Zoning of the Disaster-Inducing Environment and Driving Factors for Landslides, Collapses, and Debris Flows on the Qinghai–Tibet Plateau

Abstract

The Qinghai–Tibet Plateau is one of the most geologically active regions in the world, characterized by significant geomorphic variation and a wide range of geological hazards. The multifactorial coupling of tectonic movements, geomorphological evolution, climate variability, and lithological characteristics contributes to the pronounced spatial heterogeneity of the disaster-inducing environment. Identifying key controlling factors and their driving mechanisms is crucial for effective regional disaster prevention and mitigation. This study adopts a systematic framework based on regional disaster systems theory, integrating tectonic activity, engineering geology, topography, and precipitation to construct a multi-factor zoning system. Using the Random Forest model, we quantify factor contributions and delineate eight distinct disaster-inducing environment zones. Zones I–III (Himalayas–Hengduan Mountains–Qilian Mountains) are characterized by a dominant coupling mechanism of “tectonic fragmentation—topographic relief—precipitation erosion” and account for the majority of large-scale disasters. In contrast, Zones IV–VIII, primarily located in the central–western Plateau basins, are constrained by limited material sources, resulting in lower disaster densities. The findings indicate that geological structures and lithological fragmentation provide the material foundation for hazard occurrence, while topographic potential and hydrodynamic forces serve as critical triggering conditions. This nonlinear coupling of factors shapes a disaster geographic pattern characterized by “dense in the east and sparse in the west”. Based on these results, the targeted recommendations proposed offer valuable theoretical insights and methodological guidance for disaster mitigation and region-specific management across the Qinghai–Tibet Plateau.

1. Introduction

The Qinghai–Tibet Plateau, situated at the collision zone between the Indian and Eurasian plates, is one of the highest and most geologically active regions on Earth. It features intense topographic relief, diverse geomorphic types, and a complex tectonic framework marked by densely distributed active faults. The region is further shaped by the interaction of monsoonal, non-monsoonal, and westerly atmospheric circulations, resulting in uneven spatiotemporal precipitation, frequent extreme weather events, and a highly fragile ecosystem. Under this complex natural setting, slope-related geological hazards—including collapses, landslides, and debris flows—occur frequently, characterized by sudden onset, spatial discreteness, and cascade evolution processes [1]. From the perspective of regional disaster systems theory, such hazards result from the coupling of the disaster-inducing environment, triggering factors, and disaster-bearing bodies. The disaster-inducing environment provides the geological and geomorphic foundation for hazard development and acts as the key spatial carrier for the accumulation of controlling conditions. Particularly in Plateau regions, the intricate geomorphic and tectonic settings give rise to highly heterogeneous hazard-inducing environments, with clear spatial variations in dominant controlling mechanisms [2,3]. Therefore, establishing a zoning system tailored to the Plateau’s natural geographic diversity is urgently needed to identify zone-specific dominant factors and support regionally differentiated hazard risk management strategies.

In recent years, rapid urbanization and infrastructure expansion across the Plateau have amplified regional disaster risks. Major events such as the 2022 flash flood and debris flow in Datong (Qinghai), the 2023 glacial avalanche and landslide chain disaster in Sedongpu Gully (Tibet), and the 2024 landslide in Zhenxiong (Yunnan) have posed serious threats to local communities [4]. Against the backdrop of climate warming and ongoing tectonic activity, such frequent and abrupt hazard events are increasing. Their occurrence is driven by the coupling of multiple natural and anthropogenic factors, ranging from tectonic fragmentation, geomorphic potential, and hydro-meteorological processes to subsurface heating and thermal field evolution induced by engineering activities [5]. However, the structure and dominant mechanisms of the hazard-inducing environment on the Plateau remain insufficiently understood, which has become a major gap in the implementation of precise disaster prevention and risk control.

Although substantial progress has been made in hazard susceptibility mapping and risk identification, most studies focus on predictive modeling based on topographic, geological, and historical hazard information [6,7]. Others use administrative or natural units to delineate risk zones, enhanced by remote sensing and machine learning techniques [8]. However, these approaches primarily assess risk outcomes and the vulnerability of disaster-bearing bodies [9,10], often neglecting in-depth investigation into the spatial structure, regional differentiation, and controlling mechanisms of the hazard-inducing environment itself. More recently, a few studies have attempted to construct hazard-environment zoning systems. For example, Zhang Zizhao et al. (2023) developed a zoning model for collapses, landslides, and debris flows, and Wei Luning et al. (2024) proposed a multi-factor coordinated zoning framework for identifying spatial hazard controls on the Plateau [11,12]. Nonetheless, limitations such as weak adaptability to plateau-scale complexity, insufficient attribution analysis, and limited model interpretability remain, hindering systematic hazard identification under the Plateau’s complex geophysical background.

To address these challenges, this study focuses on the Qinghai–Tibet Plateau, analyzing the spatial distribution of collapses, landslides, and debris flows from 2000 to 2020. Key disaster-inducing factors—including tectonic structure, geomorphology, precipitation, and engineering geology—are selected to construct a multi-factor multi-scale environmental zoning system. By employing a Random Forest model combined with stratified sampling and raster-based weighted overlay analysis, we identify the dominant drivers of hazard formation and quantify their spatial contributions. This enables us to reveal regional heterogeneity in the structure and the mechanisms of hazard-inducing environments across different zones. This study ultimately aims to improve our understanding of disaster-inducing environments in Plateau regions, offering theoretical support and methodological guidance for hazard mechanism analysis, risk-zoning optimization, and region-specific disaster prevention and mitigation strategies.

2. Materials and Methods

2.1. Overview of the Study Area

The Qinghai–Tibet Plateau, the highest geographic unit on Earth (Figure 1), has an average elevation exceeding 4000 m and spans approximately 2.5724 × 10⁶ km², accounting for nearly one-quarter of China’s total land area. Located at the collision zone between the Indian and Eurasian plates, the region is characterized by intense tectonic activity, well-developed fault systems, and a broad spectrum of geomorphic types—including alpine valleys, glacial landforms, and faulted basins. This geomorphic diversity underpins the spatial heterogeneity of geological hazards across the Plateau [13,14]. Climatically, the region exhibits a pronounced southeast-to-northwest gradient, transitioning from humid to arid zones. Most annual precipitation is concentrated between May and August, during which frequent episodes of heavy rainfall serve as critical triggers for landslides, debris flows, and related geological hazards. Moreover, due to the Plateau’s high altitude and cold climate, large diurnal temperature variations and recurrent freeze–thaw cycles significantly enhance rock weathering and contribute to slope instability [15,16,17]. The ecological environment is equally vulnerable, marked by low vegetation coverage dominated by alpine meadows, steppes, and sparse shrubs, with limited resilience to external disturbances [18,19]. The interplay of complex tectonics, diverse landforms, extreme climatic variability, and ecological fragility forms a classic multi-factor coupled disaster-inducing environment. As a result, the Qinghai–Tibet Plateau ranks among the most hazard-prone regions in China, frequently experiencing collapses, landslides, and debris flows, and exhibiting some of the most complex environmental conditions for disaster development.

2.2. Sources of Data and Preprocessing

The data used in this study encompass topography, geological structures, engineering geology (lithology), climate (precipitation), and geological hazard records, as summarized in

Table 1. Metadata details table.

No	Hazard-Inducing Factors and Disaster Points	Data Sources
1	Topography and Geomorphology	Based on the landform classification system of DEM [20,21].
2	Geological Structures	Geotectonic map of the Qinghai–Tibet Plateau and its adjacent areas [22].
3	Stratigraphic Lithology	Derived from ISRIC Report (2008.06). https://www.isric.org/taxonomy/term/71 (accessed on 28 September 2024)
4	Precipitation Zoning	Precipitation zoning in the Qinghai–Tibet Plateau (1978–2018) and spatiotemporal evolution characteristics [23,24].
5	Disaster Points	Resources and Environment Science Data Center, Chinese Academy of Sciences. https://www.resdc.cn/ (accessed on 25 September 2024)
6	Study Area	Boundary of the Qinghai–Tibet Plateau. https://www.geodoi.ac.cn/ (accessed on 25 September 2024)

. To ensure the scientific validity of the model analysis and the accuracy of spatial representation, all datasets were projected to the WGS84 coordinate system during preprocessing and were standardized through spatial verification to ensure consistency and suitability for spatial analysis and modeling. (1) Topographic data: A geomorphological classification system based on Digital Elevation Model (DEM) data, developed by Wang et al. (2020) [20] and Liu Yi (2022) [21], was employed. This method integrates multi-scale geomorphic factor analysis, allowing for the accurate delineation of complex landform units across various spatial scales on the Plateau. The refined classification supports the investigation of potential associations between geomorphic features and geological hazards. (2) Geological structure data: Structural data are derived from the 1:1,500,000 Geological Map of the Qinghai–Tibet Plateau compiled by Pan Guitang et al. (2013) [22]. This dataset includes information on active faults, fault density, and structural boundaries. Tectonic activity, as a fundamental internal driving force within the disaster-inducing environment, influences both rock fragmentation and the spatial patterns of hazards such as landslides and collapses. (3) Engineering geology (lithology) data: Lithological data are extracted from the ISRIC-WISE Global Soil Database (ISRIC, 2008), with a focus on lithological classifications specific to the Qinghai–Tibet Plateau. The spatial distributions of various lithological combinations were analyzed. Given the differing physical and chemical properties of rock types—such as weathering susceptibility, permeability, and structural stability—this factor significantly influences hazard sensitivity. Accordingly, spatial layers for engineering geology were developed to evaluate their contributions to hazard formation across the study area. (4) Climate (precipitation) data: Climate data are sourced from precipitation zoning results based on long-term meteorological observations (1978–2018) across Plateau stations, as reported by Gong Chengqi et al. (2022) [23] and Ma Yaoming et al. (2018) [24]. This dataset captures regional precipitation patterns, including annual totals, intensity, and seasonal variation, facilitating the quantitative analysis of precipitation as a trigger under varying geomorphic and tectonic contexts. (5) Geological hazard data: Hazard records were obtained from the national geological disaster census released by the Chinese Academy of Sciences, covering the period 2000–2020. This dataset includes 13,837 validated records of landslides, debris flows, and collapses (CLD) across the Qinghai–Tibet Plateau. Notably, disaster records are sparse in the northwestern and central Plateau due to lower population density and limited monitoring. After conducting topological validation and coordinate standardization, a temporally and spatially consistent hazard database was constructed. This database serves as the core for model validation, factor response analysis, and zoning assessment. (6) Study area boundary: The boundary of the Qinghai–Tibet Plateau was defined based on the delineation proposed by Zhang Yali et al. (2014), which accurately represents the region’s topographic extent and geological context while ensuring administrative alignment [25].

Preprocessing and standardization methods: On the GIS platform, the spatial zoning of topography, tectonics, lithology, and climate was conducted based on their natural attributes and prior zoning frameworks. The density of CLD disaster points (i.e., frequency per unit area) within each zone was calculated to construct a basic factor–response indicator system [26]. All variables were normalized and standardized to ensure numerical comparability and improve model performance in factor weight calculation and zoning reliability.

2.3. Research Methods

2.3.1. Random Forest Model and Accuracy Metrics

This study employs the Random Forest classifier from the ensemble learning library in Python (part of the Scikit-learn package) (https://colab.research.google.com/ (accessed on 5 January 2025)) as the core algorithm tool to simulate the response relationships between disaster-inducing environmental factors and geological hazard points, quantifying the contribution of each factor to landslide, collapse, and debris flow disasters [27,28,29] (Equation (1)). A random selection of 80% of the samples is designated as the training set, while 20% are used as the validation set to simulate and calculate related targets based on the associative conditions between disaster-inducing environmental factors and hazards. According to previous studies, once the number of feature trees exceeds 500, improvements in the model tend to stabilize, yet the computational costs increase significantly [30,31,32]. Therefore, this study limits the range of feature trees to 0–500, striking a good balance between computational efficiency and predictive performance.

f(m)_i,j = ω_T T + ω_Gs Gs + ω_Pr Pr + ω_Eg Eg + ∈

(1)

In the formula, f(m)_i,j denotes the geohazard prediction result at point/pixel i in zone j, determined by the random forest model; T, Gs, Pr, and Eg represent the geomorphology, geological structure, precipitation, and engineering geology factors at point/pixel i in zone j; ω_T, ω_Gs, ω_Pr, and ω_Eg are the model weights assigned to these factors in zone j; and ϵ is the model residual.

To ensure the accuracy and reliability of the Random Forest model in simulating and calculating disaster-inducing environmental factors for geological hazards, this study introduces Root Mean Squared Error (RMSE, Equation (2)) as the evaluation metric. By calculating the error values of the Random Forest model under different numbers of random feature trees, a quantitative assessment of the model’s precision is achieved. In the analysis of disaster-inducing environmental factors, RMSE can accurately quantify the predictive accuracy of each factor for disaster-inducing scenarios. Based on this, the importance distribution of each environmental factor influencing geological disasters is obtained. Guided by zoning methods, the disaster-inducing environmental zoning of landslides, collapses, and debris flows on the Qinghai–Tibet Plateau is completed, which has also benefited from lattice-based conductivity modeling [33], along with the contribution degree of environmental factors in each zone.

$$RMSE=\sqrt{{\displaystyle \frac{1}{n}}\sum _{i=1}^n{(y_i-{\widehat{y}}_i)}^2}$$

(2)

In the formula, RMSE represents the degree of deviation between predicted and actual values, n is the number of samples, $y_i$ denotes the true value of the i-th sample, and ${\widehat{y}}_i$ denotes the predicted value of the i-th sample.

2.3.2. Zoning Method

The disaster-inducing environment of geological hazards on the Qinghai–Tibet Plateau exhibits pronounced spatial heterogeneity, with its formation mechanisms governed by the coupling of multiple factors, including topography, geological structures, lithology, hydrology, and climate [34,35]. The scientific zoning of such environments is essential for identifying dominant hazard-inducing factors and elucidating the mechanisms underlying hazard development. Based on physical geography systems theory and spatial analysis methodologies, this study develops a “multi-factor–multi-perspective–multi-scale” technical approach to hazard-inducing environment zoning. This approach integrates both qualitative and quantitative analyses to reveal spatial heterogeneity in factor distributions and corresponding hazard patterns [26].

The purpose of hazard-inducing environment zoning is to support geological hazard susceptibility assessments by ensuring consistency in formation mechanisms within zones and highlighting significant differences between zones. This study adheres to three fundamental principles: first, the principle of spatial correlation and heterogeneity, which groups spatially clustered or characteristic-consistent factors into the same zone, while significantly different ones are divided into different zones; second, the principle of comprehensiveness and independence, which integrates the response relationships between multiple factors and hazards while maintaining independent analysis of the spatial structure of factors; and third, the principle of specificity, which identifies regional characteristics that conventional factors cannot reveal through field surveys and the literature, thereby enhancing the scientific validity and applicability of the zoning.

Following these principles, this study establishes a technical workflow based on a “factor-driven, model-supported, GIS-based weighted overlay” strategy. First, key hazard-inducing factors—including landform types, geological structures, lithology, and precipitation—are selected to construct a standardized input dataset. Geological hazard points (landslides, collapses, and debris flows) from 2000 to 2020 are compiled into a spatial hazard database. Second, on the ArcGIS platform, the spatial relationships between hazard points and environmental factors are analyzed, with indicators such as hazard density and frequency extracted to identify each factor’s response characteristics. Third, a Random Forest model is employed to quantify the contribution of each factor, calculate corresponding weights, and generate a composite evaluation layer through weighted overlay analysis. Subsequently, spatial zones exhibiting significant heterogeneity are extracted. Finally, a hazard-inducing environment zoning map is produced. This methodology enables the systematic identification of dominant factors and regional disparities. It also incorporates FEM-based assessments of subsurface environmental structures [36], offering a scientific basis for analyzing hazard formation mechanisms and guiding disaster prevention and mitigation strategies [37].

3. Results

3.1. Zonation of Hazard-Inducing Environments

Landslides, collapses, and debris flows (CLD), as typical slope-related geological hazards, are collectively governed by a range of factors—primarily topography and geomorphology, stratigraphy and lithology, geological structures, and atmospheric precipitation [38]. Based on a systematic analysis of the geological background and developmental characteristics of hazard-inducing environments across the Qinghai–Tibet Plateau—and incorporating relevant regional research findings [39,40]—this study selects key factors, including geomorphology, tectonics, climate, and engineering geological conditions, for comprehensive analysis.

Given that each factor exerts varying degrees of influence, assigning rational weights is essential for integrated evaluation. To this end, a Random Forest model was applied to quantitatively assess the contribution of each factor. As shown in Figure 2a, RMSE decreased to 2.901 when the number of trees reached 490, significantly below 35% of the true mean value of f(m)_i,j. This indicates high predictive accuracy and supports the reliability of the calculated factor contributions.

The model results yielded weights of 0.276, 0.306, 0.248, and 0.169 for geomorphology, geological structures, precipitation, and engineering geology, respectively, reflecting their relative importance in shaping hazard-inducing environments. Based on these weights, a comprehensive evaluation map was generated through weighted overlay operations using the GIS raster calculator. Following established zoning principles [41,42], the hazard-inducing environment of the Qinghai–Tibet Plateau was divided into eight distinct zones (Figure 2b), described as follows:

Zone I: Alpine Valley Freeze–Thaw Tectonic Zone. Located along the Yarlung Tsangpo River Valley and adjacent highlands, this zone is characterized by dense infrastructure (roads and settlements) and high anthropogenic activity. Significant inter-factor variability and a high hazard density (1.1 events/100 km²) designate it as a primary concentration area for geological hazards.

Zone II: Alpine Canyon Tectonic–Erosional Zone. Primarily distributed across the Hengduan Mountains and parts of western Sichuan, this zone exhibits steep canyon topography, complex lithologies, active tectonic movements, and intense runoff. It shows high factor variability and the highest hazard density (1.21 events/100 km²), highlighting its spatial heterogeneity.

Zone III: Alpine Freeze–Thaw Tectonically Active Zone. Centered in the Qilian Mountains, this zone features pronounced freeze–thaw weathering and active tectonics. While factor differences are somewhat less prominent, the zone maintains a high hazard density (1.04 events/100 km²) and notable spatial variability.

Zone IV: Faulted Basin Weathered Fragmentation Zone. Located in the Qaidam–Gonghe–Huangshui Basin region on the northeastern margin of the Plateau, this zone is characterized by fault-controlled basins with highly weathered and fragmented rock masses. Despite moderate inter-factor variability, it shows a hazard density of 0.3 events/100 km², with events primarily clustered in the eastern subzones, where human activity is greater.

Zone V: Permafrost–Weathered Rock Belt. This zone spans the East–West Kunlun Mountains and experiences intense freeze–thaw cycles along with loosely consolidated surficial materials. It exhibits moderate hazard density (0.2 events/100 km²) and substantial inter-factor variability.

Zone VI: Permafrost–Nival Erosion Zone. Distributed across the Qiangtang Plateau and Tanggula Mountains, this high-altitude zone is dominated by freeze–thaw and aeolian erosion. Although hazard density is low (0.13 events/100 km²), it demonstrates observable factor variability and spatial heterogeneity.

Zone VII: Alpine Hard Rock Freeze–Thaw Debris Zone. Found in the Karakoram Mountains and parts of the Qiangtang Plateau, this zone features fractured bedrock, intense freeze–thaw activity, and accumulated debris. Despite notable factor variability, hazard density remains low (0.03 events/100 km²), with sparse occurrences.

Zone VIII: Alpine Lake Basin Saline Soft Rock Zone. Located primarily in lake basins of the Qiangtang Plateau, this arid zone contains saline-softened lithologies and surface fracturing. Although inter-factor variability is present, the hazard density is low (0.13 events/100 km²), and hazard occurrences are infrequent.

3.2. Factor Contributions and Hazard-Inducing Environmental Characteristics Across Zones

To analyze the spatial heterogeneity of hazard-inducing environments for landslides, collapses, and debris flows (CLD) across the various zones of the Qinghai–Tibet Plateau, and to elucidate the disaster-forming mechanisms within each zone, Random Forest models were constructed based on the previously delineated environmental zones. The modeling and validation results (Figure 3) indicate that the optimal number of trees for the Random Forest models in Zones I–VIII are 400, 400, 70, 70, 10, 100, 100, and 400, respectively. In all zones, RMSE values were below 5 and less than 35% of the true mean values—specifically 1.563, 3.392, 2.433, 2.292, 1.228, 0.407, 0.34, and 0.253—demonstrating the high accuracy and reliability of the models in calculating factor contributions within each zone.

3.2.1. Zonal Factor Characteristics

As illustrated in Figure 4, the CLD hazard-inducing environments across Zones I–VIII display two primary patterns, geomorphological–structural dominance and climate–engineering geology co-driving, characterized by pronounced regional differentiation and multi-factor coupling. The specific characteristics of each zone are as follows:

Zone I: Alpine Valley Freeze–Thaw Tectonic Zone: Dominated by geological structures (30.884%) and precipitation (26.482%) as the principal hazard-inducing factors.

Zone II: Alpine Canyon Tectonic–Erosion Zone: Primarily controlled by geological structures (30.531%) and terrain geomorphology (27.034%), with precipitation (25.551%) exerting a synergistic influence.

Zone III: Alpine Freeze–Thaw Tectonically Active Zone: Precipitation (33.134%) is the leading hazard-inducing factor, operating in conjunction with geological structures (24.705%) and engineering geological conditions (22.13%) in a co-driven mechanism.

Zone IV: Faulted Basin Weathered Fragmentation Zone: Engineering geological conditions (27.454%) and terrain geomorphology (26.889%) serve as the dominant factors, supplemented by geological structures (23.844%) and precipitation (21.814%) through a compound interaction.

Zone V: Permafrost–Weathered Rock Belt: Jointly driven by precipitation (30.164%) and engineering geological conditions (26.34%), with terrain geomorphology (24.627%) further contributing to the hazard-inducing environment.

Zone VI: Permafrost–Nival Erosion Zone: Terrain geomorphology (29.88%) and precipitation (26.885%) are the primary controlling factors, co-driven by engineering geological conditions (23.353%).

Zone VII: Alpine Hard Rock Freeze–Thaw Debris Zone: Engineering geological conditions (42.077%) represent the core hazard-inducing factor, while precipitation (26.731%) contributes to the triggering mechanism.

Zone VIII: Alpine Lake Basin Saline Soft Rock Zone: Dominated by engineering geological conditions (31.868%) and precipitation (28.52%), with terrain geomorphology (20.218%) playing a supporting role through spatial coupling effects.

3.2.2. Hazard-Inducing Environment and Factor Contributions

The spatial differentiation of dominant factors is fundamentally rooted in the diverse natural geographical conditions of the Qinghai–Tibet Plateau. Overall, the hazard-inducing environment—characterized by geomorphology, geological structure, climatic conditions, and engineering geology—exhibits pronounced spatial heterogeneity. Notably, the thermal behavior of lithology under cyclic conditions has also been shown to influence slope responses to dynamic subsurface processes [43].

(1)

Geomorphology

The geomorphological landscape of the Plateau is typified by “crisscrossing mountains and interspersed lake basins” [44] and is subdivided into eight first-order geomorphic units [20,21]. Terrain relief exhibits a distinct spatial gradient: plateaus and low-relief mountains dominate the Northern Tibetan Plateau and Qinghai Plateau, while high-relief mountainous terrain is concentrated in the southeastern and northwestern margins (Figure 5a). Geological hazards, such as debris flows and landslides, predominantly occur in steep mountainous areas, including the Himalayan Extremely High Mountain Region, Hengduan Mountain Gorge Region, and Qaidam–Huangyuan Mountain Basin (Figure 5b). These zones correspond to the highest geomorphological contributions in the model for subregions I–IV (22.533%, 27.034%, 20.031%, and 26.889%, respectively), indicating that steep slopes constitute fundamental conditions for hazard development.

(2)

Geological structure

Situated at the collision interface of the Indian and Eurasian plates, the Plateau is marked by intense tectonic activity and the development of major fault systems such as the Bangong–Nujiang and Jinsha River fault zones [45]. These have shaped complex structural systems, including the Qiangtang–Sanjiang and Qin–Qilian–Kunlun orogenic belts [22] (Figure 6a). Fault zones and tectonic boundaries often consist of highly fractured rock masses and deeply incised terrain, forming fragile geological substrates conducive to hazard formation [46]. Model results confirm the substantial contribution of tectonic structures in subregions I–IV (30.884%, 30.531%, 24.705%, and 23.844%), aligning well with the observed concentration of disasters along active tectonic belts such as the Himalayas and Hengduan Mountains (Figure 6b).

(3)

Climate (precipitation)

Precipitation across the Plateau displays significant spatial heterogeneity [47,48,49]. Southeastern regions receive over 1000 mm of annual rainfall due to moist air masses from the Indian Ocean, while northwestern areas remain arid [50]. In addition, localized short-duration rainstorms frequently occur in the summer months [51,52]. Precipitation exerts a dominant influence in subregions III, V, VII, and VIII (33.134%, 30.164%, 26.731%, and 28.52%, respectively). These values correspond with the higher disaster densities observed in climatic zones I, II, and III (Figure 7), demonstrating the role of intense surface runoff in triggering mass movements in unconsolidated rock–soil materials [53,54].

(4)

Engineering geology

The Plateau’s engineering geological framework includes hard rock, gravel, and loose sedimentary materials (Figure 8a). Loose sediment zones are mainly concentrated in the central and western regions (Figure 8b), aligning with the high contribution values of engineering geological factors in subregions IV, V, VII, and VIII (27.454%, 26.34%, 42.077%, and 31.868%, respectively). These results highlight the influence of lithological physical properties—such as weathering susceptibility and thermal conductivity—on disaster occurrence and sensitivity [55].

In summary, the hazard-inducing environment of the Qinghai–Tibet Plateau exhibits strong spatial heterogeneity, with distinct structural compositions and dominant factors across regions. Interactions between tectonics, climate, geomorphology, and engineering geology manifest as either synergistic or individually dominant controls, ultimately shaping the spatial patterns and formation mechanisms of landslides and debris flows throughout the Plateau.

3.3. Characteristics and Causal Mechanisms of Collapse, Landslides, and Debris Flows (CLD) in Each Environmental Zone

3.3.1. Geological Disasters and Disaster-Inducing Factors

The Qinghai–Tibet Plateau’s geomorphological pattern—characterized by interlaced mountain ranges and lake basins—combined with complex geological structures, hydroclimatic conditions, and anthropogenic engineering activities, jointly shapes the unique spatial distribution characteristics of collapses, landslides, and debris flows (CLD), which exhibit pronounced spatial heterogeneity. As shown in

Table 2. Statistical table of geological hazards in integrated zoning (I–VIII).

Zoning of Disaster-Inducing Environments	Area/km2	Collapse	Landslide	Debris Flow	Small Plan (One)	Density/Place × (100 km)−2	Percentage/%
Alpine Valley Freeze–Thaw Tectonic Zone (I)	532,429	910	1012	3960	5882	1.1	42.51
Alpine Canyon Tectonic-Erosion Zone (II)	396,643	963	1739	2105	4807	1.21	34.74
Alpine Freeze–Thaw Tectonic Active Zone (III)	85,036	168	334	380	882	1.04	6.37
Faulted Basin Weathered Fragmentation Zone (IV)	303,823	180	89	632	901	0.3	6.51
Permafrost-Weathered Rock Belt (V)	254,436	255	34	213	502	0.2	3.63
Permafrost-Nival Erosion Zone (VI)	394,924	124	58	343	525	0.13	3.79
Alpine Hard Rock Freeze–Thaw Debris Zone (VII)	462,211	29	5	109	143	0.03	1.03
Alpine Lake Basin Saline Soft Rock Zone (VIII)	149,773	69	7	119	195	0.13	1.41

, Comprehensive Zones I–III represent high-density hazard areas, with CLD densities of 1.10, 1.21, and 1.04 events per 100 km², respectively, accounting for a combined 84.32% of all hazard events. Zone IV exhibits a moderate density of 0.30 events per 100 km². This pattern closely correlates with geological structures and topographic conditions.

Comprehensive Zone I, located in the Himalaya–Gangdese Mountains, is shaped by the ongoing collision between the Indian and Eurasian plates, resulting in densely developed fault systems, fragmented rock masses, and extreme topographic relief—all of which provide abundant material sources and gravitational potential for collapses and landslides. Zone II, situated in the Hengduan Mountains and western Sichuan region, is defined by the deeply incised valley landscape formed by the “Three Parallel Rivers” geomorphology, and is highly prone to debris flows triggered by concentrated monsoonal rainfall [56]. Zone III, in the Qilian Mountains, frequently experiences CLD events due to active tectonic fragmentation combined with localized precipitation and snowmelt processes [57]. Zone IV, positioned along the margins of the Qaidam–Gonghe–Huangshui Basin, is characterized by intense weathering and an abundance of loose clastic material; under the influence of snowmelt and short-duration heavy rainfall, debris flows and collapses are readily triggered [58]. In Zone V (Eastern and Western Kunlun Mountains), intense tectonic uplift and freeze–thaw weathering have led to the accumulation of thick surficial debris layers. Under the joint action of glacial meltwater and sudden precipitation, debris flows and periglacial collapses and landslides frequently occur [59,60]. By contrast, Zones VI–VIII, located in the Qiangtang Plateau, Karakoram, and Tanggula Mountains, are characterized by gentler terrain and sparse human activity, resulting in significantly lower hazard densities. This reflects the combined regulatory effect of natural environmental conditions and anthropogenic disturbance on hazard occurrence and distribution.

In terms of disaster type distribution, collapses are primarily concentrated in Zones I and II, with 910 and 963 occurrences, respectively; conditions favoring collapse development include densely developed fault zones, fractured lithologies, and steep slopes. Zones III and V recorded 168 and 255 collapses, respectively, reflecting moderate elevation gradients and rock weathering. In contrast, the flat terrain of Zone VII corresponds to just 29 recorded collapse events. Landslides are most concentrated in Zone II, with 1739 occurrences, driven by steep topography and rock softening under concentrated rainfall. Zone I follows with 1012 landslides, largely attributed to frequent tectonic activity. Zones III and IV recorded 334 and 89 landslides, respectively, concentrated in mountainous areas with pronounced relief. Zone VIII recorded only seven landslides, consistent with the limited conditions for landslide initiation in flat terrain. Debris flows are densest in Zone I, with 3960 recorded events, followed by Zone II with 2105 occurrences. Both are influenced by a combination of intense precipitation, abundant loose surface material, and steep slope convergence. Debris flows in Zone IV (632) and Zone III (380) are associated with snowmelt, short-duration rainfall, and abundant weathered debris. Zones VIII (119) and VII (109) recorded significantly fewer debris flows, primarily due to low precipitation and limited source material.

In summary, Zones I and II—characterized by active tectonic processes and complex terrain—serve as core development areas for collapses, landslides, and debris flows (CLD). Other zones, constrained by gentler slopes, limited material sources, and lower precipitation, exhibit reduced hazard development. Collectively, this distribution pattern underscores the joint control of geological background, topographic relief, climatic conditions, and engineering geological (lithological) factors over the spatial pattern of CLD hazards across the Qinghai–Tibet Plateau.

3.3.2. Analysis of Intensity and Hazard-Formation Processes

Based on field survey data and hazard scale classifications, landslides, collapses, and debris flows (CLD) on the Qinghai–Tibet Plateau are predominantly small- to medium-scale events, with large-scale disasters occurring less frequently. Additionally, 2858 hazard sites lack scale annotations. Statistical analysis (Figure 9,

Table 3. Statistics on the scale of geological hazards on the Qinghai–Tibet Plateau where collapses, landslides and debris flows (CLD) have occurred.

Scale	Disaster Type and Proportion of/%
	Collapse	Landslides	Debris Flows
Extra-large	0.10	0.08	0.51
Large	1.88	3.86	5.58
Medium	4.41	8.91	18.74
Small	15.78	8.73	31.41

) shows that debris flows represent a disproportionately high share of medium-, large-, and exceptionally large-scale events, accounting for 18.74%, 5.58%, and 0.51%, respectively, indicating their greater destructive potential. In contrast, collapses and landslides are primarily small-scale events, comprising 15.78% and 8.73% of all hazards, although a notable proportion of large-scale occurrences are still recorded (collapses: 1.88%, landslides: 3.86%), demonstrating that some collapses and landslides also have significant impacts. Overall, debris flows are more likely than collapses and landslides to generate medium- and high-intensity disasters.

A total of 1319 large and exceptionally large-scale disasters have been documented across the Plateau, predominantly concentrated in Comprehensive Zone II (48.67%), followed by Zone I (28.96%) and Zone III (17.06%), while Zones IV–VIII together account for only 5.3%. This spatial distribution is primarily controlled by geomorphological and tectonic conditions. Zones IV–VIII are situated in the central and northwestern regions of the Plateau, characterized by broad basins and relatively gentle terrain, which are generally unfavorable for the development of large-scale hazards. Conversely, Zones I–III, located within the Himalayas, Hengduan Mountains, and Qilian Mountains, feature steep topography and active tectonics, providing favorable conditions for the occurrence of large-scale disasters. Consequently, 94.69% of all large- and exceptionally large-scale events are concentrated in these three zones.

Although large-scale disasters result from the combined influence of endogenic and exogenic geological processes, they comprise only 12.01% of the total. Small- to medium-scale disasters dominate the region, with 9655 recorded events, accounting for 87.99%. These smaller-scale hazards are triggered by the interaction of tectonic structures, meteorological factors, and engineering geological conditions. The differences in hazard-inducing environmental factors across the eight comprehensive zones lead to varied and complex hazard formation mechanisms, summarized as follows: Zone I (Yarlung Tsangpo River Valley and Adjacent Highlands): Tectonically fractured lithology (30.884%) provides the material basis, precipitation (26.482%) promotes erosion, and anthropogenic disturbances amplify instability, forming a “tectonic fracturing–engineering disturbance–precipitation erosion” mechanism. Subsurface thermal gradients and seasonal heating may further exacerbate slope instability [61]. Zone II (Hengduan Mountains and Western Sichuan): Tectonic activity (30.531%) modifies rock structures; high-relief canyon terrain (27.034%) provides gravitational energy; precipitation (25.551%) accelerates weathering and saturation; and human activity contributes to a “tectonic modification–geomorphic potential–precipitation erosion–engineering disturbance” mechanism. Zone III (Qilian Mountains): Freeze–thaw cycles and tectonic activity (24.705%) cause rock fragmentation; precipitation (33.134%) triggers instability; and engineering geological fragility (22.13%) heightens susceptibility, forming a “precipitation-triggered–freeze–thaw fragmentation–tectonic activity” mechanism. Zone IV (Qaidam–Gonghe–Hehuang Basins): Intense weathering (27.454%) weakens lithology; basin-edge terrain (26.889%) provides topographic potential; snowmelt and short-duration precipitation (21.814%) serve as triggers; and tectonic activity (23.844%) contributes to a “weathered lithology–geomorphic relief–precipitation triggering–tectonic influence” mechanism. Zone V (Eastern and Western Kunlun Mountains): Freeze–thaw processes (26.34%) generate loose surface materials; glacial meltwater and episodic rainfall (30.164%) act as erosional forces; and geomorphology (24.627%) contributes to a “freeze–thaw accumulation–precipitation erosion–geomorphic potential” mechanism, primarily leading to debris flows and periglacial mass wasting. Zone VI (Qiangtang Plateau and Tanggula Mountains): Freeze–thaw and aeolian erosion processes (23.353%) reshape lithology; precipitation (26.885%) further erodes weakened materials; and gentler terrain (29.88%) reduces the likelihood of large-scale disaster formation, forming a “freeze–thaw/aolian degradation–precipitation erosion–low-relief terrain” mechanism. Zone VII (Karakoram Mountains and Qiangtang Plateau): Alpine freeze–thaw cycles result in lithological fragility (42.077%); sporadic precipitation (26.731%) triggers instability; the resulting “freeze–thaw weakening–precipitation triggering” mechanism leads to sparse disaster occurrences due to limited hydrometeorological activity and human disturbance. Zone VIII (Lake Basins of the Qiangtang Plateau): Drought-induced weathering (31.868%) leads to soft, loose lithology; precipitation (28.52%) causes erosion; and flat terrain (20.218%) limits gravitational energy. The resulting “weathered lithological fragility–precipitation erosion–gentle topography” mechanism accounts for the minimal hazard occurrence observed in this zone.

In summary, the hazard formation mechanisms across zones reflect a multifactorial coupling of “tectonic structure–lithological properties–geomorphology–climate–human activity”. Tectonic activity and lithology provide the material basis, while precipitation and terrain offer the dynamic driving forces, and anthropogenic factors serve as critical triggers. Together, these elements co-regulate the spatial differentiation of landslide, collapse, and debris flow hazards on the Qinghai–Tibet Plateau.

4. Discussion

4.1. Reliability Assessment of the Zoning Results

The dominant factor weights derived from the Random Forest model and the comprehensive zoning results generated via GIS weighted overlay analysis confirm that the disaster-inducing environments for landslides, collapses, and debris flows on the Qinghai–Tibet Plateau exhibit clear characteristics of either “geomorphological–tectonic dominance” or “climate–engineering geology synergistic driving”. These patterns display significant regional variation and are shaped by multifactorial coupling mechanisms. A comparison with the existing literature confirms the robustness and scientific validity of the zoning results. The analysis is as follows: Zone I, primarily located in the Yarlung Zangbo River valley and adjacent high mountain areas, features frequent tectonic activity, pronounced topographic relief, and short-term heavy rainfall, creating a typical debris flow-dominated disaster-inducing environment. Prior research has indicated that this zone is mainly governed by tectonic fragmentation, precipitation, and freeze–thaw processes [62,63,64], consistent with the findings of this study; Zone II, situated in the Hengduan Mountains and parts of western Sichuan, presents a synergistic hazard-inducing mechanism dominated by geomorphic potential and precipitation erosion. Studies on the southeastern Qinghai–Tibet Plateau have highlighted that deeply incised canyons, active tectonics, and concentrated precipitation contribute to frequent debris flows and landslides [65,66,67,68], which aligns with this study’s identified mechanisms; Zone III, covering the Qilian Mountains, experiences severe lithological fragmentation due to tectonic activity and freeze–thaw cycles. Short-duration rainstorms act as triggering factors. Previous studies by Wu Weijiang et al. (2021), Li et al. (2023), and Zhao et al. (2022) confirm that disaster occurrences in this zone are primarily attributed to rock degradation from freeze–thaw activity and precipitation-induced instability [69,70,71,72,73,74]; Zone IV, located in the Qaidam–Huangshui Basin within the northeastern tectonic belt, exhibits intense weathering and lithological disintegration. This region has been documented as being composed of tectonic basins with heavily weathered and loose soils, where debris flows and collapses are commonly induced by snowmelt and short-term rainfall [75,76,77,78], which validates the coupling mechanisms identified herein; Zones V–VIII lie mainly in the northwestern and central Qinghai–Tibet Plateau, where flat terrain and limited material sources result in a hazard density of less than 0.30 sites per 100 km². Although freeze–thaw processes and engineering geological factors contribute to localized rock fragmentation, sparse rainfall and low human presence limit the overall scale and frequency of hazard development [79,80,81,82].

In conclusion, the zoning results and identified dominant factors in this study are well supported by field observations, regional hazard surveys, and previous environmental mechanism analyses. This confirms the spatial applicability and reliability of the model outputs and suggests that a multi-factor coupling analytical pathway based on “geotectonic structure–geomorphic potential–precipitation triggering–lithological response” can effectively characterize the structural patterns and causative mechanisms of hazard-inducing environments on the Qinghai–Tibet Plateau.

4.2. Regional Geo-Hazard Risk Mitigation Strategies

Given the variation in dominant factors and disaster formation mechanisms across the eight identified zones, region-specific risk mitigation strategies are essential for enhancing hazard prevention and control in the Plateau’s complex environment: Zone I (tectonic fragmentation and precipitation erosion): strengthen engineering controls along fault zones, conduct pre-monsoon inspections and dynamic monitoring, and prioritize structural mitigation such as sediment retention dams and diversion channels; Zone II (steep relief and high tectonic activity with intense precipitation): enhance slope stability assessments along valley edges, establish rainfall-triggered early warning systems, increase gully maintenance frequency, and implement evacuation protocols for mountain road networks and settlements; Zone III (freeze–thaw cycles and precipitation): deploy integrated temperature and humidity sensors in high-altitude areas to detect seasonal high-risk zones, and refine engineering designs through structural hazard assessments; Zone IV (highly weathered terrain and loose soils): establish disaster risk buffer zones around towns and promote targeted slope stabilization through artificial reinforcement and ecological restoration; Zone V (precipitation and freeze–thaw-driven material accumulation): develop a regional inventory of debris sources, assess the slope stability of unconsolidated deposits, and protect buffer zones along transportation corridors; Zone VI (climate and engineering geology synergy, lower hazard levels): prioritize ecological engineering measures to control freeze–thaw erosion and enhance the resilience of infrastructure to environmental disturbances; Zone VII (engineering geology-dominant, low precipitation): implement cost-effective prevention and early warning systems, including unmanned monitoring stations and pre-emptive evacuation protocols in low-frequency hazard areas; and Zone VIII (loose lithology and flat terrain): maintain ecological stability, minimize human disturbance, and promote low-impact land use strategies. These recommendations, grounded in dominant-factor identification and tailored to specific geomorphologic and environmental settings, provide a scientific foundation for targeted risk governance and precision disaster prevention [83].

4.3. Model Accuracy Evaluation and Limitations

Previous studies suggest that an RMSE value within 10–35% of the observed mean is acceptable for geological hazard modeling [84,85,86,87]. In this study, Random Forest models were constructed using 1–500 trees to determine the optimal number of features. When the number of trees reached the optimal value, the model achieved minimum RMSE across all zones, with values remaining below 35% of the true mean. This result aligns with Liu et al. (2020)’s conclusions in rainstorm disaster modeling (RMSE ≈ 30%) [88], further affirming the robustness of the Random Forest approach in multi-hazard contexts.

However, the model does have limitations. In the northwestern and central Plateau, where terrain is flat and population density is low, disaster occurrence is infrequent, leading to sparse and unevenly distributed data. This reduces the model’s ability to learn relevant hazard patterns, introducing uncertainty into the results and weakening predictive accuracy in these areas. To address this challenge, the following optimization strategies are proposed: (1) Supplementing environmental factors: Integrate critical variables such as permafrost distribution, active layer thickness, and dynamic groundwater levels, extractable via high-resolution remote sensing technologies (e.g., InSAR, MODIS, Sentinel-2), to enhance factor completeness in data-scarce areas [89]. (2) Balanced sampling design: Implement spatially balanced sampling to expand representative datasets in key geomorphic and structural units, improving generalization and model training quality. (3) Transfer learning optimization: Apply regional adaptive transfer learning by leveraging high-quality training samples from southeastern zones as source domains. Domain adaptation techniques can improve prediction in low-sample western regions while maintaining model structure integrity [90]. (4) Multi-source data integration: Combine geological maps, historical hazard archives, UAV data, and public disaster perception reports to create a multi-scale multi-dimensional hazard-inducing factor database, improving regional adaptability and predictive robustness. These strategies will not only reduce uncertainty in model outputs for the central and northwestern Plateau, but also offer more reliable scientific support for comprehensive regional geological hazard risk assessments.

5. Conclusions

This study employs geomorphological, tectonic, climatic, and engineering geological factors as core hazard-inducing environmental variables to delineate collapse, landslide, and debris flow (CLD) hazard-prone zones across the Qinghai–Tibet Plateau, systematically analyzing dominant environmental drivers and hazard formation mechanisms. The main conclusions are as follows:

(1)

Through the application of the Random Forest model, environmental zoning for CLD hazards across the Qinghai–Tibet Plateau was conducted. The selected models achieved high accuracy, with RMSE values all below 5 and predictive errors less than 35% of observed mean values. By calculating both overall and zonal factor weights, the Plateau was divided into eight distinct environmental zones: Zone I (Alpine Valley Freeze–Thaw Tectonic Zone), Zone II (Alpine Canyon Tectonic-Erosion Zone), Zone III (Alpine Freeze–Thaw Tectonic Active Zone), Zone IV (Faulted Basin Weathered Fragmentation Zone), Zone V (Permafrost-Weathered Rock Belt), Zone VI (Permafrost-Nival Erosion Zone), Zone VII (Alpine Hard Rock Freeze–Thaw Debris Zone), and Zone VIII (Alpine Lake Basin Saline Soft Rock Zone).

(2)

The spatial distribution of CLD hazards on the Qinghai–Tibet Plateau exhibits strong heterogeneity, shaped by its complex geomorphology, tectonic structures, and hydrometeorological conditions. Zones I–III account for 84.32% of all recorded hazard events, with high densities ranging from 1.04 to 1.21 events per 100 km². Specifically, Zone I features tectonic fragmentation and significant topographic relief, resulting in 910 collapse events and 3960 debris flows; Zone II is characterized by deeply incised valleys and monsoonal precipitation, leading to 1739 landslides and numerous debris flows; and Zone III is marked by moderate hazard density, primarily triggered by snowmelt and localized rainfall. By contrast, Zone IV hosts a moderate concentration of debris flows (0.30 sites/100 km², 632 events), while Zones V–VIII, characterized by flat terrain and limited material availability, exhibit disaster densities generally below 0.30 sites/100 km². The dominant hazard types also vary by zone: collapses are concentrated in tectonically active areas (75.65%), landslides are most prevalent in deeply incised valleys (50.38%), and debris flows dominate in regions with concentrated rainfall (53.05%), highlighting the compounded effects of geology, geomorphology, and climate.

(3)

CLD hazards are predominantly small- to medium-scale (87.99%), whereas large and extremely large events (1319 sites) are highly concentrated in Zones I–III (94.69%), with Zone II accounting for 48.67% and Zone I for 28.96%. Among moderate to large-scale events, debris flows account for the largest proportion (medium: 18.74%; large: 5.58%; extremely large: 0.51%), while collapses (15.78%) and landslides (8.73%) mainly occur at smaller scales. Hazard formation mechanisms show marked regional differentiation: Zone I is primarily driven by tectonic fragmentation (30.884%) and precipitation (26.482%); Zone II is influenced by tectonic activity (30.531%), geomorphology (27.034%), and precipitation (25.551%); Zone III is controlled by precipitation (33.134%), freeze–thaw-induced lithological degradation, and active tectonics; Zone IV is affected by weathered lithology (27.454%) and snowmelt-induced erosion; Zone V exhibits debris accumulation from freeze–thaw cycles and glacial melt erosion; and Zones VI–VIII are primarily governed by engineering geological factors (23.353–42.077%) and precipitation under low-relief terrain and limited material supply. Overall, CLD hazard formation on the Plateau follows a multifactorial coupling mechanism involving geological structure, lithology, geomorphology, and climate, where geological conditions provide the material foundation, and precipitation and terrain serve as key dynamic triggers, collectively shaping the spatial heterogeneity of CLD hazard distribution.

(4)

Based on the delineated zones and their dominant driving mechanisms, this study proposes differentiated regional risk management strategies. For areas with high tectonic activity, such as Zones I and II, hazard mitigation should focus on infrastructure and site selection, stability monitoring, and construction of sediment control systems. In precipitation-prone zones, early warning systems and dynamic hazard response protocols are recommended. Regions dominated by freeze–thaw and weathering processes should prioritize surface material management, ecological restoration, and stabilization of loose deposits. This approach supports a transition from general risk identification to zone-specific targeted disaster management. The proposed strategies provide both theoretical insights and practical guidance for enhancing geological hazard resilience under the complex environmental conditions of the Qinghai–Tibet Plateau.