Susceptibility Assessment of Debris Flow Hazards from the Perspective of Watershed Units Grounded in the Random Forest (RF) Model — A Case Study in the Upper Reaches of the Fujiang River

Susceptibility Assessment of Debris Flow Hazards from the Perspective of Watershed Units Grounded in the Random Forest (RF) Model — A Case Study in the Upper Reaches of the Fujiang River | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Susceptibility Assessment of Debris Flow Hazards from the Perspective of Watershed Units Grounded in the Random Forest (RF) Model — A Case Study in the Upper Reaches of the Fujiang River Weiwei Liu, Xiaoxian Lin, Xin Zhu This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9123924/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 10 You are reading this latest preprint version Abstract As a critical ecological barrier in the upper Yangtze River basin, the upper reaches of the Fujiang River face severe constraints on sustainable development as a consequence of frequent debris flow hazards. For this reason, this study utilized 685 watershed units as evaluation units in the region to accurately identify high-risk areas and core driving factors while establishing a scientific susceptibility assessment framework. Eleven hazard-inducing factors, including watershed area, average elevation, and watershed elevation difference, were integrated to construct a Random Forest (RF) model on the Python platform for debris flow susceptibility assessment and key factor diagnosis. We ultimately arrived at the following conclusions. (1) Watershed area and mean annual precipitation are the primary factors controlling debris flow development, with contribution rates of 0.109 each, followed by the Normalized Difference Vegetation Index (NDVI) and average elevation (both exceeding 0.104), while soil type has the lowest contribution rate (0.032); (2) The RF model demonstrates excellent evaluation accuracy, with an Area Under the Curve (AUC) value of 0.82 and an Accuracy (ACC) value of 0.82. The study area is classified into five susceptibility levels, namely, very low, low, moderate, high, and very high. Distributing in a zonal pattern along the slopes of the main stem of the Fujiang River and its tributaries, such as the Huoxi River and Baicao River, the very high susceptibility areas (579.04 km²) and high susceptibility areas (58.1 km²) are primarily concentrated in unstable valley regions in the northwestern part of Maoxian-Beichuan, the eastern part of Songpan County, and the southeastern part of Pingwu County; (4) Among the 144 debris flow hazard sites within the watersheds, 56.94% (82 sites) are located in very high susceptibility areas, and 27.78% (40 sites) are in high susceptibility areas, validating the reliability of the assessment results. This study innovatively employs watershed units as evaluation units, thereby overcoming the limitations of traditional grid and administrative units. It clarifies the spatial pattern and key driving mechanisms of debris flow susceptibility in the upper reaches of the Fujiang River, providing a scientific basis and technical support for precise disaster prevention and control, land-use planning, and ecological protection in mountainous regions of southwest China. Biological sciences/Ecology Earth and environmental sciences/Ecology Earth and environmental sciences/Environmental sciences Earth and environmental sciences/Hydrology Earth and environmental sciences/Natural hazards Random Forest model (RF) Geographic Information System (GIS) susceptibility assessment debris flow upper reaches of the Fujiang River Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 1. Introduction Debris flows are sudden and destructive geological hazards in mountainous regions. Aside from that, their susceptibility assessment has become a core component of disaster risk prevention and control as well as land-use planning [1,2] . As integrated carriers of hydrological and geomorphological processes, watershed units have emerged as the mainstream spatial scale for susceptibility modeling on account of their ability to accurately capture the complex interactions between topography, geology, and hydrology [3,4] . Notwithstanding the extensive optimization of the assessment framework and integrated multi-source data by global scholars, as well as the significant role this area plays as an ecological barrier, specialized studies have rarely reported to probe deep into key tributaries in the upper reaches of the Yangtze River, especially the upper reaches of the Fu River. Research in this field has evolved from traditional statistical modeling methods, such as Frequency Ratio (FR) and Weight of Evidence (WoE) [5,6] , to advanced machine learning and multi-process coupling techniques. Models like Random Forest (RF) and Decision Tree (DT) are highly regarded for their exceptional nonlinear fitting capabilities and stability in handling multi-factor interactions [7,8] . Recent studies published in Applied Sciences have further advanced this field by exploring innovative negative sample acquisition strategies for RF-based debris flow susceptibility mapping, demonstrating that approaches such as isolation forest (IF) methods are particularly well-adapted to watershed unit datasets [46,47] . Comparative analyses of machine learning algorithms in alpine-valley regions have also confirmed that RF outperforms support vector machine (SVM) models in terms of accuracy, with relative elevation difference emerging as the most prominent evaluation factor. In China, relevant studies have immensely advanced indicator localization and model adaptability. Scholars have established multi-dimensional systems encompassing "topography-geology-hydrology-human activities", tailored to the characteristics of high-altitude regions [2,9] . Recent innovations include integrating Sentinel-1 radar data to overcome data acquisition challenges in cloudy mountainous areas [9] and optimizing watershed units through hierarchical river network features [10] , jointly confirming the effectiveness of machine learning in complex terrains. Despite the aforementioned breakthroughs, several critical issues persist: (1) Inconsistent watershed delineation standards hinder cross-regional comparisons; (2) A lack of specialized studies on the upper reaches of the Fujiang River has left its complex disaster-driving mechanisms poorly understood; (3) Existing models rarely throw light upon the synergistic effects among environmental driving factors in a quantitative manner. Located in the tectonically active transition zone between the Qinghai-Tibet Plateau and the Sichuan Basin, the upper reaches of the Fujiang River face frequent disasters and intensifying human activities [12] . In an effort to mitigate these research gaps, this present study not only adopts watershed units as the fundamental evaluation units, but also employs the RF model to assess debris flow susceptibility. By integrating 11 multi-source factors, this research is predominantly intended to identify key regional drivers and provide scientific support for disaster prevention and mitigation as well as ecological protection in the upper Yangtze River basin. 2. Overview of the Study Area Located in northwestern Sichuan Province, China, the upper reaches of the Fujiang River serve as the source and core flow region of the Fujiang River, which is a secondary tributary of the Yangtze River. This area is not only geographically unique but also historically significant as a vital corridor connecting central China with the southwest, boasting rich natural landscapes and cultural heritage. As demonstrated in Fig. 1 , the upper reaches of the Fujiang River (103°45′–105°15′ E, 31°40′–33°10′ N) refer to the watershed region from the source of the Fujiang River to the control section of the Wudu Town hydrological station in Jiangyou City. The source is located at Xuebaoding, the main peak of the Minshan Mountains, with the main stem flowing southward and major tributaries including the Huoxi River, Huya River, and Pingtong River, covering a total area of approximately 19,000 km². The Fujiang River is a primary tributary on the right bank of the upper Yangtze River, originating from Sanchazi at the northern foot of Xuebaoding, the main peak of the Minshan Mountains, in Songpan County, Aba Tibetan and Qiang Autonomous Prefecture, Sichuan Province. The regional topography exhibits pronounced step-like and transitional characteristics, transitioning from plateaus and high mountains in the northwest to basins in the southeast, which displays extreme elevation differences. The highest point is Xuebaoding in Songpan County, at 5,588 m above sea level, while the lowest point is in southern Jiangyou City, at approximately 500–600 m above sea level [11–13] . Major fault zones in the upper reaches of the Fujiang River include the Longmenshan Fault Zone, Huya Fault Zone, and Minjiang Fault Zone [14] . Through data provided by the Sichuan Provincial Department of Natural Resources, remote sensing interpretation, and field surveys, a total of 192 debris flow hazard sites have been identified in the upper reaches of the Fujiang River, predominantly distributed across 34 townships in five counties within the two cities of Mianyang and Aba Tibetan and Qiang Autonomous Prefecture [14,15] . 3. Data and Research Methods 3.1 Data Sources The primary data include debris flow hazard sites, elevation (DEM), and other datasets, with specific sources detailed in Table 1 . DEM remote sensing imagery of the Fujiang River basin was processed through geometric correction and delineated using watershed boundaries to define the upper reaches of the Fujiang River. Table 1 Data Sources Data Type Parameters Source Debris flow hazard sites 30 m resolution Sichuan Provincial Department of Natural Resources NDVI data 1 km resolution MOD13A1.006 Terra Vegetation Indices 16-Day Global 500m dataset on GEE Land use/cover data (2017–2024) 1 km resolution Wuhan University Land Cover Dataset on CLCD Meteorological data 1 km resolution National Earth System Science Data Center ( http://www.geodata.cn/ ) DEM data 30 m resolution ASTER GDEM30M Digital Elevation Data Soil type data 1: 1,000,000 scale Resource and Environment Science and Data Center, CAS ( https://www.resdc.cn/ ) Vegetation type data Hydrological data 3.2 Selection of Evaluation Units and Indicators 3.2.1 Selection of Evaluation Units As different units yield varying results, selecting appropriate evaluation units is crucial for debris flow susceptibility assessment. For the time being, grid units, watershed units, and administrative units are extensively utilized for susceptibility assessments of geological hazards like debris flows and landslides. Notwithstanding their conspicuous advantages like rapid subdivision and computation on the basis of GIS, grid units fail to link with the geological environment of debris flow regions, resulting in evaluations lacking practical attributes [16] . In contrast, watershed units effectively integrate the geological environment of debris flow regions, which not only takes into consideration topographic and material source conditions for debris flow formation and development, but also facilitates the extraction of susceptibility assessment indicators and factor values to yield reliable results. Administrative units conduct evaluations rooted in entire administrative regions, which is advantageous for government departments at all levels to formulate disaster prevention and mitigation policies, and thereby conduct disaster relief efforts [37] . Nonetheless, their results not only fail to correlate with actual topographic and geomorphic features, but also cannot truly reflect the relationship between debris flows and assessment factors. To this end, this study selects watershed units as the evaluation units for debris flow susceptibility in the upper reaches of the Fujiang River. Using ArcGIS hydrological analysis functions [17] , we compared river networks extracted under different thresholds (1000, 2000, 3000) with the actual river network distribution in the study area. We arrived at a conclusion that a threshold of 1000 yielded a river network and surrounding watersheds largely consistent with the actual distribution. For this reason, a threshold of 1000 was used to extract the river network, ultimately dividing the upper reaches of the Fujiang River into 685 watershed units with an average area of 18 km². Figure 1 depicts the distribution of debris flow watersheds in the upper reaches of the Fujiang River. 3.2.2 Selection of Evaluation Indicators The development of debris flows requires at least three conditions: ① steep slopes, ② a large amount of loose materials, and ③ an adequate water source. Steep slopes are one of the crucial conditions for debris flow development, as they provide sufficient potential energy, enabling the rapid flow of debris flows under the influence of gravity. Debris flows are typically formed by the mixture of water and mud in slopes or gullies containing a large amount of loose materials (such as soil, sand, rock debris, etc.). As substantial rainfall, snowmelt, mountain meltwater, or groundwater outbursts can provide sufficient water sources, water functions as a significant factor in debris flow development, thereby allowing loose materials to mix with water and form debris flows. As a result, the selection of evaluation factors for debris flow susceptibility should take into account the scientificity, practicality, and correctness of the factors. Grounded in the relevant explanations in Chap. 6 Survey Content of Disaster-Pregnant Geological Conditions of the Technical Requirements for Geological Survey (1:50,000) by the China Geological Survey Bureau [18] , as well as relevant research findings on the causes of debris flow disasters in China, and considering the actual situation of the study area, factors such as topography and geomorphology, meteorology and hydrology, soil type, and vegetation coverage were selected as evaluation factors for debris flow susceptibility in the upper reaches of the Fujiang River. Taking into account the availability of each evaluation factor and combining with the debris flow basin units in the study area, 11 evaluation factors were ultimately selected to construct an evaluation system for debris flow susceptibility in the upper reaches of the Fujiang River for susceptibility assessment. 3.2.3 Quantitative Grading Standards for Evaluation Factors As a result of the different evaluation criteria for each indicator, involving qualitative and quantitative, absolute and relative quantities, and differences in units of measurement, as well as considering the consistency and rationality of subsequent overlay analysis of layers, reference was made to the Technical Requirements for Geological Survey (1:50,000) by the China Geological Survey Bureau and relevant research results by Zhu Xiaolong et al. [19] . Each indicator's corresponding disaster risk level was classified, ranging from extremely low to extremely high, unified as Grades Ⅰ-Ⅴ, as revealed in Table 2 . Table 2 Quantitative Grading Table for Factor Indicators Evaluation Factor Indicator Evaluation Indicator Grading (Risk Level) Extremely Low (I) Low (II) Medium (III) High (IV) Extremely High (V) Basin Area/km² 0.006–7.387 7.38–17.63 17.63–30.45 30.45–51.99 51.99–134.34 Average Elevation/m 493–1200.07 1200.07–1816.88 1816.88–2523.955 2523.95–3291.20 3291.20–4329.25 Basin Elevation Difference/m 0–689 689–1253 1253–1693 1693–2239 2239–3388 Average Slope/(°) 0.53–10.27 10.27–21.43 21.43–26.39 26.39–30.11 30.11–45.70 Average Plan Curvature -0.79–།0.43 -0.43–0.09 -0.09–0.006 0.006–0.029 0.029–0.413 Average Profile Curvature -0.20–0.002 0.002–0.015 0.015–0.056 0.056–0.252 0.252–1.523 Melton Ratio 0–0.214 0.124–0.567 0.567–0.846 0.846–1.368 1.368–3.178 NDVI Index 0.017–0.226 0.226–0.332 0.332–0.429 0.429–0.528 0.528–0.681 Soil Type Paddy soil, swamp soil, stony soil, stony loam Yellow soil, dark brown soil, tidal brown soil, brown coniferous forest soil Limestone soil, yellow lime soil, black lime soil Brown soil, brown loam soil Purple soil, coarse-bone soil Average Annual Precipitation/mm 679.70–748.66 748.66–778.93 778.93–809.21 809.21–849.58 849.58–894.15 Average Annual Temperature/℃ -4.61–2.13 2.13–6.18 6.18–9.64 9.64–13.02 13.02–16.90 4 Results The Random Forest (RF) model is an ensemble learning method employed to solve classification and regression problems [20–23] . It consists of multiple decision trees, each constructed rooted in a randomly selected subset of features and randomly selected training samples. 4.1 Analysis of the Area Proportion and Spatial Distribution of Risk Levels for Evaluation Indicators As illustrated in Fig. 2 , the spatial distribution of risk levels for each indicator was classified and visualized in accordance with the aforementioned grading standards. By utilizing spatial statistical tools, the area and proportion of regions with different risk levels for each evaluation factor were summarized, with the results revealed in Table 3 . As evidently demonstrated by the results, debris flow disasters in the upper reaches of the Fujiang River are primarily influenced by basin area and average annual precipitation, which are the largest controlling factors. Secondary influencing factors mainly include NDVI, average elevation, and average elevation difference, while average annual temperature and soil type have the least impact. Debris flow disaster points are mainly concentrated in areas with slopes ranging from 26.39° to 30.11° and elevations between 1,200.07 and 1,816.88 m, which are medium-elevation regions. A large basin elevation difference provides suitable topographic conditions for debris flow disasters. Topographic profile within the basin exhibits strong incision, with towering peaks and deep gullies, resulting in strong hydrodynamic forces that easily carry material sources, thereby triggering debris flow disasters. The size of the basin area and average annual precipitation directly affect the amount of rainfall accumulated within the basin. The more rainfall accumulates in a debris flow basin, the more likely it is to carry enormous slope debris and sediment, further increasing the likelihood of debris flow disasters. Table 3 Area and Proportion of Risk Levels for Each Evaluation Factor Evaluation Factor Indicator Evaluation Indicator Grading (Risk Level) Extremely Low (I) Low (II) Medium (III) High (IV) Extremely High (V) Area /KM 2 Proportion % Area /KM 2 Proportion % Area /KM 2 Proportion % Area /KM 2 Proportion % Area /KM 2 Proportion % Basin Area/km² 346.71 2.85 3049.73 25.07 3864.37 31.77 2859.53 23.51 2044.38 16.80 Average Elevation/m 1401.42 11.52 4183.14 34.39 2565.78 21.09 2387.11 19.62 1627.27 13.38 Basin Elevation Difference/m 417.59 3.43 2063.73 16.96 4225.97 34.74 3873.65 31.85 1583.78 13.02 Average Slope/(°) 333.33 2.74 1499.31 12.33 3399.64 27.95 4021.61 33.06 2910.83 23.92 Average Plan Curvature 1190.71 9.79 4292.07 35.28 4591.36 37.75 1980.15 16.28 110.43 0.90 Average Profile Curvature 1460.68 12.01 7491.22 61.57 3187.87 26.21 24.08 0.2 0.87 0.01 Melton Ratio 365.16 3 2830.47 23.27 3246.75 26.69 4342.54 35.70 1379.80 11.34 NDVI Index 3680.38 30.25 2958.98 24.32 2041.04 16.78 2201.28 18.1 1283.04 10.55 Soil Type 7451.24 61.25 19.00 0.16 104.49 0.86 3097.65 25.46 1492.34 12.27 Average Annual Precipitation/mm 1803.05 14.82 2623.54 21.57 2908.72 23.91 3315.46 27.25 1513.95 12.45 Average Annual Temperature/℃ 1301.25 10.7 2517.57 20.7 2320.43 19.08 2672.62 21.97 3352.85 27.55 4.2 Analysis of Factor Contribution Rates In this study, data from 192 debris flow disasters in the upper reaches of the Fujiang River were used as research samples. Using ArcGIS10.7, DEM data for the upper reaches of the Fujiang River basin was divided into 685 basin units. A marker column attribute was added to each basin unit, with basin units where debris flows had occurred marked as "1" and those where debris flows had never occurred marked as "0". Consequently, the problem of debris flow susceptibility evaluation was transformed into a binary classification problem [24] . As the results illustrate, there were 33 basin units where debris flows had occurred and 628 basin units where debris flows had never occurred in the upper reaches of the Fujiang River. Initially, through literature review, summarizing previous experiences, and continuous testing, the basic dataset was divided into an 80% training set and a 20% test set in an 8:2 ratio for training and testing the debris flow susceptibility evaluation model in the upper reaches of the Fujiang River [26–27] . The Random Forest model was constructed by adopting the Scikit-learn framework built into the Python language. After hyperparameter tuning, the number of decision trees in the Random Forest model was set to 100, the splitting metric was "gini", the maximum tree depth was "None", the maximum number of features was "auto", and the number of parallel jobs was "None". The factor contribution rates of the evaluation indicators included in the Random Forest model after hyperparameter tuning are depicted in Fig. 3 . As suggested by an all-round observation, the factor contribution rates of basin area and average annual precipitation are the same, both at 0.109, making them the factors with the highest contribution rates among the evaluation indicators for inducing debris flow outbreaks. This also confirms that basin area and average annual precipitation have significant impacts on debris flow outbreaks in the basin. Soil type has the lowest contribution rate, at 0.032, in inducing debris flow outbreaks. The contribution rates of the two factors, watershed area and annual average precipitation, display the same and relatively high values. This phenomenon suggests that watershed area and annual average precipitation may have significant impacts on the analysis of debris flow susceptibility. Aside from that, larger watershed areas and higher annual average precipitation may be associated with the occurrence of debris flows to a certain extent. NDVI is an indicator for measuring the condition of surface vegetation. A higher NDVI value indicates more luxuriant vegetation cover. As vegetation can stabilize soil and reduce the likelihood of soil erosion and debris flows, higher NDVI values may be correlated with the mitigation or prevention of debris flows. Both the average elevation and the elevation difference within the watershed are terrain-relevant. Steep terrain and large elevation changes may give rise to soil erosion and the occurrence of debris flows. Larger profile curvature, plan curvature, and slope potentially heighten the risk of soil erosion and debris flows. A higher Melton ratio may suggest a certain correlation between steep terrain and debris flow susceptibility. The contribution rates of the two factors, annual average temperature and soil type, are relatively low, suggesting that they may not have obvious direct impacts on the occurrence of debris flows. In summary, on the basis of the given factor contribution rates, watershed area, annual average precipitation, NDVI, average elevation, watershed elevation difference, average profile curvature, average plan curvature, average slope, and Melton ratio may be important factors influencing debris flow susceptibility. Simultaneously, other factors such as annual average temperature and soil type, with relatively low contribution rates, may conduct a trivial role in the analysis of debris flow susceptibility. 4.3 Debris Flow Susceptibility Analysis Afterwards, a susceptibility assessment of debris flows in the upper reaches of the Fujiang River was conducted grounded in the Random Forest model. After the model ran, susceptibility index values were obtained for 685 watershed units in the upper reaches of the Fujiang River. By importing these susceptibility values into ArcGIS 10.7 software, the debris flow susceptibility values in the upper reaches of the Fujiang River were classified into five different levels according to the susceptibility index of the watershed units, namely, extremely low susceptibility area, low susceptibility area, moderate susceptibility area, high susceptibility area, and extremely high susceptibility area. A susceptibility assessment map of debris flow disasters in the upper reaches of the Fujiang River was subsequently generated (Fig. 4 ). As evidenced in the figure, we can draw several conclusions below: (1) The extremely high and high susceptibility areas for debris flows in the upper reaches of the Fujiang River are mainly distributed in the northwestern part from Mao County to Beichuan, the eastern part of Songpan County, and the southeastern part of Pingwu County. They are mainly concentrated in extremely unstable valley areas, mostly distributed along the slopes on both sides of the main stream of the Fujiang River (from Songpan to Pingwu section), Huoxi River, Baicao River, and other rivers. (2) The moderate susceptibility areas for debris flows in the upper reaches of the Fujiang River are mainly distributed around the high susceptibility areas. This region is mainly mountainous. Since other geological disasters are also distributed in this area, they further provide material sources for debris flow disasters. Under specific conditions such as heavy rainfall scouring, debris flow disasters are likely to occur. (3) The extremely low and low susceptibility areas for debris flows in the upper reaches of the Fujiang River are mainly distributed in the central, northern, and eastern parts of the upper reaches of the Fujiang River, as well as in most areas of Jiangyou City. The terrain in this region is relatively gentle, with bedrock exposed on the surface, lacking the conditions for debris flow development. With regard to the number of watersheds (Table 4 ), considering both the number and area of watersheds, the susceptibility zoning in the upper reaches of the Fujiang River is highly consistent. The extremely low susceptibility area occupies an absolutely dominant position, with both the number of watersheds (635) and the area (11,282.63 km²) accounting for 92.75% of the total. The extremely high susceptibility area comes next, with both indicators (31 in number and 579.04 km² in area) accounting for 4.76%. The combined proportion of the other three susceptibility areas (moderate, low, and high) is less than 2.5%. Table 4 Statistics on Debris Flow Susceptibility in the Watersheds of the Upper Reaches of the Fujiang River Risk Zone Number of Watersheds / Units Watershed Area / km² Area Proportion / % Ⅰ (Extremely Low) 635 11282.63 92.75 Ⅱ (Low) 4 74.05 0.61 Ⅲ (Moderate) 10 170.89 1.40 Ⅳ (High) 3 58.10 0.48 Ⅴ (Extremely High) 31 579.04 4.76 Total 685 12164.72 100 As suggested by comprehensive statistics on the 144 debris flow watersheds in the upper reaches of the Fujiang River and their susceptibility level zoning (Table 5 ) Table 5 Statistics on Debris Flow Susceptibility in the Upper Reaches of the Fujiang River County-level Administrative/prone Level Extremely Low Susceptibility Low Susceptibility Moderate Susceptibility Area High Susceptibility Area Extremely High Susceptibility Area Jiangyou City 1 0 0 1 0 Pingwu County 1 4 3 11 41 Beichuan County 0 2 7 14 21 Maoxian County 0 0 0 4 8 Songpan County 2 0 2 10 12 4.4 Model Accuracy Evaluation The Receiver Operating Characteristic (ROC) curve (Fig. 5 ) is a tool employed to assess the performance of classification models. In this study, with the upper reaches of the Fujiang River as the target area, an innovative approach was adopted by using watershed units as the basic evaluation units. Rooted in a catchment threshold of 1000, the study area was divided into 685 watersheds, among which extremely low/low susceptibility areas accounted for 93.36%, and extremely high/high susceptibility areas accounted for 5.24%. By integrating 11 disaster-causing factors from four major categories, namely topography and geomorphology, meteorology and hydrology, vegetation coverage, and geology and soil, a Random Forest (RF) model was constructed on the Python platform to evaluate debris flow susceptibility [33–35] . Afterwards, the ROC curve and the Area Under the Curve (AUC) were employed to evaluate the model's performance. The ROC curve measures classification effectiveness by plotting the relationship between the True Positive Rate (TPR) and the False Positive Rate (FPR) at different thresholds, with a curve closer to the top-left corner indicating better performance. Verified by 144 historical debris flow disaster points, the model achieved an AUC value of 0.944 and an Accuracy (ACC) value of 0.82. On top of that, 56.94% of the disaster points were concentrated in extremely high susceptibility areas, and 27.78% were located in high susceptibility areas, thereby confirming the reliability and favorable applicability of the evaluation results, which can provide a scientific basis for precise prevention and control of debris flows in mountainous regions of Southwest China. 5 Discussion 5.1 Optimization and Rationality Comparison of Evaluation Units The rational selection of evaluation units is crucial for the elevation of assessment accuracy. In general, traditional grid units are not only disadvantageous for the integrity of the geological environment, but are also challenging to reflect the continuous process of debris flow, which involves "catchment-sediment production-transportation". As a result of their artificial boundaries, administrative units fail to align with the natural continuity of the terrain, often leading to distorted results [12,13] . In this study, river networks were extracted using a threshold of 1000, and 685 watershed units (with an average area of 18 km²) were delineated through the hydrological analysis function of ArcGIS. This approach not only preserves the coupling relationships among terrain, hydrology, and material sources but also matches the patterns of disaster development, effectively overcoming the limitations of traditional units. In contrast to the study by Wang Shige et al. [8] in the Dadu River basin, this study incorporates additional factors such as average planar curvature, Melton ratio, and annual average temperature, providing a more comprehensive coverage of the material sources, hydrodynamic forces, and topographic base conditions for debris flow formation, aligning with the complex mountainous characteristics of the upper reaches of the Fujiang River. In comparison with the study by Zhu Xin et al. [28] , which used a catchment threshold of 3000 to divide 2948 fragmented units (with an average area of 4 km²), the units in this study are of moderate size, reducing data processing redundancy costs while avoiding the dissection of natural watershed coupling relationships. The susceptibility zoning is clear (extremely low/low susceptibility areas account for 93.36%, and extremely high/high susceptibility areas account for 5.24%), facilitating the precise identification of core risk areas and the formulation of prevention and control measures. 5.2 Advantages of Model Selection and Identification of Core Disaster-Causing Factors The choice of model and factor combination can exert direct influences on the objectivity of the evaluation. Traditional methods such as the Analytic Hierarchy Process (AHP) and the Frequency Ratio (FR) method rely on subjective weighting or single statistical rules, thereby rendering it challenging to quantify the nonlinear interactions among multiple factors [29,30] . Some machine learning models suffer from inadequate factor screening specificity; for example, Wilford et al.'s model focused solely on topographic factors [43] , and Rowbotham's logistic regression model lacked vegetation and soil parameters [44] . The RF model selected in this study possesses strong feature screening and nonlinear fitting capabilities, enabling it to automatically identify core driving factors and avoid human interference. This advantage aligns with recent findings in Applied Sciences, where machine learning-based feature importance analysis has been increasingly integrated with hyperparameter optimization techniques such as Bayesian Optimization to enhance both predictive performance and model explainability in geohazard assessments [48] . The integration of feature importance analysis with susceptibility modeling, as demonstrated in this study, provides clearer insight into the model's decision-making process and enables more reliable engineering interpretation [48] .As the above results demonstrate, watershed area and annual average precipitation are the primary driving factors (both with a contribution rate of 0.109), aligning with the regional characteristics of "high mountains, deep valleys, and concentrated precipitation". Larger watershed areas provide space for the accumulation of loose materials and water collection, while concentrated precipitation supplies the hydrodynamic force for debris flow initiation. The Normalized Difference Vegetation Index (NDVI) and average elevation are secondary key factors (both with a contribution rate exceeding 0.104), reflecting the regulation of soil stability by vegetation and the disaster-causing characteristics of steep terrain and abundant material sources in high-altitude areas. Soil type has the lowest contribution rate (0.032), which can be attributable to the fact that the soil in the study area is predominantly purple soil and skeletal soil with weak erosion resistance, and the impact of soil type differences on disasters is relatively uniform. This result accurately reveals the region-specific driving mechanisms, successfully addressing gaps in previous studies [15,45] . The contribution patterns observed in our study—where topographic conditions serve as prerequisites while precipitation and vegetation factors modulate disaster occurrence—are consistent with recent machine learning-based susceptibility assessments in alpine-valley regions documented in Applied Sciences and related publications. These studies have similarly emphasized that topographic conditions are fundamental prerequisites for debris flow development, while factors such as precipitation, vegetation cover, and anthropogenic influence play critical roles in determining the spatial distribution of high-susceptibility zones . 5.3 Spatial Pattern Characteristics of Susceptibility and Causal Analysis The spatial pattern of debris flow susceptibility in the upper reaches of the Fujiang River is tightly correlated with the geological environment and human activities. Extremely high susceptibility areas (579.04 km²) and high susceptibility areas (58.1 km²) are concentrated in unstable valley regions in the northwestern part of Maoxian-Beichuan, the eastern part of Songpan County, and the southeastern part of Pingwu County, exhibiting a zonal distribution along the slopes on both sides of the main stream and tributaries of the Fujiang River. Characterized by active geological structures and fragmented rock masses, this region lies within the influence range of the Longmenshan and Huya fault zones [10] . Coupled with significant topographic relief (up to 3388 m), average slopes ranging from 26.39° to 45.70°, and annual average precipitation between 849.58 and 894.15 mm, multiple factors contribute to the high incidence of disasters. Moderate susceptibility areas are distributed in the high mountains surrounding the high susceptibility areas, where material sources and hydrodynamic conditions are weaker, and disasters are likely to occur only under extreme rainfall conditions. Extremely low/low susceptibility areas account for 93.36% of the total area and are concentrated in the gentle regions of Jiangyou City and the central and northern parts of the basin, where the terrain is flat and bedrock is exposed, lacking the conditions for debris flow development. This pattern is consistent with the geomorphic characteristics and fault activity response laws revealed by Chen Hao et al. [15] , further confirming the rationality of the evaluation results. 5.4 Future Research Directions and Prospects Altogether, future research can be deepened in the following aspects. First and foremost, it is advisable to integrate high-resolution Sentinel-2 imagery and unmanned aerial vehicle (UAV) survey data to unify factor resolution and enhance evaluation accuracy in local small watersheds. Apart from that, it is preferable to introduce dynamic factors such as seismic ground motion parameters and land use changes, and construct a dynamic evaluation system combined with time-series NDVI to quantify the coupled effects of human activities and natural changes. In line with emerging trends in Applied Sciences special issues on "Intelligent Technologies in Geotechnical Engineering and Geological Hazards" and "Applications of Artificial Intelligence in Geotechnics and Engineering Geology," future work should also explore the integration of physics-informed AI-driven methods that combine machine learning with hydro-mechanical coupling models[48,50]. Such approaches can better capture the complex fracture–seepage interactions and permeability evolution processes that govern debris flow initiation under rainfall and seismic triggers [49] .Moreover, it holds pivotal significance to compare the performance of the RF model with other models, such as XGBoost and Support Vector Machine (SVM) to screen the optimal model for mountainous basins. Last but not least, grounded in the evaluation results, delineate differentiated prevention and control zones and propose comprehensive prevention and control schemes that combine engineering governance with non-engineering measures, considering terrain and socioeconomic characteristics. This will provide more practical technical support for the construction of ecological barriers and land space planning in the upper reaches of the Yangtze River. Declarations Author Contribution Scholar Liu Weiwei is responsible for writing the main manuscript, Professor Lin Xiaoxian is responsible for reviewing and revising the manuscript, and Zhu Xin is responsible for providing data. Data Availability The data used in this study are derived from the following public sources:1. In addition to the mudslide disaster sites in the upper reaches of the Weijiang River, which are obtained by the Sichuan Provincial Department of Natural Resources, the applicability of these data is limited. They involve coordinates and are authorized by this study, so the disaster point data is not disclosed.2.Data source of China's administrative divisions:https://cloudcenter.tianditu.gov.cn/administrativeDivision(This data cannot be linked due to the restrictions of the external network, but the author confirms that the data set used during the current research is obtained and used at reasonable request.)3.DEM data set source:https://www.gebco.net/data\_and\_products/gridded\_bathymetry\_data/4.Source of river water system data:https://zenodo.org/records/138419105.Source of temperature data:https://www.ncei.noaa.gov/data/global-summary-of-the-day/archive/6.Precipitation data:https://www.ncei.noaa.gov/data/global-summary-of-the-day/archive/7.NDVI data set:https://doi.org/10.11888/Terre.tpdc.300328.Further details can be found in the original publications. References Fell R. Landslide risk assessment and acceptable risk[J]. Canadian Geotechnical Journal, 1994, 31(2): 261-272. Cui P, Chen X Q, Wang S G. Progress and prospects of research on debris flow disaster prevention and control in China[J]. Acta Geographica Sinica, 2020, 75(1): 3-18. Montgomery D R, Dietrich W E, Thompson S E. A runoff generation threshold for landsliding in the Oregon Coast Range[J]. Water Resources Research, 1995, 31(1): 115-127. Yang Q K, Wang L, Li W L. Impact of watershed unit division on the accuracy of debris flow susceptibility evaluation[J]. Geographical Research, 2021, 40(5): 1365-1378. Kanungo D P, Arora M K, Gupta R P, et al. Landslide risk assessment using concepts of danger pixels and fuzzy set theory in Darjeeling Himalayas[J]. Landslides, 2008, 5(4): 407-416. Pourghasemi H R, Pradhan B, Gokceoglu C. Application of fuzzy logic and analytical hierarchy process (AHP) to landslide susceptibility mapping at Haraz watershed Iran[J]. Natural Hazards, 2012, 63(3): 965-996. Khosravi K, Pham B T, Chapi K, et al. A comparative assessment of decision trees algorithms for flash flood susceptibility modeling at Haraz watershed, northern Iran[J]. Science of the Total Environment, 2018, 627: 744-755. Wu C Y, Lin S Y. Event-based landslide susceptibility models in Shihmen watershed, Taiwan: accounting for the characteristics of rainfall events[J]. Environmental Monitoring and Assessment, 2022, 194(5): 405. Wang S G, Li W L, Yang Q K. Evaluation of Debris Flow Susceptibility in Southwest Mountainous Areas Based on Basin Units: A Case Study of the Dadu River Basin [J]. Acta Geologica Sinica, 2022, 96 (8): 2654-2666. Chen J, Zhang G R, Liu H H. Debris flow susceptibility evaluation in the mountainous regions of Southwest China based on Sentinel-1 and Random Forest model[J]. Chinese Journal of Rock Mechanics and Engineering, 2021, 40(S1): 3012-3020. Tian C S. Debris flow hazard assessment under future climate change in the upper reaches of the Minjiang River[D]. Chengdu: Chengdu University of Technology, 2016. Hu B, Zhang Y, Wang H R, et al. Spatial distribution characteristics of fuel and fire risk assessment in typical forest ecosystems in the upper reaches of the Yangtze River: A case study of Pingwu County, Sichuan Province[J/OL]. Journal of Catastrophology, 1-10[2025-10-26]. https://link.cnki.net/urlid/61.1097.P.20251010.1541.004. Li M Y, Pan H Z, Chen P, et al. NDVI changes in the upper reaches of the Yangtze River and their impact on the sediment and water processes entering the Three Gorges Reservoir[J/OL]. Resources and Environment in the Yangtze Basin, 1-21[2025-10-26]. https://link.cnki.net/urlid/42.1320.X.20250912.1921.002. Li K, Zhao J S, Lin Y L, et al. Debris flow susceptibility evaluation in Dongchuan based on RF and SVM models[J]. Journal of Yunnan University (Natural Sciences Edition), 2022, 44(1): 107-115. Sichuan Provincial Department of Natural Resources. Sichuan Provincial Geological Disaster Prevention and Control Plan (2021-2025)[R]. Chengdu: Sichuan Provincial Department of Natural Resources, 2022. Chen H, Dong T X, Li Y, et al. Geomorphic characteristics of the upper reaches of the Fujiang River basin and their response to fault activity[J]. Journal of Mountain Science, 2020, 38(4): 542-551. Li K, Zhao J S, Lin Y L, et al. Debris flow susceptibility evaluation based on different slope unit division methods and BP neural network[J]. Bulletin of Surveying and Mapping, 2022, 1(8): 68-74. Pan R. Debris flow susceptibility evaluation in Shimian County after the "8•11" disaster[D]. Sichuan Agricultural University, 2022. DOI: 10.27345/d.cnki.gsnyu.2022.000530. China Geological Survey. Technical standards for geological surveys[S]. Technical requirements for geological disaster surveys (1:50000). China Geological Survey, Ministry of Natural Resources. DD.2019.08. Zhu X L, Ma B H, Zhang W X, Chen Y. Main disaster-causing factors and their control and impact on the distribution of geological disasters in the Zhangjiakou area of Hebei Province[J]. Chinese Journal of Geological Hazard Control, 2015, (01): 103-107. Wang L J. Morphological characteristics and empirical hazard models of debris flows[D]. Beijing: Beijing Forestry University, 2021. Li Y M, Yang L, Wei S H, et al. Debris flow susceptibility evaluation in Nujiang Prefecture based on small watershed units[J]. Resources and Environment in the Yangtze Basin, 2019, 28(10): 2419-2428. Wang, X. P. Research on typical fault diagnosis methods for chillers based on random forest [D]. Guangzhou: South China University of Technology, 2020. Li, J. Y., Guo, S. S. An improved random forest algorithm for load balancing in parallel rendering systems [J]. Modern Computer, 2020, 26(8): 22-25. Lin, X. X., Chen, H., Yin, M. H., et al. Theory and practice of geological disaster prevention and control in seismic disturbance zones [M]. Beijing: China Environmental Publishing Group, 2022. Li, K., Zhao, J. S., Lin, Y. L., et al. Debris flow susceptibility assessment based on different slope unit division methods and BP neural network [J]. Bulletin of Surveying and Mapping, 2022, (8): 68-74. Chen, Y. J., Yu, B., Zhu, Y., et al. Variation characteristics of critical rainfall for debris flow after earthquakes: A case study of Xiaogangjian Gully in the Wenchuan earthquake zone [J]. Journal of Mountain Science, 2013, 31(3): 356-361. Xu, R. G., Zheng, W. M., Ba, R. J., et al. Study on the development characteristics of debris flow in the Dadu River Basin [J]. Yangtze River, 2013, 44(S1): 117-119+131.DOI: 10.16232/j.cnki.1001-4179.2013.s1.007. Zhu, X., Lin, X. X., Chen, H., et al. Distribution pattern and risk prevention and control strategies of debris flow in the upper reaches of the Fujiang River [D]. Mianyang: Mianyang Teachers' College, 2024. Li, W. L., Yang, Q. K., Wang, L. Application of an improved frequency ratio method in debris flow susceptibility assessment in the Qinba Mountain area [J]. Journal of Natural Resources, 2019, 34(7): 1492-1504. Zhong, X. W. Research on rail transit flow prediction based on AFC ticket checking big data [D]. Chongqing: Chongqing University of Technology, 2021. Mosaffaie J, Salehpour Jam A, Sarfaraz F, et al. Landslide risk assessment based on susceptibility and vulnerability[J]. Environment, Development and Sustainability, 2024, 26(12): 9285-9303. Petschko H, Brenning A, Bell R, et al. Assessing the quality of landslide susceptibility maps–case study Lower Austria[J]. Natural Hazards and Earth System Sciences, 2014, 14(1): 95-118. Tien Bui D, Tsangaratos P, Nguyen V T, et al. Comparing the prediction performance of a Deep Learning Neural Network model with conventional machine learning models in landslide susceptibility assessment[J]. Catena, 2020, 188: 104426. Zhang, G. R., Chen, J., Li, W. L. Comparison of different machine learning models in debris flow susceptibility assessment in the Yalong River Basin [J]. Journal of Mountain Science, 2023, 41(2): 321-330. Meng, X. M., Chen, G., Guo, P., et al. Research progress and prospects of landslide and debris flow disasters in the Bailong River Basin [J]. Marine Geology & Quaternary Geology, 2013, 33(4): 1-15. Xia, J. W., Guo, H. Z. Discussion on the distribution characteristics and main controlling factors of landslides in the upper reaches of the Yangtze River [J]. Hydrogeology & Engineering Geology, 1997, (1): 19-22+32. Song, G. J., Huang, J. C., Zhang, G., et al. Research on zoning evaluation of mine geological environment in Xuchang City, Henan Province [J]. Geology and Resources, 2022, 31(2): 221-227. Feng, J. Y. Research on geological disaster risk assessment and risk management in Luolong County, Tibet [D]. Sichuan Agricultural University, 2025. DOI: 10.27345/d.cnki.gsnyu.2025.000332. Li, H., Zhai, X., Li, C. X., Pan, Y., Li, J. P., Yuan, W. Cloud model evaluation method for debris flow disaster susceptibility in Hebei Province: A case study of debris flow in Zhaogou Village, Xingtai [J]. Science Technology and Engineering, 2024, 24(25): 10884-10891. Wang, Q. Y. Debris flow risk assessment in Xinzhou area based on semi-supervised learning [D]. 10.27162/d.cnki.gjlin.2024.003279. Chen, J. S. Research on performance optimization of three-dimensional network-on-chip for multi-core processors [D]. Guangzhou: Guangdong University of Technology, 2022. Yin, H. L. Debris flow susceptibility assessment of the China-Nepal traffic corridor under the background of climate change [D]. Lanzhou University, 2024. DOI: 10.27204/d.cnki.glzhu.2024.000967. Wilford D , Sakals M E , Innes J L ,et al. Recognition of debris flow, debris flood and flood hazard through watershed morphometrics[J]. Landslides, 2004, 1(1): 61-66. Rowbotham D, Louis S J.The identification of debris torrent basins using morphometric measures derived within a gis[J]. Geografiska Annaler, 2005, 87(4): 527- 537. Perera E, Jayawardana D T, Jayasinghe P,et al. Landslide vulnerability assessment based on entropy method: a case study from Kegalle district, Sri Lanka[J]. Modeling Earth Systems and Environment, 2019, 5(4): 1635-1649. Gao R Y, Wu D, Liu H L, et al. Multiple negative-sample acquisition strategies for machine learning-based debris flow susceptibility mapping considering different sample representation forms[J]. Applied Sciences, 2024, 14(20): 9240. Zhou J Y, Huang J C, Sun Z B, et al. Machine learning-based debris flow susceptibility mapping in the alpine-valley area of the Nujiang River Basin[J]. PeerJ, 2024, 12: e17352. (Cited in Applied Sciences special issue context) Karahan H, Alkaya D. Integrating SVR optimization and machine learning-based feature importance for TBM penetration rate prediction[J]. Applied Sciences, 2026, 16(1): 355. Sun Z H, Tan Q W, Wang L Q, et al. New advances in rock fractures and landslide forecasting[J]. Applied Sciences, 2026, Special Issue. Wang Y J, Cai H P, Li Q Q, et al. Applications of artificial intelligence in geotechnics and engineering geology—2nd Edition[J]. Applied Sciences, 2025, Special Issue. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 14 Apr, 2026 Reviews received at journal 13 Apr, 2026 Reviews received at journal 30 Mar, 2026 Reviewers agreed at journal 29 Mar, 2026 Reviewers agreed at journal 24 Mar, 2026 Reviewers invited by journal 24 Mar, 2026 Editor assigned by journal 23 Mar, 2026 Editor invited by journal 23 Mar, 2026 Submission checks completed at journal 21 Mar, 2026 First submitted to journal 21 Mar, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9123924","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":612300930,"identity":"06595a98-87cb-41a7-bdfd-a6b46e3c5d0e","order_by":0,"name":"Weiwei Liu","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Weiwei","middleName":"","lastName":"Liu","suffix":""},{"id":612300931,"identity":"5f320b6e-824b-43cb-8dac-41892578d8c5","order_by":1,"name":"Xiaoxian Lin","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA1klEQVRIie3PsQqCQBjA8QvhXL689SvDZ/AQpMdRBCfdHUKabormeotAiMZAuskHcCvoBWppiuh0FrWt4f7DwcH9uO8jRKf7wyxDHZM3AjPOzR0Xg4S2ZL10ZiJoCQwT0pLMc2VLyAhiTqU9PWFYSEPO69USiFleDv2DWRHfVxgeJY28RKrBII7rfgL89qCKXDf8nlBFEPwh4p4DRQoBbpl8xhF+ewr0XAr8nopxxOO7Ch1sdkm3CHRoF8Yq34YsByYMaSev3GFmKXtJx7+/PdfpdDpdV1+TdTp3LdqfwwAAAABJRU5ErkJggg==","orcid":"","institution":"","correspondingAuthor":true,"prefix":"","firstName":"Xiaoxian","middleName":"","lastName":"Lin","suffix":""},{"id":612300932,"identity":"1d5f6886-4910-4932-9dc5-732491948ca2","order_by":2,"name":"Xin Zhu","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Xin","middleName":"","lastName":"Zhu","suffix":""}],"badges":[],"createdAt":"2026-03-14 16:23:24","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9123924/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9123924/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":105519000,"identity":"ca26f594-5733-410b-bf0a-800579f34b54","added_by":"auto","created_at":"2026-03-27 01:46:28","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":117231,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSchematic Diagram of the Division of Debris Flow Basin Units in the Upper Reaches of the Fujiang River\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-9123924/v1/8975570183648d3c62b1ae5c.jpeg"},{"id":105519005,"identity":"582e96c8-f4bd-441a-a58a-5ca39aefcee3","added_by":"auto","created_at":"2026-03-27 01:46:28","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":249262,"visible":true,"origin":"","legend":"\u003cp\u003eGrading Map of Debris Flow Basin Factors\u003c/p\u003e","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-9123924/v1/0c4640e3f1347fa3ad015ad9.jpeg"},{"id":105567087,"identity":"4cf37556-fe7b-4b4c-8d2d-5119c2921b41","added_by":"auto","created_at":"2026-03-27 12:58:16","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":74782,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFactor Contribution Rates\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-9123924/v1/e140ff4057aeaaf5f802860b.png"},{"id":105519002,"identity":"b6d17f66-cb4e-4096-9ce0-6c0837fdcc61","added_by":"auto","created_at":"2026-03-27 01:46:28","extension":"jpeg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":125836,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSusceptibility Assessment Map of Debris Flow Disasters in the Upper Reaches of the Fujiang River\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-9123924/v1/181caeffb84a2c7f1b3f59b3.jpeg"},{"id":105519001,"identity":"bb0d6d0f-2226-4437-a223-c7c4d555aaeb","added_by":"auto","created_at":"2026-03-27 01:46:28","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":59010,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eROC Curve\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9123924/v1/104e933c3cc2cbbe8c8c7a7a.jpg"},{"id":105570372,"identity":"7da702ea-9790-4905-bf62-04db8712a09a","added_by":"auto","created_at":"2026-03-27 13:16:31","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1793749,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9123924/v1/bbb6bd87-34a0-4c1b-a6dc-adaba5ea745c.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"\u003cp\u003eSusceptibility Assessment of Debris Flow Hazards from the Perspective of Watershed Units Grounded in the Random Forest (RF) Model — A Case Study in the Upper Reaches of the Fujiang River\u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eDebris flows are sudden and destructive geological hazards in mountainous regions. Aside from that, their susceptibility assessment has become a core component of disaster risk prevention and control as well as land-use planning \u003csup\u003e[1,2]\u003c/sup\u003e. As integrated carriers of hydrological and geomorphological processes, watershed units have emerged as the mainstream spatial scale for susceptibility modeling on account of their ability to accurately capture the complex interactions between topography, geology, and hydrology \u003csup\u003e[3,4]\u003c/sup\u003e. Notwithstanding the extensive optimization of the assessment framework and integrated multi-source data by global scholars, as well as the significant role this area plays as an ecological barrier, specialized studies have rarely reported to probe deep into key tributaries in the upper reaches of the Yangtze River, especially the upper reaches of the Fu River.\u003c/p\u003e \u003cp\u003eResearch in this field has evolved from traditional statistical modeling methods, such as Frequency Ratio (FR) and Weight of Evidence (WoE) \u003csup\u003e[5,6]\u003c/sup\u003e, to advanced machine learning and multi-process coupling techniques. Models like Random Forest (RF) and Decision Tree (DT) are highly regarded for their exceptional nonlinear fitting capabilities and stability in handling multi-factor interactions \u003csup\u003e[7,8]\u003c/sup\u003e. Recent studies published in Applied Sciences have further advanced this field by exploring innovative negative sample acquisition strategies for RF-based debris flow susceptibility mapping, demonstrating that approaches such as isolation forest (IF) methods are particularly well-adapted to watershed unit datasets\u003csup\u003e[46,47]\u003c/sup\u003e. Comparative analyses of machine learning algorithms in alpine-valley regions have also confirmed that RF outperforms support vector machine (SVM) models in terms of accuracy, with relative elevation difference emerging as the most prominent evaluation factor. In China, relevant studies have immensely advanced indicator localization and model adaptability. Scholars have established multi-dimensional systems encompassing \"topography-geology-hydrology-human activities\", tailored to the characteristics of high-altitude regions \u003csup\u003e[2,9]\u003c/sup\u003e. Recent innovations include integrating Sentinel-1 radar data to overcome data acquisition challenges in cloudy mountainous areas \u003csup\u003e[9]\u003c/sup\u003e and optimizing watershed units through hierarchical river network features \u003csup\u003e[10]\u003c/sup\u003e, jointly confirming the effectiveness of machine learning in complex terrains.\u003c/p\u003e \u003cp\u003eDespite the aforementioned breakthroughs, several critical issues persist: (1) Inconsistent watershed delineation standards hinder cross-regional comparisons; (2) A lack of specialized studies on the upper reaches of the Fujiang River has left its complex disaster-driving mechanisms poorly understood; (3) Existing models rarely throw light upon the synergistic effects among environmental driving factors in a quantitative manner. Located in the tectonically active transition zone between the Qinghai-Tibet Plateau and the Sichuan Basin, the upper reaches of the Fujiang River face frequent disasters and intensifying human activities \u003csup\u003e[12]\u003c/sup\u003e. In an effort to mitigate these research gaps, this present study not only adopts watershed units as the fundamental evaluation units, but also employs the RF model to assess debris flow susceptibility. By integrating 11 multi-source factors, this research is predominantly intended to identify key regional drivers and provide scientific support for disaster prevention and mitigation as well as ecological protection in the upper Yangtze River basin.\u003c/p\u003e"},{"header":"2. Overview of the Study Area","content":"\u003cp\u003eLocated in northwestern Sichuan Province, China, the upper reaches of the Fujiang River serve as the source and core flow region of the Fujiang River, which is a secondary tributary of the Yangtze River. This area is not only geographically unique but also historically significant as a vital corridor connecting central China with the southwest, boasting rich natural landscapes and cultural heritage. As demonstrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, the upper reaches of the Fujiang River (103\u0026deg;45\u0026prime;\u0026ndash;105\u0026deg;15\u0026prime; E, 31\u0026deg;40\u0026prime;\u0026ndash;33\u0026deg;10\u0026prime; N) refer to the watershed region from the source of the Fujiang River to the control section of the Wudu Town hydrological station in Jiangyou City. The source is located at Xuebaoding, the main peak of the Minshan Mountains, with the main stem flowing southward and major tributaries including the Huoxi River, Huya River, and Pingtong River, covering a total area of approximately 19,000 km\u0026sup2;. The Fujiang River is a primary tributary on the right bank of the upper Yangtze River, originating from Sanchazi at the northern foot of Xuebaoding, the main peak of the Minshan Mountains, in Songpan County, Aba Tibetan and Qiang Autonomous Prefecture, Sichuan Province. The regional topography exhibits pronounced step-like and transitional characteristics, transitioning from plateaus and high mountains in the northwest to basins in the southeast, which displays extreme elevation differences. The highest point is Xuebaoding in Songpan County, at 5,588 m above sea level, while the lowest point is in southern Jiangyou City, at approximately 500\u0026ndash;600 m above sea level \u003csup\u003e[11\u0026ndash;13]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eMajor fault zones in the upper reaches of the Fujiang River include the Longmenshan Fault Zone, Huya Fault Zone, and Minjiang Fault Zone \u003csup\u003e[14]\u003c/sup\u003e. Through data provided by the Sichuan Provincial Department of Natural Resources, remote sensing interpretation, and field surveys, a total of 192 debris flow hazard sites have been identified in the upper reaches of the Fujiang River, predominantly distributed across 34 townships in five counties within the two cities of Mianyang and Aba Tibetan and Qiang Autonomous Prefecture \u003csup\u003e[14,15]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"3. Data and Research Methods","content":"\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Data Sources\u003c/h2\u003e \u003cp\u003eThe primary data include debris flow hazard sites, elevation (DEM), and other datasets, with specific sources detailed in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. DEM remote sensing imagery of the Fujiang River basin was processed through geometric correction and delineated using watershed boundaries to define the upper reaches of the Fujiang River.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eData Sources\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eData Type\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eParameters\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSource\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDebris flow hazard sites\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e30 m resolution\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSichuan Provincial Department of Natural Resources\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNDVI data\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1 km resolution\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMOD13A1.006 Terra Vegetation Indices 16-Day Global 500m dataset on GEE\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLand use/cover data (2017\u0026ndash;2024)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1 km resolution\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWuhan University Land Cover Dataset on CLCD\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMeteorological data\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1 km resolution\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNational Earth System Science Data Center (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.geodata.cn/\u003c/span\u003e\u003cspan address=\"http://www.geodata.cn/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDEM data\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e30 m resolution\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eASTER GDEM30M\u003c/p\u003e \u003cp\u003eDigital Elevation Data\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSoil type data\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e1: 1,000,000 scale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eResource and Environment Science and Data Center, CAS (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.resdc.cn/\u003c/span\u003e\u003cspan address=\"https://www.resdc.cn/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVegetation type data\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHydrological data\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Selection of Evaluation Units and Indicators\u003c/h2\u003e \u003cdiv id=\"Sec6\" class=\"Section3\"\u003e \u003ch2\u003e3.2.1 Selection of Evaluation Units\u003c/h2\u003e \u003cp\u003eAs different units yield varying results, selecting appropriate evaluation units is crucial for debris flow susceptibility assessment. For the time being, grid units, watershed units, and administrative units are extensively utilized for susceptibility assessments of geological hazards like debris flows and landslides. Notwithstanding their conspicuous advantages like rapid subdivision and computation on the basis of GIS, grid units fail to link with the geological environment of debris flow regions, resulting in evaluations lacking practical attributes \u003csup\u003e[16]\u003c/sup\u003e. In contrast, watershed units effectively integrate the geological environment of debris flow regions, which not only takes into consideration topographic and material source conditions for debris flow formation and development, but also facilitates the extraction of susceptibility assessment indicators and factor values to yield reliable results. Administrative units conduct evaluations rooted in entire administrative regions, which is advantageous for government departments at all levels to formulate disaster prevention and mitigation policies, and thereby conduct disaster relief efforts \u003csup\u003e[37]\u003c/sup\u003e. Nonetheless, their results not only fail to correlate with actual topographic and geomorphic features, but also cannot truly reflect the relationship between debris flows and assessment factors. To this end, this study selects watershed units as the evaluation units for debris flow susceptibility in the upper reaches of the Fujiang River. Using ArcGIS hydrological analysis functions \u003csup\u003e[17]\u003c/sup\u003e, we compared river networks extracted under different thresholds (1000, 2000, 3000) with the actual river network distribution in the study area. We arrived at a conclusion that a threshold of 1000 yielded a river network and surrounding watersheds largely consistent with the actual distribution. For this reason, a threshold of 1000 was used to extract the river network, ultimately dividing the upper reaches of the Fujiang River into 685 watershed units with an average area of 18 km\u0026sup2;. Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e depicts the distribution of debris flow watersheds in the upper reaches of the Fujiang River.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section3\"\u003e \u003ch2\u003e3.2.2 Selection of Evaluation Indicators\u003c/h2\u003e \u003cp\u003eThe development of debris flows requires at least three conditions: ① steep slopes, ② a large amount of loose materials, and ③ an adequate water source. Steep slopes are one of the crucial conditions for debris flow development, as they provide sufficient potential energy, enabling the rapid flow of debris flows under the influence of gravity. Debris flows are typically formed by the mixture of water and mud in slopes or gullies containing a large amount of loose materials (such as soil, sand, rock debris, etc.). As substantial rainfall, snowmelt, mountain meltwater, or groundwater outbursts can provide sufficient water sources, water functions as a significant factor in debris flow development, thereby allowing loose materials to mix with water and form debris flows. As a result, the selection of evaluation factors for debris flow susceptibility should take into account the scientificity, practicality, and correctness of the factors. Grounded in the relevant explanations in Chap.\u0026nbsp;6 \u003cem\u003eSurvey Content of Disaster-Pregnant Geological Conditions\u003c/em\u003e of the Technical Requirements for Geological Survey (1:50,000) by the China Geological Survey Bureau \u003csup\u003e[18]\u003c/sup\u003e, as well as relevant research findings on the causes of debris flow disasters in China, and considering the actual situation of the study area, factors such as topography and geomorphology, meteorology and hydrology, soil type, and vegetation coverage were selected as evaluation factors for debris flow susceptibility in the upper reaches of the Fujiang River. Taking into account the availability of each evaluation factor and combining with the debris flow basin units in the study area, 11 evaluation factors were ultimately selected to construct an evaluation system for debris flow susceptibility in the upper reaches of the Fujiang River for susceptibility assessment.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section3\"\u003e \u003ch2\u003e3.2.3 Quantitative Grading Standards for Evaluation Factors\u003c/h2\u003e \u003cp\u003eAs a result of the different evaluation criteria for each indicator, involving qualitative and quantitative, absolute and relative quantities, and differences in units of measurement, as well as considering the consistency and rationality of subsequent overlay analysis of layers, reference was made to the Technical Requirements for Geological Survey (1:50,000) by the China Geological Survey Bureau and relevant research results by Zhu Xiaolong et al. \u003csup\u003e[19]\u003c/sup\u003e. Each indicator's corresponding disaster risk level was classified, ranging from extremely low to extremely high, unified as Grades Ⅰ-Ⅴ, as revealed in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eQuantitative Grading Table for Factor Indicators\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eEvaluation Factor Indicator\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"5\" nameend=\"c6\" namest=\"c2\"\u003e \u003cp\u003eEvaluation Indicator Grading (Risk Level)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eExtremely Low (I)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLow (II)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMedium (III)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHigh (IV)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eExtremely High (V)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBasin Area/km\u0026sup2;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.006\u0026ndash;7.387\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7.38\u0026ndash;17.63\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e17.63\u0026ndash;30.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e30.45\u0026ndash;51.99\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e51.99\u0026ndash;134.34\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage Elevation/m\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e493\u0026ndash;1200.07\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1200.07\u0026ndash;1816.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1816.88\u0026ndash;2523.955\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2523.95\u0026ndash;3291.20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e3291.20\u0026ndash;4329.25\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBasin Elevation Difference/m\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u0026ndash;689\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e689\u0026ndash;1253\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1253\u0026ndash;1693\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1693\u0026ndash;2239\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2239\u0026ndash;3388\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage Slope/(\u0026deg;)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.53\u0026ndash;10.27\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e10.27\u0026ndash;21.43\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e21.43\u0026ndash;26.39\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e26.39\u0026ndash;30.11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e30.11\u0026ndash;45.70\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage Plan Curvature\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-0.79\u0026ndash;།0.43\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-0.43\u0026ndash;0.09\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-0.09\u0026ndash;0.006\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.006\u0026ndash;0.029\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.029\u0026ndash;0.413\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage Profile Curvature\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-0.20\u0026ndash;0.002\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.002\u0026ndash;0.015\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.015\u0026ndash;0.056\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.056\u0026ndash;0.252\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.252\u0026ndash;1.523\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMelton Ratio\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u0026ndash;0.214\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.124\u0026ndash;0.567\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.567\u0026ndash;0.846\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.846\u0026ndash;1.368\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.368\u0026ndash;3.178\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNDVI Index\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.017\u0026ndash;0.226\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.226\u0026ndash;0.332\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.332\u0026ndash;0.429\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.429\u0026ndash;0.528\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.528\u0026ndash;0.681\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSoil Type\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePaddy soil, swamp soil, stony soil, stony loam\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eYellow soil, dark brown soil, tidal brown soil, brown coniferous forest soil\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLimestone soil, yellow lime soil, black lime soil\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eBrown soil, brown loam soil\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003ePurple soil, coarse-bone soil\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage Annual Precipitation/mm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e679.70\u0026ndash;748.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e748.66\u0026ndash;778.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e778.93\u0026ndash;809.21\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e809.21\u0026ndash;849.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e849.58\u0026ndash;894.15\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage Annual Temperature/℃\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-4.61\u0026ndash;2.13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2.13\u0026ndash;6.18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e6.18\u0026ndash;9.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e9.64\u0026ndash;13.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e13.02\u0026ndash;16.90\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"4 Results","content":"\u003cp\u003eThe Random Forest (RF) model is an ensemble learning method employed to solve classification and regression problems \u003csup\u003e[20\u0026ndash;23]\u003c/sup\u003e. It consists of multiple decision trees, each constructed rooted in a randomly selected subset of features and randomly selected training samples.\u003c/p\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Analysis of the Area Proportion and Spatial Distribution of Risk Levels for Evaluation Indicators\u003c/h2\u003e \u003cp\u003eAs illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, the spatial distribution of risk levels for each indicator was classified and visualized in accordance with the aforementioned grading standards. By utilizing spatial statistical tools, the area and proportion of regions with different risk levels for each evaluation factor were summarized, with the results revealed in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. As evidently demonstrated by the results, debris flow disasters in the upper reaches of the Fujiang River are primarily influenced by basin area and average annual precipitation, which are the largest controlling factors. Secondary influencing factors mainly include NDVI, average elevation, and average elevation difference, while average annual temperature and soil type have the least impact. Debris flow disaster points are mainly concentrated in areas with slopes ranging from 26.39\u0026deg; to 30.11\u0026deg; and elevations between 1,200.07 and 1,816.88 m, which are medium-elevation regions. A large basin elevation difference provides suitable topographic conditions for debris flow disasters. Topographic profile within the basin exhibits strong incision, with towering peaks and deep gullies, resulting in strong hydrodynamic forces that easily carry material sources, thereby triggering debris flow disasters. The size of the basin area and average annual precipitation directly affect the amount of rainfall accumulated within the basin. The more rainfall accumulates in a debris flow basin, the more likely it is to carry enormous slope debris and sediment, further increasing the likelihood of debris flow disasters.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eArea and Proportion of Risk Levels for Each Evaluation Factor\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"11\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"3\"\u003e \u003cp\u003eEvaluation Factor Indicator\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"10\" nameend=\"c11\" namest=\"c2\"\u003e \u003cp\u003eEvaluation Indicator Grading (Risk Level)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eExtremely Low (I)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003eLow (II)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c7\" namest=\"c6\"\u003e \u003cp\u003eMedium (III)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e \u003cp\u003eHigh (IV)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c11\" namest=\"c10\"\u003e \u003cp\u003eExtremely High (V)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eArea /KM\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eProportion %\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eArea /KM\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eProportion %\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eArea /KM\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eProportion %\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eArea /KM\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eProportion %\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c10\"\u003e \u003cp\u003eArea /KM\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c11\"\u003e \u003cp\u003eProportion %\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBasin Area/km\u0026sup2;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e346.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3049.73\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e25.07\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3864.37\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e31.77\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e2859.53\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e23.51\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e2044.38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e16.80\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage Elevation/m\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1401.42\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e11.52\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4183.14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e34.39\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e2565.78\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e21.09\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e2387.11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e19.62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e1627.27\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e13.38\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBasin Elevation Difference/m\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e417.59\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3.43\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2063.73\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e16.96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e4225.97\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e34.74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e3873.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e31.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e1583.78\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e13.02\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage Slope/(\u0026deg;)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e333.33\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2.74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1499.31\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e12.33\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3399.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e27.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e4021.61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e33.06\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e2910.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e23.92\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage Plan Curvature\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1190.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e9.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4292.07\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e35.28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e4591.36\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e37.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1980.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e16.28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e110.43\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.90\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage Profile Curvature\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1460.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e12.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e7491.22\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e61.57\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3187.87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e26.21\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e24.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.01\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMelton Ratio\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e365.16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2830.47\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e23.27\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3246.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e26.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e4342.54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e35.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e1379.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e11.34\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNDVI Index\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3680.38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e30.25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2958.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e24.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e2041.04\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e16.78\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e2201.28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e18.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e1283.04\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e10.55\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSoil Type\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e7451.24\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e61.25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e19.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e104.49\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e3097.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e25.46\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e1492.34\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e12.27\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage Annual Precipitation/mm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1803.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e14.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2623.54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e21.57\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e2908.72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e23.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e3315.46\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e27.25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e1513.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e12.45\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage Annual Temperature/℃\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1301.25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e10.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2517.57\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e20.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e2320.43\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e19.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e2672.62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e21.97\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e3352.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e27.55\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Analysis of Factor Contribution Rates\u003c/h2\u003e \u003cp\u003eIn this study, data from 192 debris flow disasters in the upper reaches of the Fujiang River were used as research samples. Using ArcGIS10.7, DEM data for the upper reaches of the Fujiang River basin was divided into 685 basin units. A marker column attribute was added to each basin unit, with basin units where debris flows had occurred marked as \"1\" and those where debris flows had never occurred marked as \"0\". Consequently, the problem of debris flow susceptibility evaluation was transformed into a binary classification problem \u003csup\u003e[24]\u003c/sup\u003e. As the results illustrate, there were 33 basin units where debris flows had occurred and 628 basin units where debris flows had never occurred in the upper reaches of the Fujiang River. Initially, through literature review, summarizing previous experiences, and continuous testing, the basic dataset was divided into an 80% training set and a 20% test set in an 8:2 ratio for training and testing the debris flow susceptibility evaluation model in the upper reaches of the Fujiang River \u003csup\u003e[26\u0026ndash;27]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThe Random Forest model was constructed by adopting the Scikit-learn framework built into the Python language. After hyperparameter tuning, the number of decision trees in the Random Forest model was set to 100, the splitting metric was \"gini\", the maximum tree depth was \"None\", the maximum number of features was \"auto\", and the number of parallel jobs was \"None\".\u003c/p\u003e \u003cp\u003eThe factor contribution rates of the evaluation indicators included in the Random Forest model after hyperparameter tuning are depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. As suggested by an all-round observation, the factor contribution rates of basin area and average annual precipitation are the same, both at 0.109, making them the factors with the highest contribution rates among the evaluation indicators for inducing debris flow outbreaks. This also confirms that basin area and average annual precipitation have significant impacts on debris flow outbreaks in the basin. Soil type has the lowest contribution rate, at 0.032, in inducing debris flow outbreaks.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe contribution rates of the two factors, watershed area and annual average precipitation, display the same and relatively high values. This phenomenon suggests that watershed area and annual average precipitation may have significant impacts on the analysis of debris flow susceptibility. Aside from that, larger watershed areas and higher annual average precipitation may be associated with the occurrence of debris flows to a certain extent. NDVI is an indicator for measuring the condition of surface vegetation. A higher NDVI value indicates more luxuriant vegetation cover. As vegetation can stabilize soil and reduce the likelihood of soil erosion and debris flows, higher NDVI values may be correlated with the mitigation or prevention of debris flows. Both the average elevation and the elevation difference within the watershed are terrain-relevant. Steep terrain and large elevation changes may give rise to soil erosion and the occurrence of debris flows. Larger profile curvature, plan curvature, and slope potentially heighten the risk of soil erosion and debris flows. A higher Melton ratio may suggest a certain correlation between steep terrain and debris flow susceptibility. The contribution rates of the two factors, annual average temperature and soil type, are relatively low, suggesting that they may not have obvious direct impacts on the occurrence of debris flows. In summary, on the basis of the given factor contribution rates, watershed area, annual average precipitation, NDVI, average elevation, watershed elevation difference, average profile curvature, average plan curvature, average slope, and Melton ratio may be important factors influencing debris flow susceptibility. Simultaneously, other factors such as annual average temperature and soil type, with relatively low contribution rates, may conduct a trivial role in the analysis of debris flow susceptibility.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e4.3 Debris Flow Susceptibility Analysis\u003c/h2\u003e \u003cp\u003eAfterwards, a susceptibility assessment of debris flows in the upper reaches of the Fujiang River was conducted grounded in the Random Forest model. After the model ran, susceptibility index values were obtained for 685 watershed units in the upper reaches of the Fujiang River. By importing these susceptibility values into ArcGIS 10.7 software, the debris flow susceptibility values in the upper reaches of the Fujiang River were classified into five different levels according to the susceptibility index of the watershed units, namely, extremely low susceptibility area, low susceptibility area, moderate susceptibility area, high susceptibility area, and extremely high susceptibility area. A susceptibility assessment map of debris flow disasters in the upper reaches of the Fujiang River was subsequently generated (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAs evidenced in the figure, we can draw several conclusions below: (1) The extremely high and high susceptibility areas for debris flows in the upper reaches of the Fujiang River are mainly distributed in the northwestern part from Mao County to Beichuan, the eastern part of Songpan County, and the southeastern part of Pingwu County. They are mainly concentrated in extremely unstable valley areas, mostly distributed along the slopes on both sides of the main stream of the Fujiang River (from Songpan to Pingwu section), Huoxi River, Baicao River, and other rivers. (2) The moderate susceptibility areas for debris flows in the upper reaches of the Fujiang River are mainly distributed around the high susceptibility areas. This region is mainly mountainous. Since other geological disasters are also distributed in this area, they further provide material sources for debris flow disasters. Under specific conditions such as heavy rainfall scouring, debris flow disasters are likely to occur. (3) The extremely low and low susceptibility areas for debris flows in the upper reaches of the Fujiang River are mainly distributed in the central, northern, and eastern parts of the upper reaches of the Fujiang River, as well as in most areas of Jiangyou City. The terrain in this region is relatively gentle, with bedrock exposed on the surface, lacking the conditions for debris flow development.\u003c/p\u003e \u003cp\u003eWith regard to the number of watersheds (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e), considering both the number and area of watersheds, the susceptibility zoning in the upper reaches of the Fujiang River is highly consistent. The extremely low susceptibility area occupies an absolutely dominant position, with both the number of watersheds (635) and the area (11,282.63 km\u0026sup2;) accounting for 92.75% of the total. The extremely high susceptibility area comes next, with both indicators (31 in number and 579.04 km\u0026sup2; in area) accounting for 4.76%. The combined proportion of the other three susceptibility areas (moderate, low, and high) is less than 2.5%.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eStatistics on Debris Flow Susceptibility in the Watersheds of the Upper Reaches of the Fujiang River\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRisk Zone\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNumber of Watersheds / Units\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWatershed Area / km\u0026sup2;\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eArea Proportion / %\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eⅠ (Extremely Low)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e635\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e11282.63\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e92.75\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eⅡ (Low)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e74.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.61\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eⅢ (Moderate)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e170.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.40\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eⅣ (High)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e58.10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.48\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eⅤ (Extremely High)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e31\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e579.04\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4.76\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTotal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e685\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e12164.72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAs suggested by comprehensive statistics on the 144 debris flow watersheds in the upper reaches of the Fujiang River and their susceptibility level zoning (Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e)\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eStatistics on Debris Flow Susceptibility in the Upper Reaches of the Fujiang River\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCounty-level Administrative/prone Level\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eExtremely Low Susceptibility\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLow Susceptibility\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eModerate Susceptibility Area\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHigh Susceptibility Area\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eExtremely High Susceptibility Area\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJiangyou City\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePingwu County\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e41\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBeichuan County\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e21\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMaoxian County\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSongpan County\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e12\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e4.4 Model Accuracy Evaluation\u003c/h2\u003e \u003cp\u003eThe Receiver Operating Characteristic (ROC) curve (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e) is a tool employed to assess the performance of classification models. In this study, with the upper reaches of the Fujiang River as the target area, an innovative approach was adopted by using watershed units as the basic evaluation units. Rooted in a catchment threshold of 1000, the study area was divided into 685 watersheds, among which extremely low/low susceptibility areas accounted for 93.36%, and extremely high/high susceptibility areas accounted for 5.24%. By integrating 11 disaster-causing factors from four major categories, namely topography and geomorphology, meteorology and hydrology, vegetation coverage, and geology and soil, a Random Forest (RF) model was constructed on the Python platform to evaluate debris flow susceptibility \u003csup\u003e[33\u0026ndash;35]\u003c/sup\u003e. Afterwards, the ROC curve and the Area Under the Curve (AUC) were employed to evaluate the model's performance. The ROC curve measures classification effectiveness by plotting the relationship between the True Positive Rate (TPR) and the False Positive Rate (FPR) at different thresholds, with a curve closer to the top-left corner indicating better performance. Verified by 144 historical debris flow disaster points, the model achieved an AUC value of 0.944 and an Accuracy (ACC) value of 0.82. On top of that, 56.94% of the disaster points were concentrated in extremely high susceptibility areas, and 27.78% were located in high susceptibility areas, thereby confirming the reliability and favorable applicability of the evaluation results, which can provide a scientific basis for precise prevention and control of debris flows in mountainous regions of Southwest China.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"5 Discussion","content":"\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e5.1 Optimization and Rationality Comparison of Evaluation Units\u003c/h2\u003e \u003cp\u003eThe rational selection of evaluation units is crucial for the elevation of assessment accuracy. In general, traditional grid units are not only disadvantageous for the integrity of the geological environment, but are also challenging to reflect the continuous process of debris flow, which involves \"catchment-sediment production-transportation\". As a result of their artificial boundaries, administrative units fail to align with the natural continuity of the terrain, often leading to distorted results \u003csup\u003e[12,13]\u003c/sup\u003e. In this study, river networks were extracted using a threshold of 1000, and 685 watershed units (with an average area of 18 km\u0026sup2;) were delineated through the hydrological analysis function of ArcGIS. This approach not only preserves the coupling relationships among terrain, hydrology, and material sources but also matches the patterns of disaster development, effectively overcoming the limitations of traditional units. In contrast to the study by Wang Shige et al. \u003csup\u003e[8]\u003c/sup\u003e in the Dadu River basin, this study incorporates additional factors such as average planar curvature, Melton ratio, and annual average temperature, providing a more comprehensive coverage of the material sources, hydrodynamic forces, and topographic base conditions for debris flow formation, aligning with the complex mountainous characteristics of the upper reaches of the Fujiang River. In comparison with the study by Zhu Xin et al. \u003csup\u003e[28]\u003c/sup\u003e, which used a catchment threshold of 3000 to divide 2948 fragmented units (with an average area of 4 km\u0026sup2;), the units in this study are of moderate size, reducing data processing redundancy costs while avoiding the dissection of natural watershed coupling relationships. The susceptibility zoning is clear (extremely low/low susceptibility areas account for 93.36%, and extremely high/high susceptibility areas account for 5.24%), facilitating the precise identification of core risk areas and the formulation of prevention and control measures.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e5.2 Advantages of Model Selection and Identification of Core Disaster-Causing Factors\u003c/h2\u003e \u003cp\u003eThe choice of model and factor combination can exert direct influences on the objectivity of the evaluation. Traditional methods such as the Analytic Hierarchy Process (AHP) and the Frequency Ratio (FR) method rely on subjective weighting or single statistical rules, thereby rendering it challenging to quantify the nonlinear interactions among multiple factors \u003csup\u003e[29,30]\u003c/sup\u003e. Some machine learning models suffer from inadequate factor screening specificity; for example, Wilford et al.'s model focused solely on topographic factors \u003csup\u003e[43]\u003c/sup\u003e, and Rowbotham's logistic regression model lacked vegetation and soil parameters \u003csup\u003e[44]\u003c/sup\u003e. The RF model selected in this study possesses strong feature screening and nonlinear fitting capabilities, enabling it to automatically identify core driving factors and avoid human interference. This advantage aligns with recent findings in Applied Sciences, where machine learning-based feature importance analysis has been increasingly integrated with hyperparameter optimization techniques such as Bayesian Optimization to enhance both predictive performance and model explainability in geohazard assessments\u003csup\u003e[48]\u003c/sup\u003e. The integration of feature importance analysis with susceptibility modeling, as demonstrated in this study, provides clearer insight into the model's decision-making process and enables more reliable engineering interpretation\u003csup\u003e[48]\u003c/sup\u003e.As the above results demonstrate, watershed area and annual average precipitation are the primary driving factors (both with a contribution rate of 0.109), aligning with the regional characteristics of \"high mountains, deep valleys, and concentrated precipitation\". Larger watershed areas provide space for the accumulation of loose materials and water collection, while concentrated precipitation supplies the hydrodynamic force for debris flow initiation. The Normalized Difference Vegetation Index (NDVI) and average elevation are secondary key factors (both with a contribution rate exceeding 0.104), reflecting the regulation of soil stability by vegetation and the disaster-causing characteristics of steep terrain and abundant material sources in high-altitude areas. Soil type has the lowest contribution rate (0.032), which can be attributable to the fact that the soil in the study area is predominantly purple soil and skeletal soil with weak erosion resistance, and the impact of soil type differences on disasters is relatively uniform. This result accurately reveals the region-specific driving mechanisms, successfully addressing gaps in previous studies \u003csup\u003e[15,45]\u003c/sup\u003e. The contribution patterns observed in our study\u0026mdash;where topographic conditions serve as prerequisites while precipitation and vegetation factors modulate disaster occurrence\u0026mdash;are consistent with recent machine learning-based susceptibility assessments in alpine-valley regions documented in Applied Sciences and related publications. These studies have similarly emphasized that topographic conditions are fundamental prerequisites for debris flow development, while factors such as precipitation, vegetation cover, and anthropogenic influence play critical roles in determining the spatial distribution of high-susceptibility zones .\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e5.3 Spatial Pattern Characteristics of Susceptibility and Causal Analysis\u003c/h2\u003e \u003cp\u003eThe spatial pattern of debris flow susceptibility in the upper reaches of the Fujiang River is tightly correlated with the geological environment and human activities. Extremely high susceptibility areas (579.04 km\u0026sup2;) and high susceptibility areas (58.1 km\u0026sup2;) are concentrated in unstable valley regions in the northwestern part of Maoxian-Beichuan, the eastern part of Songpan County, and the southeastern part of Pingwu County, exhibiting a zonal distribution along the slopes on both sides of the main stream and tributaries of the Fujiang River. Characterized by active geological structures and fragmented rock masses, this region lies within the influence range of the Longmenshan and Huya fault zones \u003csup\u003e[10]\u003c/sup\u003e. Coupled with significant topographic relief (up to 3388 m), average slopes ranging from 26.39\u0026deg; to 45.70\u0026deg;, and annual average precipitation between 849.58 and 894.15 mm, multiple factors contribute to the high incidence of disasters. Moderate susceptibility areas are distributed in the high mountains surrounding the high susceptibility areas, where material sources and hydrodynamic conditions are weaker, and disasters are likely to occur only under extreme rainfall conditions. Extremely low/low susceptibility areas account for 93.36% of the total area and are concentrated in the gentle regions of Jiangyou City and the central and northern parts of the basin, where the terrain is flat and bedrock is exposed, lacking the conditions for debris flow development. This pattern is consistent with the geomorphic characteristics and fault activity response laws revealed by Chen Hao et al. \u003csup\u003e[15]\u003c/sup\u003e, further confirming the rationality of the evaluation results.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e5.4 Future Research Directions and Prospects\u003c/h2\u003e \u003cp\u003eAltogether, future research can be deepened in the following aspects. First and foremost, it is advisable to integrate high-resolution Sentinel-2 imagery and unmanned aerial vehicle (UAV) survey data to unify factor resolution and enhance evaluation accuracy in local small watersheds. Apart from that, it is preferable to introduce dynamic factors such as seismic ground motion parameters and land use changes, and construct a dynamic evaluation system combined with time-series NDVI to quantify the coupled effects of human activities and natural changes. In line with emerging trends in Applied Sciences special issues on \"Intelligent Technologies in Geotechnical Engineering and Geological Hazards\" and \"Applications of Artificial Intelligence in Geotechnics and Engineering Geology,\" future work should also explore the integration of physics-informed AI-driven methods that combine machine learning with hydro-mechanical coupling models[48,50]. Such approaches can better capture the complex fracture\u0026ndash;seepage interactions and permeability evolution processes that govern debris flow initiation under rainfall and seismic triggers\u003csup\u003e[49]\u003c/sup\u003e .Moreover, it holds pivotal significance to compare the performance of the RF model with other models, such as XGBoost and Support Vector Machine (SVM) to screen the optimal model for mountainous basins. Last but not least, grounded in the evaluation results, delineate differentiated prevention and control zones and propose comprehensive prevention and control schemes that combine engineering governance with non-engineering measures, considering terrain and socioeconomic characteristics. This will provide more practical technical support for the construction of ecological barriers and land space planning in the upper reaches of the Yangtze River.\u003c/p\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eScholar Liu Weiwei is responsible for writing the main manuscript, Professor Lin Xiaoxian is responsible for reviewing and revising the manuscript, and Zhu Xin is responsible for providing data.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe data used in this study are derived from the following public sources:1. In addition to the mudslide disaster sites in the upper reaches of the Weijiang River, which are obtained by the Sichuan Provincial Department of Natural Resources, the applicability of these data is limited. They involve coordinates and are authorized by this study, so the disaster point data is not disclosed.2.Data source of China's administrative divisions:https://cloudcenter.tianditu.gov.cn/administrativeDivision(This data cannot be linked due to the restrictions of the external network, but the author confirms that the data set used during the current research is obtained and used at reasonable request.)3.DEM data set source:https://www.gebco.net/data\\_and\\_products/gridded\\_bathymetry\\_data/4.Source of river water system data:https://zenodo.org/records/138419105.Source of temperature data:https://www.ncei.noaa.gov/data/global-summary-of-the-day/archive/6.Precipitation data:https://www.ncei.noaa.gov/data/global-summary-of-the-day/archive/7.NDVI data set:https://doi.org/10.11888/Terre.tpdc.300328.Further details can be found in the original publications.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eFell R. Landslide risk assessment and acceptable risk[J]. Canadian Geotechnical Journal, 1994, 31(2): 261-272.\u003c/li\u003e\n\u003cli\u003eCui P, Chen X Q, Wang S G. Progress and prospects of research on debris flow disaster prevention and control in China[J]. Acta Geographica Sinica, 2020, 75(1): 3-18.\u003c/li\u003e\n\u003cli\u003eMontgomery D R, Dietrich W E, Thompson S E. A runoff generation threshold for landsliding in the Oregon Coast Range[J]. Water Resources Research, 1995, 31(1): 115-127.\u003c/li\u003e\n\u003cli\u003eYang Q K, Wang L, Li W L. Impact of watershed unit division on the accuracy of debris flow susceptibility evaluation[J]. Geographical Research, 2021, 40(5): 1365-1378.\u003c/li\u003e\n\u003cli\u003eKanungo D P, Arora M K, Gupta R P, et al. Landslide risk assessment using concepts of danger pixels and fuzzy set theory in Darjeeling Himalayas[J]. Landslides, 2008, 5(4): 407-416.\u003c/li\u003e\n\u003cli\u003ePourghasemi H R, Pradhan B, Gokceoglu C. Application of fuzzy logic and analytical hierarchy process (AHP) to landslide susceptibility mapping at Haraz watershed Iran[J]. Natural Hazards, 2012, 63(3): 965-996.\u003c/li\u003e\n\u003cli\u003eKhosravi K, Pham B T, Chapi K, et al. A comparative assessment of decision trees algorithms for flash flood susceptibility modeling at Haraz watershed, northern Iran[J]. Science of the Total Environment, 2018, 627: 744-755.\u003c/li\u003e\n\u003cli\u003eWu C Y, Lin S Y. Event-based landslide susceptibility models in Shihmen watershed, Taiwan: accounting for the characteristics of rainfall events[J]. Environmental Monitoring and Assessment, 2022, 194(5): 405.\u003c/li\u003e\n\u003cli\u003eWang S G, Li W L, Yang Q K. Evaluation of Debris Flow Susceptibility in Southwest Mountainous Areas Based on Basin Units: A Case Study of the Dadu River Basin [J]. Acta Geologica Sinica, 2022, 96 (8): 2654-2666.\u003c/li\u003e\n\u003cli\u003eChen J, Zhang G R, Liu H H. Debris flow susceptibility evaluation in the mountainous regions of Southwest China based on Sentinel-1 and Random Forest model[J]. Chinese Journal of Rock Mechanics and Engineering, 2021, 40(S1): 3012-3020.\u003c/li\u003e\n\u003cli\u003eTian C S. Debris flow hazard assessment under future climate change in the upper reaches of the Minjiang River[D]. Chengdu: Chengdu University of Technology, 2016.\u003c/li\u003e\n\u003cli\u003eHu B, Zhang Y, Wang H R, et al. Spatial distribution characteristics of fuel and fire risk assessment in typical forest ecosystems in the upper reaches of the Yangtze River: A case study of Pingwu County, Sichuan Province[J/OL]. Journal of Catastrophology, 1-10[2025-10-26]. https://link.cnki.net/urlid/61.1097.P.20251010.1541.004.\u003c/li\u003e\n\u003cli\u003eLi M Y, Pan H Z, Chen P, et al. NDVI changes in the upper reaches of the Yangtze River and their impact on the sediment and water processes entering the Three Gorges Reservoir[J/OL]. Resources and Environment in the Yangtze Basin, 1-21[2025-10-26]. https://link.cnki.net/urlid/42.1320.X.20250912.1921.002.\u003c/li\u003e\n\u003cli\u003eLi K, Zhao J S, Lin Y L, et al. Debris flow susceptibility evaluation in Dongchuan based on RF and SVM models[J]. Journal of Yunnan University (Natural Sciences Edition), 2022, 44(1): 107-115.\u003c/li\u003e\n\u003cli\u003eSichuan Provincial Department of Natural Resources. Sichuan Provincial Geological Disaster Prevention and Control Plan (2021-2025)[R]. Chengdu: Sichuan Provincial Department of Natural Resources, 2022.\u003c/li\u003e\n\u003cli\u003eChen H, Dong T X, Li Y, et al. Geomorphic characteristics of the upper reaches of the Fujiang River basin and their response to fault activity[J]. Journal of Mountain Science, 2020, 38(4): 542-551.\u003c/li\u003e\n\u003cli\u003eLi K, Zhao J S, Lin Y L, et al. Debris flow susceptibility evaluation based on different slope unit division methods and BP neural network[J]. Bulletin of Surveying and Mapping, 2022, 1(8): 68-74.\u003c/li\u003e\n\u003cli\u003ePan R. Debris flow susceptibility evaluation in Shimian County after the \u0026quot;8\u0026bull;11\u0026quot; disaster[D]. Sichuan Agricultural University, 2022. DOI: 10.27345/d.cnki.gsnyu.2022.000530.\u003c/li\u003e\n\u003cli\u003eChina Geological Survey. Technical standards for geological surveys[S]. Technical requirements for geological disaster surveys (1:50000). China Geological Survey, Ministry of Natural Resources. DD.2019.08.\u003c/li\u003e\n\u003cli\u003eZhu X L, Ma B H, Zhang W X, Chen Y. Main disaster-causing factors and their control and impact on the distribution of geological disasters in the Zhangjiakou area of Hebei Province[J]. Chinese Journal of Geological Hazard Control, 2015, (01): 103-107.\u003c/li\u003e\n\u003cli\u003eWang L J. Morphological characteristics and empirical hazard models of debris flows[D]. Beijing: Beijing Forestry University, 2021.\u003c/li\u003e\n\u003cli\u003eLi Y M, Yang L, Wei S H, et al. Debris flow susceptibility evaluation in Nujiang Prefecture based on small watershed units[J]. Resources and Environment in the Yangtze Basin, 2019, 28(10): 2419-2428.\u003c/li\u003e\n\u003cli\u003eWang, X. P. Research on typical fault diagnosis methods for chillers based on random forest [D]. Guangzhou: South China University of Technology, 2020.\u003c/li\u003e\n\u003cli\u003eLi, J. Y., Guo, S. S. An improved random forest algorithm for load balancing in parallel rendering systems [J]. Modern Computer, 2020, 26(8): 22-25.\u003c/li\u003e\n\u003cli\u003eLin, X. X., Chen, H., Yin, M. H., et al. Theory and practice of geological disaster prevention and control in seismic disturbance zones [M]. Beijing: China Environmental Publishing Group, 2022.\u003c/li\u003e\n\u003cli\u003eLi, K., Zhao, J. S., Lin, Y. L., et al. Debris flow susceptibility assessment based on different slope unit division methods and BP neural network [J]. Bulletin of Surveying and Mapping, 2022, (8): 68-74.\u003c/li\u003e\n\u003cli\u003eChen, Y. J., Yu, B., Zhu, Y., et al. Variation characteristics of critical rainfall for debris flow after earthquakes: A case study of Xiaogangjian Gully in the Wenchuan earthquake zone [J]. Journal of Mountain Science, 2013, 31(3): 356-361.\u003c/li\u003e\n\u003cli\u003eXu, R. G., Zheng, W. M., Ba, R. J., et al. Study on the development characteristics of debris flow in the Dadu River Basin [J]. Yangtze River, 2013, 44(S1): 117-119+131.DOI: 10.16232/j.cnki.1001-4179.2013.s1.007.\u003c/li\u003e\n\u003cli\u003eZhu, X., Lin, X. X., Chen, H., et al. Distribution pattern and risk prevention and control strategies of debris flow in the upper reaches of the Fujiang River [D]. Mianyang: Mianyang Teachers\u0026apos; College, 2024.\u003c/li\u003e\n\u003cli\u003eLi, W. L., Yang, Q. K., Wang, L. Application of an improved frequency ratio method in debris flow susceptibility assessment in the Qinba Mountain area [J]. Journal of Natural Resources, 2019, 34(7): 1492-1504.\u003c/li\u003e\n\u003cli\u003eZhong, X. W. Research on rail transit flow prediction based on AFC ticket checking big data [D]. Chongqing: Chongqing University of Technology, 2021.\u003c/li\u003e\n\u003cli\u003eMosaffaie J, Salehpour Jam A, Sarfaraz F, et al. Landslide risk assessment based on susceptibility and vulnerability[J]. Environment, Development and Sustainability, 2024, 26(12): 9285-9303.\u003c/li\u003e\n\u003cli\u003ePetschko H, Brenning A, Bell R, et al. Assessing the quality of landslide susceptibility maps\u0026ndash;case study Lower Austria[J]. Natural Hazards and Earth System Sciences, 2014, 14(1): 95-118.\u003c/li\u003e\n\u003cli\u003eTien Bui D, Tsangaratos P, Nguyen V T, et al. Comparing the prediction performance of a Deep Learning Neural Network model with conventional machine learning models in landslide susceptibility assessment[J]. Catena, 2020, 188: 104426.\u003c/li\u003e\n\u003cli\u003eZhang, G. R., Chen, J., Li, W. L. Comparison of different machine learning models in debris flow susceptibility assessment in the Yalong River Basin [J]. Journal of Mountain Science, 2023, 41(2): 321-330.\u003c/li\u003e\n\u003cli\u003eMeng, X. M., Chen, G., Guo, P., et al. Research progress and prospects of landslide and debris flow disasters in the Bailong River Basin [J]. Marine Geology \u0026amp; Quaternary Geology, 2013, 33(4): 1-15.\u003c/li\u003e\n\u003cli\u003eXia, J. W., Guo, H. Z. Discussion on the distribution characteristics and main controlling factors of landslides in the upper reaches of the Yangtze River [J]. Hydrogeology \u0026amp; Engineering Geology, 1997, (1): 19-22+32.\u003c/li\u003e\n\u003cli\u003eSong, G. J., Huang, J. C., Zhang, G., et al. Research on zoning evaluation of mine geological environment in Xuchang City, Henan Province [J]. Geology and Resources, 2022, 31(2): 221-227.\u003c/li\u003e\n\u003cli\u003eFeng, J. Y. Research on geological disaster risk assessment and risk management in Luolong County, Tibet [D]. Sichuan Agricultural University, 2025. DOI: 10.27345/d.cnki.gsnyu.2025.000332.\u003c/li\u003e\n\u003cli\u003eLi, H., Zhai, X., Li, C. X., Pan, Y., Li, J. P., Yuan, W. Cloud model evaluation method for debris flow disaster susceptibility in Hebei Province: A case study of debris flow in Zhaogou Village, Xingtai [J]. Science Technology and Engineering, 2024, 24(25): 10884-10891.\u003c/li\u003e\n\u003cli\u003eWang, Q. Y. Debris flow risk assessment in Xinzhou area based on semi-supervised learning [D]. 10.27162/d.cnki.gjlin.2024.003279.\u003c/li\u003e\n\u003cli\u003eChen, J. S. Research on performance optimization of three-dimensional network-on-chip for multi-core processors [D]. Guangzhou: Guangdong University of Technology, 2022.\u003c/li\u003e\n\u003cli\u003eYin, H. L. Debris flow susceptibility assessment of the China-Nepal traffic corridor under the background of climate change [D]. Lanzhou University, 2024. DOI: 10.27204/d.cnki.glzhu.2024.000967.\u003c/li\u003e\n\u003cli\u003eWilford D , Sakals M E , Innes J L ,et al. Recognition of debris flow, debris flood and flood hazard through watershed morphometrics[J]. Landslides, 2004, 1(1): 61-66.\u003c/li\u003e\n\u003cli\u003eRowbotham D, Louis S J.The identification of debris torrent basins using morphometric measures derived within a gis[J]. Geografiska Annaler, 2005, 87(4): 527- 537.\u003c/li\u003e\n\u003cli\u003ePerera E, Jayawardana D T, Jayasinghe P,et al. Landslide vulnerability assessment based on entropy method: a case study from Kegalle district, Sri Lanka[J]. Modeling Earth Systems and Environment, 2019, 5(4): 1635-1649.\u003c/li\u003e\n\u003cli\u003eGao R Y, Wu D, Liu H L, et al. Multiple negative-sample acquisition strategies for machine learning-based debris flow susceptibility mapping considering different sample representation forms[J]. Applied Sciences, 2024, 14(20): 9240. \u003c/li\u003e\n\u003cli\u003eZhou J Y, Huang J C, Sun Z B, et al. Machine learning-based debris flow susceptibility mapping in the alpine-valley area of the Nujiang River Basin[J]. PeerJ, 2024, 12: e17352. (Cited in Applied Sciences special issue context)\u003c/li\u003e\n\u003cli\u003eKarahan H, Alkaya D. Integrating SVR optimization and machine learning-based feature importance for TBM penetration rate prediction[J]. Applied Sciences, 2026, 16(1): 355. \u003c/li\u003e\n\u003cli\u003eSun Z H, Tan Q W, Wang L Q, et al. New advances in rock fractures and landslide forecasting[J]. Applied Sciences, 2026, Special Issue. \u003c/li\u003e\n\u003cli\u003eWang Y J, Cai H P, Li Q Q, et al. Applications of artificial intelligence in geotechnics and engineering geology\u0026mdash;2nd Edition[J]. Applied Sciences, 2025, Special Issue. \u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Random Forest model (RF), Geographic Information System (GIS), susceptibility assessment, debris flow, upper reaches of the Fujiang River","lastPublishedDoi":"10.21203/rs.3.rs-9123924/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9123924/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e \u003cb\u003eAs\u003c/b\u003e a critical ecological barrier in the upper Yangtze River basin, the upper reaches of the Fujiang River face severe constraints on sustainable development as a consequence of frequent debris flow hazards. For this reason, this study utilized 685 watershed units as evaluation units in the region to accurately identify high-risk areas and core driving factors while establishing a scientific susceptibility assessment framework. Eleven hazard-inducing factors, including watershed area, average elevation, and watershed elevation difference, were integrated to construct a Random Forest (RF) model on the Python platform for debris flow susceptibility assessment and key factor diagnosis. We ultimately arrived at the following conclusions. (1) Watershed area and mean annual precipitation are the primary factors controlling debris flow development, with contribution rates of 0.109 each, followed by the Normalized Difference Vegetation Index (NDVI) and average elevation (both exceeding 0.104), while soil type has the lowest contribution rate (0.032); (2) The RF model demonstrates excellent evaluation accuracy, with an Area Under the Curve (AUC) value of 0.82 and an Accuracy (ACC) value of 0.82. The study area is classified into five susceptibility levels, namely, very low, low, moderate, high, and very high. Distributing in a zonal pattern along the slopes of the main stem of the Fujiang River and its tributaries, such as the Huoxi River and Baicao River, the very high susceptibility areas (579.04 km\u0026sup2;) and high susceptibility areas (58.1 km\u0026sup2;) are primarily concentrated in unstable valley regions in the northwestern part of Maoxian-Beichuan, the eastern part of Songpan County, and the southeastern part of Pingwu County; (4) Among the 144 debris flow hazard sites within the watersheds, 56.94% (82 sites) are located in very high susceptibility areas, and 27.78% (40 sites) are in high susceptibility areas, validating the reliability of the assessment results. This study innovatively employs watershed units as evaluation units, thereby overcoming the limitations of traditional grid and administrative units. It clarifies the spatial pattern and key driving mechanisms of debris flow susceptibility in the upper reaches of the Fujiang River, providing a scientific basis and technical support for precise disaster prevention and control, land-use planning, and ecological protection in mountainous regions of southwest China.\u003c/p\u003e","manuscriptTitle":"Susceptibility Assessment of Debris Flow Hazards from the Perspective of Watershed Units Grounded in the Random Forest (RF) Model — A Case Study in the Upper Reaches of the Fujiang River","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-27 01:46:23","doi":"10.21203/rs.3.rs-9123924/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-04-14T08:51:10+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-14T02:56:20+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-31T02:57:24+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"131848115554356215333301531881080619658","date":"2026-03-29T11:51:31+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"233829944099304923860578423662727111967","date":"2026-03-24T04:23:13+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-03-24T04:10:28+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-23T23:07:17+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-03-23T17:53:18+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-21T05:13:04+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2026-03-21T05:00:54+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"2bed81c2-c045-4751-a8ef-722e034917e2","owner":[],"postedDate":"March 27th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":65213030,"name":"Biological sciences/Ecology"},{"id":65213031,"name":"Earth and environmental sciences/Ecology"},{"id":65213032,"name":"Earth and environmental sciences/Environmental sciences"},{"id":65213033,"name":"Earth and environmental sciences/Hydrology"},{"id":65213034,"name":"Earth and environmental sciences/Natural hazards"}],"tags":[],"updatedAt":"2026-04-22T11:40:12+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-27 01:46:23","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9123924","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9123924","identity":"rs-9123924","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}