA4	ICA5	ICA6	ICA7
ID	11	12	13	14	15	16	17	18	19	20
band name	MNF5	Variance(Coastal aerosol)	Correlation(Coastal aerosol	Correlation(Blue)	Correlation(Green)	Correlation(NIR)	Variance(SWIR1)	Homogeneity(SWIR1)	Correlation(SWIR1)	Contrast(SWIR2)

Table 4

Characteristic image optimization result.

Image	20 April 2015	25 July 2015	14 November 2015
preferred characteristic band	ICA5, MNF5, PCA1, ICA7, ICA4, PCA1, PCA2, ICA6, PCA1, GDFI, ICA4, ICA3
comprehensive results	ICA5, MNF5, PCA1, ICA7, GDFI, ICA4, ICA3, PCA2, ICA6

Table 5

Three-stage classification precision.

	Image Classification in 20 April 2015	Image Classification in 25 July 2015	Image Classification in 14 November 2015
overall accuracy/%	83.52	81.47	85.29
kappa coefficient	0.818	0.895	0.841

Table 6

Validation of the veracity of glacial debris flow gully data extraction.

Gully Survey Information			Gully Determination Information
Total Number	Glacial Type	Other Types	Ture	False	Correct Rate(%)
81	57	24	75	6	92.59

Table 7

Statistics of glaciation debris flow area extraction using different methods.

No.	Visual Interpretation/km2	Single-Phase Image/km2						Method of This Article/km2	Deviation/km2
		20 April 2015	Deviation	25 July 2015	Deviation	14 November 2015	Deviation
1	18.401	8.794	−9.607	17.812	−0.589	11.365	−7.035	18.097	−0.303
2	3.960	0.719	−3.241	3.109	−0.851	2.096	−1.864	3.436	−0.524
3	3.137	1.004	−2.133	2.731	−0.406	2.442	−0.695	3.215	0.078

Table 8

Shadow coverage of single-range remote sensing images.

	20 April 2015	25 July 2015	14 November 2015	14 November 2015 (Sub-Region I + II)
Shadow area/km2	49.868	29.696	237.046	5.189
Ratio of shaded area/%	3.50	2.08	16.62	9.69
Shaded area belongs to	Glacier, Other, Glacial Debris Flow	Glacier, Other, Glacial Debris Flow	Glacier, Other, Glacial Debris Flow	Glacier, Other, Glacial Debris Flow

Table 9

Validation of three methods.

	Single-Phase Image		Method of This Article
	Neural Networks	Support Vector Machines
Overall accuracy/%	84.7	89.4	93.6
User accuracy/%	90.1	82.5	92.1
Charting accuracy/%	90.6	92.7	93.2

References

1. Wang, X.; Liu, S.Y.; Ding, Y.J.; Guo, W.; Jiang, Z.; Lin, J.; Han, Y. An approach for estimating the breach probabilities of moraine-dammed lakes in the Chinese Himalayas using remote-sensing data. Nat. Hazards Earth Syst. Sci.; 2012; 12, pp. 3109-3122. [DOI: https://dx.doi.org/10.5194/nhess-12-3109-2012]

2. Wang, X.; Yang, C.D.; Zhang, Y.L.; Chai, K.G.; Liu, S.Y.; Ding, Y.J.; Wei, J.F.; Zhang, Y.; Han, Y.S. Monitoring and simulation of hydrothermal conditions indicating the deteriorating stability of a perennially frozen moraine dam in the Himalayas. J. Glaciol.; 2018; 64, pp. 407-416. [DOI: https://dx.doi.org/10.1017/jog.2018.38]

3. Guo, W.Q.; Liu, S.Y.; Xu, L.; Wu, L.Z.; Shangguan, D.H.; Yao, X.J.; Wei, J.F.; Bao, W.J.; Yu, P.C.; Liu, Q. et al. The second Chinese glacier inventory: Data, methods and results. J. Glaciol.; 2015; 61, pp. 357-372. [DOI: https://dx.doi.org/10.3189/2015JoG14J209]

4. Liu, X.J.; Cong, L.; Li, X.Z.; Madsen, D.; Wang, Y.X.; Liu, Y.G.; Peng, J. Climate conditions on the Tibetan Plateau during the Last Glacial Maximum and implications for the survival of paleolithic foragers. Front. Earth Sci.; 2020; 8, 606051. [DOI: https://dx.doi.org/10.3389/feart.2020.606051]

5. Yang, R.T.; Hock, R.; Kang, S.C.; Guo, W.Q.; Shangguan, D.H.; Jiang, Z.L.; Zhang, Q.B. Glacier surface speed variations on the Kenai Peninsula, Alaska, 2014–2019. J. Geophys. Res. Earth Surf.; 2022; 127, 3. [DOI: https://dx.doi.org/10.1029/2022JF006599]

6. Takahashi, T. Debris Flow: Mechanics, Prediction and Countermeasures; 2nd ed. CRC Press: Boca Raton, FL, USA, 2014.

7. Davies, T.; Shroder, J.F.; Rosser, N. Landslide hazards, risks and disasters: Introduction. Landslide Hazards, Risks, and Disasters; 2nd ed. Hazards and Disasters Series; Elsevier: Amsterdam, The Netherlands, 2022; pp. 1-10.

8. Korup, O.; Clague, J.J. Natural hazards, extreme events, and mountain topography. Quat. Sci. Rev.; 2009; 28, pp. 977-990. [DOI: https://dx.doi.org/10.1016/j.quascirev.2009.02.021]

9. Cui, P.; Su, F.H.; Zou, S.; Chen, N.S.; Zhang, Y. Risk assessment and disaster reduction strategies for mountainous and meteorological hazards in Tibetan Plateau. Chin. Sci. Bull.; 2015; 60, 3067.

10. Wei, D.; Yang, W. Detecting damaged buildings using a texture feature contribution index from post-earthquake remote sensing images. Remote Sens. Lett.; 2020; 11, pp. 127-136. [DOI: https://dx.doi.org/10.1080/2150704X.2019.1692389]

11. Teng, F.; Wang, Y.J.; Wang, M.J.; Wang, L.Q. Monitoring and Analysis of Population Distribution in China from 2000 to 2020 Based on Remote Sensing Data. Remote Sens.; 2022; 14, 6019. [DOI: https://dx.doi.org/10.3390/rs14236019]

12. Wei, R.Q.; Zeng, Q.L.; Davies, T.; Yuan, G.X.; Wang, K.Y.; Xue, X.Y.; Yin, Q.F. Geohazard cascade and mechanism of large debris flows in Tianmo gully, SE Tibetan Plateau and implications to hazard monitoring. Eng. Geol.; 2018; 233, pp. 172-182. [DOI: https://dx.doi.org/10.1016/j.enggeo.2017.12.013]

13. Kim, Y.; Nakagawa, H.; Kawaike, K.; Zhang, H. Study on hydraulic characteristics of sabodam with a flap structure for debris flow. Int. J. Sediment Res.; 2017; 3.

14. Han, Y.; Wang, Z.L.; Chang, Y.L.; Zhang, D.S.; Li, L.L.; Qiu, Z.T.; Xia, Y.D.L. Post-Shock Gravitational Erosion and Sediment Yield: A Case Study of Landscape Transformation along the Wenchuan–Yingxiu Section of the Minjiang River, Sichuan, China. Remote Sens.; 2023; 15, 3506. [DOI: https://dx.doi.org/10.3390/rs15143506]

15. Hurlimann, M.; McArdell, B.W.; Rickli, C. Field and laboratory analysis of the runout characteristics of hillslope debris flows in Switzerland. Geomorphology; 2015; 232, pp. 20-32. [DOI: https://dx.doi.org/10.1016/j.geomorph.2014.11.030]

16. Martin, Y.E.; Johnson, E.A.; Chaikina, O. Gully recharge rates and debris flows: A combined numerical modeling and field-based investigation, Haida Gwaii, British Columbia. Geomorphology; 2017; 278, [DOI: https://dx.doi.org/10.1016/j.geomorph.2016.11.012]

17. Wang, S.H.; Chen, Q.; Wang, Y.J.; Chu, N.; Li, C.K. Stand density extraction and analysis of plantations based on QuickBird and Worldview-2 images. J. Appl. Remote Sens.; 2020; 14, 022214. [DOI: https://dx.doi.org/10.1117/1.JRS.14.022214]

18. Jiang, S.; Jiang, W.S.; Guo, B.X.; Li, L.L.; Wang, L.Z. Learned local features for structure from motion of uav images: A comparative evaluation. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens.; 2021; 14, pp. 10583-10597. [DOI: https://dx.doi.org/10.1109/JSTARS.2021.3119990]

19. Chen, L.; Yang, Z.H.; Wang, Z.H.; Zhang, J.; Zhou, S.W.; Wu, L.H. Lightweight Underwater Target Detection Algorithm Based on Dynamic Sampling Transformer and Knowledge-Distillation Optimization. J. Mar. Sci. Eng.; 2023; 11, 426. [DOI: https://dx.doi.org/10.3390/jmse11020426]

20. Hu, J.Y.; Wu, W.H.; Motagh, M.; Qin, F.; Wang, J.Y.; Pan, S.Y.; Guo, J.M.; Zhang, C.Y. FIM-based DSInSAR method for mapping and monitoring of reservoir bank landslides: An application along the Lancang River in China. Landslides; 2023; 20, pp. 2479-2495. [DOI: https://dx.doi.org/10.1007/s10346-023-02097-5]

21. Riley, K.L.; Bendick, R.; Hyde, K.D.; Gabet, E.J. Frequency–magnitude distribution of debris flows compiled from global data, and comparison with post-fire debris flows in the western U.S. Geomorphology; 2013; 191, pp. 118-128. [DOI: https://dx.doi.org/10.1016/j.geomorph.2013.03.008]

22. Marra, F.; Nikolopoulos, E.I.; Creutin, J.D.; Borga, M. Space–time organization of debris flows-triggering rainfall and its effect on the identification of the rainfall threshold relationship. J. Hydrol.; 2015; 541, pp. 246-255. [DOI: https://dx.doi.org/10.1016/j.jhydrol.2015.10.010]

23. Chen, X.Q.; Cui, P.; Zhao, W.Y. Optimal Timing for the Control of Debris Flow in Wenchuan Earthquake Area. J. Sichuan Univ. (Eng. Sci. Ed.); 2009; 41, pp. 125-130.

24. Du, S.H.; Li, D.Y.; Zhang, C.S.; Mao, D.W.; Ruan, B. Deformation and strength properties of completely decomposed granite in a fault zone. Geomech. Geophys. Geo-Energy Geo-Resour.; 2021; 7, 13.

25. Chen, B.; Zhang, J.L.; Zhao, Y.L.; Lu, Y.W. Experimental Study on Mechanism of Graphene Oxide-Modified Coral Sand Cement Mortar to Resist Sulfate Erosion. Geofluids; 2022; 2022, 1905439.

26. Uchida, T.; Sakurai, W.; Iuchi, T.; Izumiyama, H.; Borgatti, L.; Marcato, G.; Pasuto, A. Effects of episodic sediment supply on bedload transport rate in mountain rivers. Detecting debris flow activity using continuous monitoring. Geomorphology; 2018; 306, pp. 198-203.

27. Marchi, L.; Arattano, M.; Deganutti, A.M. Ten years of debris-flow monitoring in the Moscardo Torrent (Italian Alps). Geomorphology; 2002; 46, pp. 1-17.

28. Navratil, O.; Liébault, F.; Bellot, H.; Travaglini, E.; Theule, J.; Chambon, G.; Laigle, D. High-frequency monitoring of debris-flow propagation along the Réal Torrent, Southern French Prealps. Geomorphology; 2013; 201, pp. 157-171.

29. Chen, Z.C.; Zhang, B.; Han, Y.S.; Zuo, Z.L.; Zhang, X.Y. Modeling accumulated volume of landslides using remote sensing and DTM data. Remote Sens.; 2014; 6, pp. 1514-1537. [DOI: https://dx.doi.org/10.3390/rs6021514]

30. Fan, H.D.; Lu, Y.F.; Shao, S.W.; Li, L.; Wang, Y.J.; Lu, M.; Li, J.; Yao, K.Z.; Sun, Y. Evaluation and analysis of statistical and coupling models for highway landslide susceptibility. Geomat. Nat. Hazards Risk; 2023; 14, 2167612.

31. Chen, J.; Peng, S.; Chen, H.; Zhao, X.S.; Ge, Y.J.; Li, Z.L. A comprehensive measurement of progress toward local SDGs with geospatial information: Methodology and lessons learned. ISPRS Int. J. Geo-Inf.; 2020; 9, 522. [DOI: https://dx.doi.org/10.3390/ijgi9090522]

32. Cama, M.; Lombardo, L.; Conoscenti, C.; Rotigliano, E. Improving transferability strategies for debris flow susceptibility assessment: Application to the Saponara and Itala catchments (Messina, Italy). Geomorphology; 2017; 288, pp. 52-65.

33. Han, Y.S.; Dong, S.K.; Chen, Z.C.; Hu, K.H.; Su, F.H.; Huang, P. Assessment of Secondary Mountain Hazards along a Section of the Dujiangyan-Wenchuan Highway. J. Mt. Sci.; 2014; 11, pp. 51-65. [DOI: https://dx.doi.org/10.1007/s11629-012-2516-1]

34. Aronica, G.T.; Biondi, G.; Brigandì, G.; Cascone, E.; Lanza, S.; Randazzo, G. Assessment and mapping of debris-flow risk in a small catchment in eastern Sicily through integrated numerical simulations and GIS. Phys. Chem. Earth Parts A/B/C; 2012; 49, pp. 52-63. [DOI: https://dx.doi.org/10.1016/j.pce.2012.04.002]

35. Han, Y.S.; Liu, H.J.; Cui, P.; Su, F.H.; Du, D.S. Hazard assessment on secondary mountain-hazards triggered by the Wenchuan earthquake. J. Appl. Remote Sens.; 2009; 3, 1.

36. Bernhardt, H.; Reiss, D.; Hiesinger, H.; Hauber, E.; Johnsson, A. Debris flow recurrence periods and multi-temporal observations of colluvial fan evolution in Central Spitsbergen (Svalbard). Geomorphology; 2017; 296, pp. 132-141. [DOI: https://dx.doi.org/10.1016/j.geomorph.2017.08.049]

37. Crowley, J.K.; Hubbard, B.E.; Mars, J.C. Analysis of potential debris flow source areas on Mount Shasta, California, by using airborne and satellite remote sensing data. Remote Sens. Environ.; 2003; 87, pp. 345-358. [DOI: https://dx.doi.org/10.1016/j.rse.2003.08.003]

38. Lin, C.W.; Shieh, C.L.; Yuan, B.D.; Shieh, Y.C.; Liu, S.H.; Lee, S.Y. Impact of Chi-Chi earthquake on the occurrence of landslides and debris flows: Example from the Chenyulan River watershed, Nantou, Taiwan. Eng. Geol.; 2004; 71, pp. 49-61. [DOI: https://dx.doi.org/10.1016/S0013-7952(03)00125-X]

39. Zhou, Z.; Zhu, J.; Liu, Y. Debris Flow Provenance Information Extraction and Analysis of Typical Landslides in Tibetan Southeastern Based on RS and GIS. Water Resour. Power; 2015; 33, pp. 127-131.

40. Mwaniki, M.W.; Kuria, D.N.; Boitt, M.K.; Ngigi, T.G. Image enhancements of Landsat 8 (OLI) and SAR data for preliminary landslide identification and mapping applied to the central region of Kenya. Geomorphology; 2017; 282, pp. 162-175. [DOI: https://dx.doi.org/10.1016/j.geomorph.2017.01.015]

41. Hervás, J.; Barredo, J.I.; Rosin, P.L.; Pasuto, A.; Mantovani, F.; Silvano, S. Monitoring landslides from optical remotely sensed imagery: The case history of Tessina landslide, Italy. Geomorphology; 2003; 54, pp. 63-75. [DOI: https://dx.doi.org/10.1016/S0169-555X(03)00056-4]

42. Long, S.C.; Tong, A.X.; Yuan, Y.; Li, Z.H.; Wu, W.H.; Zhu, C.G. New Approaches to Processing Ground-Based SAR (GBSAR) Data for Deformation Monitoring. Remote Sens.; 2018; 10, 1936. [DOI: https://dx.doi.org/10.3390/rs10121936]

43. Xu, D.; Feng, Q. Studies on Catastrophes of Glacial Debris flow and Glacial Lake Outburst Flood in China. J. Glaciol. Geocryol.; 1988; 3, pp. 284-289.

44. Wang, L. Study on Formation Conditions and Activity Features of Glacial Debris in the Jiacha to Milin Section along Yarlung Zangbo River. Master’s Thesis; Chengdu University of Technology: Chengdu, China, 2011.

45. Feng, Y.; Chen, J.; Jiang, Q.G.; Yang, L.J. THE Application of 3D Remote Sensing Image Model of Debris Flow Disaster Warming in Guxiang Gully, Tibet. J. Geol. Hazards Environ. Preserv.; 2009; 20, pp. 114-119.

46. Chi, Y.F.; Lia, R.W.; Yan, Q.; Yu, L.L.; Su, Y.Q. Detection and Extraction of Mountain Shadow Information from LandSat8 OLI data. Mounta Res.; 2017; 35, pp. 580-589.

47. Tan, B.; Masek, J.G.; Wolfe, R.; Gao, F.; Huang, C.Q.; Vermote, E.F.; Sexton, J.O.; Ederer, G. Improved Forest Change Detection with Terrain Illumination Corrected Landsat Images. Remote Sens. Environ.; 2013; 136, pp. 469-483. [DOI: https://dx.doi.org/10.1016/j.rse.2013.05.013]

48. Vanonckelen, S.; Lhermitte, S.; Rompaey, A.V. The Effect of Atmospheric and Topographic Correction on Pixel-based Image Composites; Improved Forest Cover Detection in Mountain Environments. Int. J. Appl. Earth Obs. Geoinf.; 2015; 35, pp. 320-328. [DOI: https://dx.doi.org/10.1016/j.jag.2014.10.006]

49. Gao, Y.; Zhang, W. A simple empirical topographic correction method for ETM+ imagery. Int. J. Remote Sens.; 2009; 30, pp. 2259-2275. [DOI: https://dx.doi.org/10.1080/01431160802549336]

50. Yamazaki, F.; Liu, W.; Takasaki, M. Characteristics of shadow and removal of its effects for remote sensing imagery. IEEE Int. Geosci. Remote Sens. Symp.; 2009; 4, pp. IV-426-IV-429.

51. Liu, W.; Yamazaki, F. Object-Based Shadow Extraction and Correction of High-Resolution Optical Satellite Images. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens.; 2012; 5, pp. 1296-1302. [DOI: https://dx.doi.org/10.1109/JSTARS.2012.2189558]

52. Tang, Z.G.; Wang, J.; Li, H.Y.; Liang, J.; Li, C.K.; Wang, X. Extraction and assessment of snowline altitude over the Tibetan plateau using MODIS fractional snow cover data (2001 to 2013). J. Appl. Remote Sens.; 2014; 8, 084689. [DOI: https://dx.doi.org/10.1117/1.JRS.8.084689]

53. Necsoiu, M.; Onaca, A.; Wigginton, S.; Urdea, P. Rock glacier dynamics in Southern Carpathian Mountains from high-resolution optical and multi-temporal SAR satellite imagery. Remote Sens. Environ.; 2016; 177, pp. 21-36.

54. Fang, L.; Xu, Y.S.; Yao, W.; Stilla, U. Estimation of glacier surface motion by robust phase correlation and point like features of SAR intensity images. ISPRS J. Photogramm. Remote Sens.; 2016; 121, pp. 92-112. [DOI: https://dx.doi.org/10.1016/j.isprsjprs.2016.08.012]

55. Zaginaev, V.; Petrakov, D.; Erokhin, S.; Meleshko, A.; Stoffel, M.; Ballesteros-Cánovas, J.A. Geomorphic control on regional glacier lake outburst flood and debris flow activity over northern Tien Shan. Glob. Planet. Chang.; 2019; 176, pp. 50-59. [DOI: https://dx.doi.org/10.1016/j.gloplacha.2019.03.003]

56. Thouret, J.C.; Antoine, S.; Magill, C.; Ollier, C. Lahars and debris flows: Characteristics and impacts. Earth-Sci. Rev.; 2020; 201, 103003. [DOI: https://dx.doi.org/10.1016/j.earscirev.2019.103003]

57. Qu, Y.P.; Tang, C.; Liu, Y.; Chang, M.; Tang, D.S. Investigation and Analysis of Glacier Debris Flow in Nyingchi Area, Tibet. Chin. J. Rock Mech. Eng.; 2015; 34, pp. 4013-4022.

58. Cheng, Z.L.; Tian, J.C.; Zhang, Z.B.; Qiang, B. Analysis on Environment of Disasters Resulting from River Blockage in Tibet. J. Catastrophology; 2009; 24, pp. 26-30.

59. Lu, R.R.; Li, D.J. Ice-Snow-Melt Water Debris Flows in the Dongru Longba(Gully) Bomi County, Tibet. J. Glaciol. Geocryol.; 1989; 11, pp. 148-160.

60. Zhang, Z.J.; Zhang, H.; Chang, Y.G.; Chen, Z.C. Review of radiometric calibration methods of Landsat series optical remote sensors. J. Remote Sens.; 2015; 19, pp. 719-732.

61. Adeline, K.R.M.; Chen, M.; Briottet, X.; Pang, S.K.; Paparoditis, N. Shadow detection in very high spatial resolution aerial images: A comparative study. J. Photogramm. Remote Sens.; 2013; 80, pp. 21-38. [DOI: https://dx.doi.org/10.1016/j.isprsjprs.2013.02.003]

62. Belgiu, M.; Tomljenovic, I.; Lampoltshammer, T.J.; Blaschke, T.; Höfle, B. Ontology-based classification of building types detected from airborne laser scanning data. Remote Sens.; 2014; 6, pp. 1347-1366. [DOI: https://dx.doi.org/10.3390/rs6021347]

63. Du, P.J.; Samat, A.; Waske, B.; Liu, S.C.; Li, Z.H. Random forest and rotation forest for fully polarized SAR image classification using polarimetric and spatial features. ISPRS J. Photogram. Remote Sens.; 2015; 105, pp. 38-53. [DOI: https://dx.doi.org/10.1016/j.isprsjprs.2015.03.002]

64. Belgiu, M.; Drăguţ, L. Random forest in remote sensing: A review of applications and future directions. J. Photogramm. Remote Sens.; 2016; 114, pp. 24-31. [DOI: https://dx.doi.org/10.1016/j.isprsjprs.2016.01.011]

65. Jin, H.R.; Li, P.J.; Cheng, T.; Song, B.Q. Land cover classification using CHRIS/PROBA images and multi-temporal texture. Int. J. Remote Sens.; 2012; 33, pp. 101-119. [DOI: https://dx.doi.org/10.1080/01431161.2011.584077]