Assessing Large-Scale Flood Risks: A Multi-Source Data Approach

Abstract

Flood hazards caused by intense short-term precipitation have led to significant social and economic losses and pose serious threats to human life and property. Accurate disaster risk assessment plays a critical role in verifying disaster statistics and supporting disaster recovery and reconstruction processes. In this study, a novel Large-Scale Flood Risk Assessment Model (LS-FRAM) is proposed, incorporating the dimensions of hazard, exposure, vulnerability, and coping capacity. Multi-source heterogeneous data are utilized for evaluating the flood risks. Soil erosion modeling is incorporated into the assessment framework to better understand the interactions between flood intensity and land surface degradation. An index system comprising 12 secondary indicators is constructed and screened using Pearson correlation analysis to minimize redundancy. Subsequently, the Analytic Hierarchy Process (AHP) is utilized to determine the weights of the primary-level indicators, while the entropy weight method, Fuzzy Analytic Hierarchy Process (FAHP), and an integrated weighting approach are combined to calculate the weights of the secondary-level indicators. This model addresses the complexity of large-scale flood risk assessment and management by incorporating multiple perspectives and leveraging diverse data sources. The experimental results demonstrate that the flood risk assessment model, utilizing multi-source data, achieves an overall accuracy of 88.49%. Specifically, the proportions of areas classified as high and very high flood risk are 54.11% in Henan, 31.74% in Shaanxi, and 18.2% in Shanxi. These results provide valuable scientific support for enhancing flood control, disaster relief capabilities, and risk management in the middle and lower reaches of the Yellow River. Furthermore, they can furnish the necessary data support for post-disaster reconstruction efforts in impacted areas.

1. Introduction

In recent years, extreme climate events have exhibited a new trend of increased frequency and intensity. Short-term rainstorm floods caused by meteorological factors, such as extreme precipitation and typhoons, have increased to varying degrees, posing serious threats to people’s property and safety. According to the “2022 Global Assessment Report on Disaster Risk Reduction” [1] published by the United Nations Office for Disaster Risk Reduction, global natural disasters were complex in 2022, with 60 major floods accounting for 48% of the total frequency of natural disasters. According to the “Climate Conditions of Asia 2023” [2] report released by the World Meteorological Organization (WMO), Asia is the most disaster-prone region in the world, with over 80% of disasters related to flood and storm events, which cause the highest number of casualties and economic losses. According to the 2024 statistical data [3], heavy rainfall and flooding events were widespread globally in 2024. Between April and May alone, catastrophic rainfall occurred in regions such as South China, the United Arab Emirates and Oman, Central Asia, southern Brazil, East Africa, and South Asia. Notably, extreme rainfall events were observed in typically arid areas such as Dubai and the Taklamakan Desert, highlighting the increasingly unpredictable nature of hydrometeorological hazards. These developments underscore the growing complexity of flood formation mechanisms and the challenges posed by the short lead times available for effective flood forecasting and emergency response. Given the frequent occurrence of rainstorm-induced flood disasters, timely and accurate assessment of flood-affected areas is critical for disaster verification, emergency management, and post-disaster recovery. In this context, the integration of multi-source data has emerged as a crucial technical approach for large-scale disaster analysis, supporting both immediate emergency response and long-term disaster risk reduction efforts. Furthermore, precise flood risk assessment not only enhances emergency response efficiency but also provides a scientific foundation for recovery planning and resource allocation.

Recent advancements in data-driven flood risk mapping frameworks have demonstrated the feasibility of large-scale applications. Waleed et al. [4] developed a national-scale flood susceptibility model for Pakistan based on XGBoost and LightGBM algorithms, highlighting the scalability of machine learning methods for flood risk assessments. Similarly, Zhao et al. [5] integrated the hydrodynamic model (HiPIMS) with land use scenarios in the Luan River Basin, China, revealing that urban expansion increased flood risk by 15%. While these approaches have advanced risk mapping capabilities, limitations persist, particularly in addressing multi-dimensional risk factors.

In addition to these challenges, soil erosion—a critical process influencing surface hydrology and flood generation—has yet to be fully integrated into large-scale flood risk assessments. Soil erosion alters infiltration rates, increases surface runoff, and exacerbates sediment transport, thereby significantly affecting the intensity, duration, and spatial extent of flood events [6,7,8,9]. Previous studies [10,11,12,13] have demonstrated that soil erosion models effectively reflect hydrological changes, particularly in regions characterized by high precipitation and limited soil conservation measures. However, the integration of soil erosion dynamics into comprehensive flood risk frameworks remains limited.

The theoretical foundations for disaster risk assessment emphasize a holistic view of risk components. The Intergovernmental Panel on Climate Change (IPCC) Sixth Assessment Report (AR6) [14,15] and the Sendai Framework for Disaster Risk Reduction 2015–2030 [16] define disaster risk as the product of hazard, exposure, and vulnerability, with each element influenced by climatic and anthropogenic factors. Despite these theoretical advances, operationalizing a comprehensive multi-dimensional risk assessment framework remains a major research challenge.

Assessment methodologies have evolved from qualitative, expert-driven approaches to more quantitative, integrated methods. Traditional expert judgment methods [17] offered simplicity but suffered from subjectivity and data limitations. Probabilistic Risk Assessment (PRA) techniques [18,19] have since quantified disaster probabilities and consequences based on historical datasets; however, their effectiveness is constrained by data incompleteness and inability to predict unprecedented extreme events [20,21,22]. Recent years have witnessed a shift towards semi-quantitative and integrated approaches [23,24,25], including risk matrix methods [26,27,28,29], the Analytic Hierarchy Process (AHP) [30,31,32], and Geographic Information Systems (GIS)-based assessments [33,34,35,36], providing enhanced spatial and dynamic analysis capabilities. Nonetheless, single-method approaches often fall short in capturing the inherent complexity of disaster risks, thus promoting the rise in integrated frameworks such as multi-criteria decision analysis (MCDA) [37,38,39,40,41].

In conclusion, based on the aforementioned research background, this paper proposes a new large-scale flood disaster risk assessment framework (LS-FRAM), aimed at enhancing the accuracy of flood risk assessment in the Loess Plateau region by integrating multi-source heterogeneous remote sensing data and soil erosion models. The framework combines the Analytic Hierarchy Process (AHP) with the entropy weight method and Fuzzy Analytic Hierarchy Process (FAHP), considering multiple evaluation criteria and scientifically calculating the weight of each indicator. This allows for a comprehensive evaluation of the complex impacts of large-scale flood disasters. The contributions of this study are twofold:

(1)

It constructs a large-scale flood disaster risk assessment framework based on multi-source heterogeneous data, covering natural, social, and economic dimensions, and proposes a risk indicator system for large-scale rainstorm-type floods suitable for the Loess Plateau region.

(2)

It introduces a new flood risk assessment model (LS-FRAM) that systematically combines soil erosion models with multi-criteria decision analysis, enabling a more detailed analysis of the geographic factors influencing flood occurrence and their feedback mechanisms with soil loss, thereby improving the accuracy of disaster prediction and assessment.

2. Study Area and Multi-Source Data

The Yellow River originates from the Bayan Kara Mountains on the Qinghai–Tibet Plateau, traversing nine provinces and regions, namely Qinghai, Sichuan, Inner Mongolia, Henan, Shaanxi, and Shanxi, before ultimately reaching the Bohai Sea. The main stream spans approximately 5464 km in length and covers a drainage basin area of approximately 752,000 km². This study focuses on the middle and lower reaches of Henan, Shaanxi, and Shanxi as the primary research areas. The region’s landscape is characterized by plateaus, plains, and mountains. The Loess Plateau experiences significant soil erosion and flooding in its middle reaches due to the loose soil and sparse vegetation coverage. This is exacerbated during heavy rainstorms, leading to the transportation of substantial sediment. The sediment accumulation downstream has reached a critical level, leading to a gradual reduction in the flood discharge capacity of the river channel, resulting in the formation of an elevated riverbed. The damages resulting from floods are more severe. The middle and lower reaches are characterized by a temperate monsoon climate, which is often accompanied by frequent meteorological disasters. During the summer season, numerous rainstorms occur, primarily concentrated within the flood season. This region is among the areas in China that are most severely impacted by floods. Therefore, performing a flood risk analysis of the area holds significant reference value for enhancing future emergency and governance measures. The specific geographical location overview is depicted in Figure 1.

This study examines disaster causation factors and disaster-prone environments in the middle and lower reaches of the Yellow River. Utilizing a variety of data sources and considering geographical features, population distribution, economic conditions, and other relevant factors, the research integrates 19 data sources. Therein, slope, slope length, rainfall erosivity, soil erodibility, cover and management factors, and soil and water conservation measure factors are used to calculate the soil erosion modulus. Therefore, the experimental required 14 original indicators including metrics such as the rainstorm days, flood season precipitation, precipitation variability, soil erosion modulus, elevation variation coefficient (ECV), relief amplitude, impervious areas, drainage density, population density, GDP density, cultivated land, emergency shelter, medical institution, and road network density. The particular origins and circumstances of their data are delineated in

Table 1. Data resources and indicators for flood risk assessment.

Original Data	Indicators	Resolution	Data Resource
DEM	Slope	30 m	Geospatial Data Cloud site, Computer Network Information Center, Chinese Academy of Sciences (http://www.gscloud.cn, accessed on 1 January 2024.)
	Slope length	30 m
	ECV	30 m
	Relief amplitude	30 m
	Drainage density	30 m
Precipitation in 2010–2020	Precipitation variability	1 km	National Earth System Science Data Center, National Science and Technology Infrastructure of China (http://www.geodata.cn, accessed on 1 January 2024.)
	Rainfall Erosivity	1 km
Monthly precipitation in 2010–2020 [42]	Flood season precipitation	1 km
Soil types	Soil erodibility	1 km	Harmonized World Soil Database (https://www.fao.org/soils-portal/soil-survey/soil-maps-and-databases/harmonized-world-soil-database-v12/en/, accessed on 1 February 2024)
NDVI [43]	Cover and management factor	1 km	NASA (https://search.earthdata.nasa.gov/search, accessed on 1 February 2024)
Land utilization [44]	Cultivated land	30 m	ESRI (https://www.arcgis.com/apps/instant/media/index.html, accessed on 1 February 2024)
	Factors of soil and water conservation measures	30 m
Impervious surface [45]	Impervious areas	30 m	Zenodo (https://zenodo.org/record/5220816#.YrUCEPnraly, accessed on 1 February 2024)
Population [46]	Population density	1 km	ORNL (https://landscan.ornl.gov, accessed on 1 February 2024)
Economy [47]	GDP density	30 m	GitHub (https://github.com/thestarlab/ChinaGDP, accessed on 1 February 2024)
POI	Emergency shelter	-	Amap (https://lbs.amap.com/, accessed on 1 January 2024)
	Medical institutions	-
Road	Road network density	-	OpenStreeMap (https://www.openstreetmap.org, accessed on 1 January 2024)
Daily precipitation in 2010–2020	Rainstorm days	-	NOAA (https://www.ncei.noaa.gov/data/global-summary-of-the-day/archive/, accessed on 1 January 2024)

Note: The blue indicators represent the factors required for calculating the soil erosion modulus.

.

3. Methodology

This paper proposes a large-scale flood disaster risk assessment framework (LS-FRAM), which enhances the accuracy of flood risk assessment in the Loess Plateau region. This framework comes from four dimensions: hazard, exposure, vulnerability, and coping capacity. Multi-source heterogeneous data are integrated to achieve comprehensive risk assessment. Key indicators are selected through Pearson correlation analysis, resulting in the identification of 12 secondary-level indicators. Based on this, the AHP is utilized to determine the weights of the primary-level indicators, while the entropy weight method, FAHP, and an integrated weighting approach are combined to calculate the weights of the secondary-level indicators. The technical process is shown in Figure 2.

The large-scale flood disaster risk assessment framework developed in this study is constructed based on four key dimensions: hazard, exposure, vulnerability, and coping capacity, incorporating a total of 14 secondary indicators. Following a Pearson correlation analysis, two indicators exhibiting high correlation with the number of heavy rain days and flood season precipitation were excluded to reduce redundancy. Consequently, 12 secondary indicators were ultimately selected for analysis, as illustrated in Figure 3. Specifically, the hazard dimension includes annual precipitation variability, degree of soil erosion, and river network density. The exposure dimension comprises elevation variation coefficient, terrain ruggedness, and impervious surface proportion. The vulnerability dimension involves population density, GDP density, and proportion of arable land. Finally, the coping capacity dimension encompasses the number of medical institutions, the number of emergency shelters, and road network density. Given that the selected indicators are expressed in different units, dimensionless normalization was performed to eliminate the effects of varying scales and to achieve uniformity across all indicators. Through this process, a comprehensive and standardized framework for large-scale flood disaster risk assessment was established.

3.1. Large-Scale Flood Disaster Risk Assessment Framework

The United Nations International Strategy for Disaster Reduction (UNISDR) defines risk as the potential loss of life or property that a social system may encounter during a specific period. By combining the definition of risk in the Sendai Framework [8] and the expression of natural disaster risk, a risk assessment model for rainstorm-induced flood disasters has been established. To better reflect the overall characteristics of the indicators, weights are introduced into the model, completing the analysis of the risk in the study area. The mathematical expression is shown in Equation (1).

$$F_{risk}^{H-E-V-C}=\sigma \times F^H+\mu \times F^E+\gamma \times F^V-\kappa \times F^C$$

(1)

where $F_{risk}^{H-E-V-C}$ is the flood risk result, $F^H$ is the hazard result, $F^E$ is the exposure result, $F^V$ is the vulnerability result, $F^C$ is the coping capacity result. $\sigma , \mu , \gamma$ and $\kappa$ are the primary-level indicator weights, respectively. $F^H, F^E, F^V$, and $F^C$ represent the evaluation results for hazard, exposure, vulnerability, and coping capacity, respectively, obtained using the fuzzy comprehensive evaluation method. The computation methods for $F^E, F^V$, and $F^C$ are consistent with that of $F^H$ as illustrated in Equation (20). In the fuzzy comprehensive evaluation framework, indicator weights correspond to the secondary-level indicator weights within the risk assessment model. The weights of the secondary indicators were determined using a combination of the FAHP, the entropy weight method, and the integrated weighting approach.

(1)

Hazard

The hazard is primarily considered from two aspects: precipitation and soil erosion. Ultimately, three indicators were selected: precipitation variability, soil erosion modulus, and river network density. Therein, soil erosion is the phenomenon characterized by the degradation, removal, transportation, and deposition of soil and its parent material due to external forces such as water, wind, freezing-thawing cycles, or gravity. Soil erosion exacerbates flood disasters by increasing surface runoff, causing river channel sedimentation, elevating flood sediment content, and weakening ecological buffering capacity. The soil erosion modulus calculation is derived from the revised universal soil loss equation (RUSLE) [48,49,50,51], with its specific mathematical representation presented in Equation (2):

$$A=R_A\times K\times L\times S\times C\times P$$

(2)

where A is the soil erosion modulus, in t/(hm²·a); R_A is the rainfall erosivity factor, in MJ·mm/(hm²·h·a); K is the soil erodibility factor, in t·hm²·h(hm² MJ·mm); L is the slope length factor; S is the slope factor; C is the vegetation coverage factor; and P is the factors of soil and water conservation measures.

In addition, extreme precipitation is also a major cause of rainstorm-induced flooding. As precipitation increases, the risk of flooding also rises. Therefore, daily, monthly, and annual rainfall data were selected to calculate the number of rainstorm days from 2010 to 2020, the total rainfall during the flood season (June–September), and annual precipitation variability. Pearson correlation analysis showed a strong correlation between the number of rainstorm days and flood season precipitation. Ultimately, precipitation variability was chosen as the precipitation-related factor for analysis. The term “precipitation variability” denotes the potential range of fluctuations or oscillations in precipitation occurrences. A higher degree of variability indicates a greater likelihood of abnormal precipitation events transpiring [52]. In this study, merge and convert the daily precipitation data obtained from NOAA meteorological stations by employing the inverse distance weighting (IDW) method for interpolation.

(2)

Exposure

Exposure refers to the probability of a population, socio-economic factors, and infrastructure being exposed to flooding during a disaster. In this study, exposure is assessed using evaluation factors selected from both natural and social disaster-prone environments, including the coefficient of variation in elevation, topographic relief, and impervious surfaces. During extreme rainfall events, a large amount of surface runoff is generated, and the poor or nonexistent water infiltration capacity of impervious surfaces is one of the main causes of large-scale flooding.

(3)

Vulnerability

Vulnerability reflects the degree to which a region is adversely affected by flood disasters and its ability to mitigate their impacts, making it an important factor in assessing regional climate change adaptation. Vulnerability is influenced by factors such as population, economy, and environment. Therefore, three indicators—population density, GDP density, and the proportion of cultivated land—were selected as vulnerability factors.

(4)

Coping capacity

Coping capacity refers to a region’s ability to respond to disasters, which consists of pre-disaster prediction, emergency response during the disaster, and post-disaster recovery. The evaluation of disaster coping capacity is conducted from three aspects: emergency sheltering, transportation, and medical services. A well-developed transportation network and comprehensive emergency facilities can ensure effective evacuation and timely disaster avoidance. POI data are primarily obtained from the A map (Gaode) Open Platform and converted into batches using iterative tools in GIS. After conversion, kernel density analysis in spatial analysis is applied. Kernel density analysis utilizes a kernel function to calculate the value per unit area based on point or polyline features, fitting individual points or polylines into a smooth conical surface to compute the density of point features surrounding each output raster cell.

3.2. Determining the Indicator Weights for Large-Scale Flood Disaster Risk Assessment

In this paper, AHP is employed to compute the primary-level weights of flood risk factors [53]. Additionally, FAHP and the entropy weight method are utilized to ascertain the secondary-level weights [54,55]. The study employs a combination of subjective and objective weighting methods. This approach addresses the limitations of using a single AHP to assess the consistency of the judgment matrix and mitigates the errors commonly associated with separate subjective and objective weighting methods.

(1) The specific steps for determining the weights of the primary-level indicators using the AHP are as follows:

First, a hierarchical structure is established. Second, a pairwise comparison matrix is constructed by using a scale of 1–9 and their reciprocals to quantify the relative importance between each pair of elements. Third, the subjective weights of the indicators are calculated. Fourth, a consistency check is performed. If the consistency ratio (CR) is less than 0.1, the matrix is considered to have acceptable consistency; if CR ≥ 0.1, the consistency of the judgment matrix is deemed unsatisfactory and needs to be revised and re-evaluated.

(2) The specific steps of FAHP are outlined as follows:

(a) The scale was determined to develop a fuzzy judgment matrix, and the 0.1–0.9 scaling method was used to compare and establish a fuzzy judgment matrix pairwise r_ij, where r_ij is the importance of the i-th element relative to the j-th element within the fuzzy logic framework.

(b) The fuzzy judgment matrix was normalized to calculate the eigenvectors, eigenvalues, and weights of the matrix. For the determination of the weights, the least squares method was used through Equation (3),

$$W_{a\_i}={\displaystyle \frac{1}{n}}-{\displaystyle \frac{1}{2a}}+{\displaystyle \frac{1}{n}}{\displaystyle \sum _{j=1}^n{r}_{ij}}$$

(3)

where n is the number of indicators, i.e., the dimension of the matrix. a is a tuning parameter introduced in the least squares method to balance the error term. Where $i=1,2,3\dots n,a\ge {\displaystyle \frac{n-1}{2}}$, W_{a_i} represents the weight of the secondary indicator calculated using the Fuzzy Analytic Hierarchy Process (FAHP).

(c) A positive complementary matrix was converted to a positive reciprocal matrix, and the consistency ratio (CR) values were used for the consistency check analysis. CR < 0.1 indicates completion of the consistency check, while CR < 0.1 indicates failure. The consistency indicator (CI) was calculated through Equation (4):

$$CI={\displaystyle \frac{{\lambda }_{\max }-n}{n-1}}, {\lambda }_{\max }={\displaystyle \sum _{i=1}^n\frac{R_{ai}{W}_{a\_i}}{n{W}_{a\_i}}}$$

(4)

where λ_max is the maximum eigenvalue, n is the total number of indicators. R_aiW_{a_i} is the difference between the product of the judgment matrix and the weight vector, i is the index of different indicator factors.

The analysis indicates a positive correlation between the value and the error, suggesting that as the error increases, so does its magnitude. Therefore, in the process of inspection, the random CR calculation Equation (5) is utilized, with the necessary values being acquired through reference to

Table 2. Random CRs.

Order/t	3	4	5	6	7	8	9	10	11	12	13	14	15
RI	0.52	0.89	1.12	1.26	1.36	1.41	1.46	1.49	1.52	1.54	1.56	1.58	1.59

for random consistency.In Table 2, t represents the order of the judgment matrix. When the random CR falls below 0.1, the calculated hierarchical ranking weights are accurate and justifiable. Otherwise, the judgment matrix must be recalibrated until it meets the required standards.

$$CR={\displaystyle \frac{CI}{RI}}$$

(5)

The specific process of the entropy weight method is given as follows:

Data standardization processing involves the selection of the forward normalization Equation (6) for positive indicators and the backward normalization Equation (7) for negative indicators.

$$M_{ij}={\displaystyle \frac{X_{ij}-X_{\min }}{X_{\max }-X_{\min }}}$$

(6)

$$N_{ij}={\displaystyle \frac{X_{\max }-X_{ij}}{X_{\max }-X_{\min }}}$$

(7)

where M_ij and N_ij are the values of the indicators after forward and backward normalization, respectively. M_ij and N_ij represent the normalized values of the j-th indicator of the i-th sample after positive and negative normalization, respectively. X_ij is the original value of the j-th indicator for the i-th sample, where j = 1, 2, 3, …, n. X_max and X_min represent the maximum and minimum values among the n indicators, respectively.

The proportion of the j indicator value in sample i was calculated through Equation (8):

$$Y_{ij}={\displaystyle \frac{X_{ij}}{{\displaystyle \sum _{i=1}^m{X}_{ij}}}}$$

(8)

where m is the number of samples participating in the evaluation. X_ij is the original value of the j-th indicator for the i-th sample, while Y_ij is the proportion of the j-th indicator value in the i-th sample.

The indicator information entropy was calculated according to Equation (9).

$$g_j=-k{\displaystyle \sum _{i=1}^m\left(Y_{ij}\ln Y_{ij}\right)}, k={\displaystyle \frac{1}{\ln n}}$$

(9)

where g_j is the entropy value of the j-th indicator, with a range of [0, 1]. Y_ij is the proportion of the j-th indicator value for the i-th sample, and m is the total number of samples.

The information entropy redundancy was obtained according to Equation (10), and q_j represents the entropy redundancy of the j-th indicator.

$$q_j=1-g_j$$

(10)

The indicator weights were calculated through Equation (11):

$$W_{b\_j}={\displaystyle \frac{q_j}{{\displaystyle \sum _{j=1}^n{q}_j}}}$$

(11)

where n is the number of indicators, and W_{b_j} is the weight of the j-th secondary indicator obtained through the entropy weight method. q_j is the entropy redundancy of the j-th indicator.

The single indicator evaluation score was determined using Equation (12), and S_j is the evaluation score of the j-th indicator.

$$S_j=W_{b\_j}{X}_{ij}$$

(12)

The combination weights can be obtained based on the calculation of the subjective and objective weights. The concept of a distance function is introduced to mitigate the impact of significant fluctuations in the data. The equation for calculating the integrated weight is presented using Equation (13):

$$w_{c\_i}=\alpha w_{a\_i}+\beta w_{b\_j},\alpha +\beta =1$$

(13)

where α and β are the allocation coefficients, the sum of α and β is equal to 1. w_{a_i} and w_{b_i} represent the subjective and objective weights, respectively. w_{c_i} denotes the combined weight calculated based on both subjective and objective weights using a distance function. The results of this calculation are presented in the “integrated weight” column of

Table 3. Weights and attributes of indicators in flood risk evaluation.

First Index		S-Index				Index Attribute
Index	Weight	Index	FAHP Weight	Entropy Weight	Integrated Weight
Hazard	30.05%	Precipitation variability/% (u1)	0.2500	0.1392	22.45%	positive
		Soil erosion (u2)	0.4000	0.8137	49.50%	positive
		Drainage density/km/km2 (u3)	0.3500	0.0471	28.04%	positive
Exposure	23.69%	ECV (u4)	0.2833	0.5206	33.78%	negative
		Relief amplitude (u5)	0.3334	0.1408	28.91%	negative
		Impervious surface/% (u6)	0.3833	0.3385	37.30%	positive
Vulnerability	26.52%	Population density/person/km2 (u7)	0.3933	0.4289	40.15%	positive
		GDP density/ten thousand/km2 (u8)	0.2858	0.4553	32.48%	positive
		Cultivated land/% (u9)	0.3208	0.1159	27.38%	positive
Coping capacity	19.74%	Emergency shelter(u10)	0.2800	0.5796	34.88%	negative
		Medical institutions(u11)	0.4000	0.3105	37.94%	negative
		Road network density/km/km2 (u12)	0.3200	0.1099	27.17%	negative

.

The difference between subjective and objective weights is represented by a distance function as shown in Equation (14):

$$\left\{\begin{array}{l}D{\left(\alpha ,\beta \right)}^2=D{\left(w_{a\_i},w_{b\_j}\right)}^2\\ D{\left(w_{a\_i},w_{b\_j}\right)}^2={\left[{\displaystyle \frac{1}{2}}{\displaystyle \sum _{i=1}^n{\left(w_{a\_i}-w_{b\_j}\right)}^2}\right]}^{{\displaystyle \frac{1}{2}}}\end{array}\right.$$

(14)

Accordingly, using the sample point data as input, the calculated weights of the primary-level and secondary-level indicators in this study are presented in Table 3.

3.3. Fuzzy Comprehensive Evaluation of Large-Scale Flood Disaster Risk

Fuzzy comprehensive evaluation is a method based on fuzzy mathematics that utilizes the principle of fuzzy relation synthesis to quantify factors that are unclear and difficult to quantify, followed by a comprehensive evaluation [56,57,58]. There is a lack of a standardized and specific measurement criterion for defining and assessing flood risks, characterized by significant fuzziness. Therefore, the fuzzy comprehensive evaluation method is widely used in risk assessment. The specific evaluation process is outlined as follows:

(1)

Factor set: Based on the evaluation system established previously, the factor set was determined, and the second category indicators were used as the factor set U = {u₁, u₂, u₃, …, u_n}, n = 1, 2, 3, …, 12.

(2)

Comments collection: The assessment of risk levels was generally divided into 5 categories: very low, low, medium, high, and very high.

(3)

Weight set: The weight set used to assess the risk based on the weights calculated in the preceding section was determined as $A_H=\left\{w_{c\_1},w_{c\_2},w_{c\_3}\right\}$. The weight set for vulnerability, exposure, and coping capacity was calculated as $A_V=\left\{w_{c\_4},w_{c\_5},w_{c\_6}\right\}, A_E=\left\{w_{c\_7},w_{c\_8},w_{c\_9}\right\}, A_C=\left\{w_{c\_10},w_{c\_11},w_{c\_12}\right\}$. Where $A_H, A_E, A_V, A_C$ represents the set of secondary indicator weight vectors corresponding to hazard, exposure, vulnerability, and coping capacity, respectively.

(4)

Classification criteria and membership functions: The natural discontinuity method was employed to classify the normalized data. The expression of the membership function is given through Equations (15)–(19). Where u_n is the evaluation set constructed by the secondary indicators, where n is the secondary indicators. u_nv_l refers to the membership degree of the n-th secondary indicator to the evaluation set v₁, l = 1, 2, …, 5. Before calculating the membership degree, the dataset of each secondary indicator must be divided by the discontinuity points e_i (i = 1, 2, …, 5), which serve as the defining points for the five membership degree levels. The membership function can be expressed as follows:

①

v₁ can be determined as follows:

$$u_n{v}_1=\left\{\begin{array}{c}1,x\le e_1\\ {\displaystyle \frac{e_1-x}{e_2-e_1}},e_1<x<e_2\\ 0,x\ge e_2\end{array}\right.$$

(15)

②

v₂ can be computed as follows:

$$u_n{v}_2=\left\{\begin{array}{c} 0,x\le e_1 \mathit{or} x\ge e_3\\ {\displaystyle \frac{x-e_1}{e_2-e_1}},e_1<x<e_2\\ {\displaystyle \frac{e_3-x}{e_3-e_2}},e_2<x<e_3\\ 1,x=e_2\end{array}\right.$$

(16)

③

v₃ can be obtained as follows:

$$u_n{v}_3=\left\{\begin{array}{c} 0,x\le e_2 \mathit{or} x\ge e_4\\ {\displaystyle \frac{x-e_2}{e_3-e_2}},e_2<x<e_3\\ {\displaystyle \frac{e_4-x}{e_4-e_3}},e_3<x<e_4\\ 1,x=e_3\end{array}\right.$$

(17)

④

v₄ can be calculated as follows:

$$u_n{v}_4=\left\{\begin{array}{c} 0,x\le e_3 \mathit{or} x\ge e_5\\ {\displaystyle \frac{x-e_3}{e_4-e_3}},e_3<x<e_4\\ 1,x=e_4\\ {\displaystyle \frac{e_5-x}{e_5-e_4}},e_4<x<e_5\end{array}\right.$$

(18)

⑤

v₅ can be achieved as follows:

$$u_n{v}_5=\left\{\begin{array}{c}0,x\le e_4\\ {\displaystyle \frac{x-e_4}{e_5-e_4}},e_4<x<e_5\\ 1,x\ge e_5\end{array}\right.$$

(19)

(5)

Membership levels across different levels: The established membership function was implemented using the Con function in the grid calculator, determining 5 membership levels based on the specified level discontinuity points. The evaluation level was determined through the principle of maximum membership degree. The different membership levels of hazard, exposure, vulnerability, and coping capacity layers can be calculated using Equation (20).

$$F^{\ast }=A\circ R_q=\left(A_H,A_E,A_V,A_C\right)\circ \left(\begin{array}{cccc}{u}_1{v}_1& u_2{v}_1& \dots & u_n{v}_1\\ u_1{v}_2& u_2{v}_2& \dots & u_n{v}_2\\ ⋮& ⋮& \ddots & ⋮\\ u_1{v}_j& u_2{v}_j& \dots & u_n{v}_j\end{array}\right)=\left(b_1,b_2,\dots ,b_n\right)$$

(20)

where F is the fuzzy comprehensive evaluation result matrix, and the subscripts * represent hazard, exposure, vulnerability, and coping capacity, respectively. A is the fuzzy weight vector of the secondary indicators under hazard, exposure, vulnerability, and coping capacity. Specifically, A_H, A_E, A_V, and A_C are vectors composed of the secondary weights w_{a_i}, w_{b_i}, and α. R_q is the fuzzy relationship matrix, and u_nv_j is the membership degree of the secondary indicator un to the evaluation set v_j. b_n is the degree of membership of the rated object to the fuzzy subset of v_j level as a whole, n = 1, 2, …, 5. The symbol $\circ$ represents the fuzzy operator.

(6)

Disaster risk assessment results: Based on the fuzzy matrix calculation and membership degree layer analysis, the evaluation result matrix for the 4 primary indicators, namely hazard, exposure, vulnerability, and coping capacity, has been derived. The grid layer containing 4 primary indicator factors and 5 levels of membership degrees, along with the weight of the primary indicator, is fully determined by applying Equation (1) to finalize the calculation of the risk of rainstorm-induced flood disasters.

4. Experimental Results and Analysis

4.1. Index Weight Calculation and Risk Classification

Using AHP for determining the weights of primary indicators requires the classification of normalized secondary indicators to establish a membership function for fuzzy comprehensive evaluation. The specific calculation results are presented as follows:

(1)

The risk assessment in this model framework involves the weighted multiplication of hazard, exposure, vulnerability, and coping capacity. In Equation (1), $\sigma ,\mu ,\gamma ,\kappa$ are the weights of the primary indicators, determined through AHP. The specific calculation results are shown in

Table 4. AHP analysis results for flood risk evaluation.

Items	Eigenvector	Weight	Maximum Eigenvalue	CI
Hazard	1.202	30.050%	4.031	0.01
Exposure	0.947	23.686%
Vulnerability	1.061	26.521%
Coping capacity	0.790	19.742%

.

(2)

The normalized data for classification criteria and membership functions were classified using the natural discontinuity method, and the classification values are detailed in

Table 5. Normalized grid grading point values for flood risk indicators.

Indicators	e1	e2	e3	e4	e5
Relief amplitude	0.7255	0.8157	0.8824	0.9373	1.0000
ECV	0.0000	0.5412	0.5608	0.7333	1.0000
Precipitation variability	0.2745	0.4001	0.5529	0.7725	1.0000
Soil erosion	0.0157	0.0627	0.1843	0.4667	1.0000
Drainage density	0.3412	0.4824	0.5922	0.7216	1.0000
Impervious surface	0.0000	0.0039	0.0118	0.0314	1.0000
Population density	0.0157	0.0941	0.2745	0.6275	1.0000
GDP density	0.0078	0.0667	0.2039	0.4588	1.0000
Cultivated land	0.1200	0.3500	0.6100	0.8700	1.0000
Emergency shelter	0.2510	0.5137	0.7490	0.9216	1.0000
Medical institutions	0.4510	0.7451	0.8980	0.9686	1.0000
Road network density	0.4000	0.6157	0.7686	0.8902	1.0000

.

4.2. Extraction of Indicator Layer Elements and Comprehensive Calculation Results

The middle and lower reaches of the Yellow River flow through the Loess Plateau area, and the soil erosion is serious, which can easily cause sediment deposition on the ground during the flood season. Therefore, the model indicating the degree of soil and water erosion, namely the soil erosion mode, is added to the risk. The results are shown in Figure 4.

According to the theory of fuzzy comprehensive evaluation, 12 secondary indicators were processed and analyzed utilizing GeoScenePro 4.0 software. The results of the kernel density analysis are depicted in Figure 5j,k, while the results of factor normalization are presented in Figure 5. Furthermore, 12 secondary indicators were established with discontinuity points at different levels, and a grid calculator was employed to compute the five membership levels for each indicator, resulting in a total of 60 grid layers. After the completion of the weight set calculation, the fuzzy comprehensive evaluation models, which consist of hazard, exposure, vulnerability, and coping capacity components, are solved individually. In accordance with the principle of maximum membership degree, the fuzzy superposition tool is employed for accomplishing the level division.

4.3. Comprehensive Risk Assessment of Large-Scale Flood Disaster

This study conducts a statistical analysis of flood risk from four dimensions: hazard, exposure, vulnerability, and coping capacity. The experimental results are presented in Figure 6. By comprehensively analyzing the spatial distribution characteristics of these factors, the formation mechanisms and spatial heterogeneity of high-risk areas are systematically revealed. The results indicate that flood risk exhibits significant spatial heterogeneity, with the proportion of high-risk areas decreasing sequentially from Henan Province to Shaanxi Province and Shanxi Province. The distribution of high-risk areas highly coincides with the proportion of cultivated land and impermeable surfaces (such as in the eastern plains of Henan Province), confirming the amplifying effect of human activities on risk.

Therein, the hazard of large-scale flood disasters is one of the key factors in risk assessment. The higher-hazard areas in Henan Province are primarily concentrated in the southern region, with high-hazard areas dispersed across the province. Cities such as Zhengzhou, Jiaozuo, and Xinxiang are characterized by drainage density, high precipitation variability, and unstable rainfall, which increase the likelihood of extreme precipitation events. Xinyang, located in the Nanyang Basin, has low-lying, flat terrain and poor drainage and flood control capabilities, contributing to higher flood hazards in areas like Zhengzhou, Jiaozuo, Xinxiang, Xinyang, and Kaifeng. In contrast, the proportion of high-hazard areas in Shanxi and Shaanxi provinces is relatively lower, with the areas of greatest concern mainly located in the Yellow River Basin and the Loess Plateau.

Exposure is influenced by natural environmental factors such as topography and terrain, and it exhibits distinct spatial characteristics, as shown in Figure 6. The high-exposure areas for large-scale flood disasters are primarily located in flood-prone areas, low-lying regions, or areas with poor drainage. The elevation variation coefficient and topographic roughness significantly affect the formation of floods. In areas with higher elevations and more rugged terrain, floods are typically more concentrated, with stronger destructive power but a limited impact range due to the terrain constraints. Conversely, in low-lying areas, floods spread faster, last longer, and cause greater economic losses. Therefore, the elevation variation coefficient and topographic roughness are considered negative indicators, with higher variation and roughness indicating a lower flood disaster risk in those regions. The high exposure areas in the study region are mainly distributed in the eastern plains of Henan and the low-lying areas in the northern parts of Shanxi and Shaanxi, where the population density is high, and the proportion of impervious surfaces is also large. The landforms are primarily mountainous and hilly, making it a low-exposure area.

Vulnerability denotes the extent to which a geographical area is negatively impacted by flood disasters and its capacity to alleviate their consequences [59]. As provincial capitals, Zhengzhou, Xi’an, and Taiyuan have more concentrated populations and economies. Assuming the same level of flood disaster, areas with higher population and economic density will face greater losses, and thus their vulnerability is higher. Different land use types exhibit varying levels of sensitivity to flood disasters. As an agricultural province, Henan, with cities such as Kaifeng, Zhoukou, Jiaozuo, Puyang, and Shangqiu located in the North China Plain, has a high proportion of cultivated land and is highly agricultural. Therefore, its vulnerability is higher during flood events, and the resulting damage is more severe. Future flood disaster prevention and control efforts should focus on the treatment of vulnerable populations and the construction of flood control infrastructure in high-vulnerability areas.

Coping capacity includes predictive capabilities before the disaster, emergency rescue capabilities during the disaster, and recovery abilities after the disaster. As shown in Figure 6, Zhengzhou, Xi’an, and Taiyuan, as provincial capitals, are important transportation hubs with dense road networks and abundant medical resources, resulting in stronger overall disaster response capabilities. In contrast, economically underdeveloped regions have sparse road networks and limited medical resources, leading to lower disaster response capabilities. Cities with convenient transportation, dense road networks, and abundant medical resources can respond quickly and launch effective rescue operations during a disaster. Additionally, economically developed regions generally have more robust flood control facilities, stronger disaster warning systems, and faster post-disaster reconstruction. Therefore, in risk assessments, coping capacity is negatively correlated with disaster risk.

For better analysis, statistical and quantitative calculations of the visualization results are presented in

Table 6. Statistics results by regions.

Items	Severity Level	Shanxi		Shaanxi		Henan
		Area (km2)	Percentage (%)	Area (km2)	Percentage (%)	Area (km2)	Percentage (%)
Hazard	Very low	2667.01	1.70	3666.22	1.78	80.74	0.05
	Low	53,490.24	34.14	32,021.72	15.56	3870.08	2.32
	Medium	76,128.85	48.58	85,196.55	41.40	36,598.48	21.92
	High	23,445.10	14.96	73,519.40	35.72	106,357.71	63.69
	Very high	968.81	0.62	11,396.10	5.54	20,092.99	12.03
Exposure	Very low	25,956.39	16.56	52,056.58	25.29	26,735.01	16.01
	Low	38,967.11	24.87	40,950.65	19.90	35,108.72	21.02
	Medium	31,733.26	20.25	35,425.34	17.21	38,418.36	23.01
	High	29,278.62	18.68	34,923.59	16.97	36,888.42	22.09
	Very high	30,764.61	19.63	42,443.84	20.62	29,849.49	17.87
Vulnerability	Very low	58,097.27	37.08	10,2830.61	49.97	22,338.97	13.38
	Low	20,471.05	13.06	26,772.34	13.01	5933.84	3.55
	Medium	65,528.18	41.82	61,642.86	29.95	74,800.75	44.79
	High	5945.73	3.79	6325.28	3.07	34,453.71	20.63
	Very high	6657.77	4.25	8228.90	4.00	29,472.73	17.65
Coping capacity	Very low	11,841.65	7.56	64,872.84	31.52	12,996.47	7.78
	Low	47,292.01	30.18	65,719.83	31.93	20,033.37	12.00
	Medium	36,053.07	23.01	35,728.91	17.36	36,378.46	21.78
	High	37,681.82	24.05	24,982.11	12.14	52,348.06	31.35
	Very high	23,831.45	15.21	14,496.30	7.04	45,243.64	27.09
Flood risk	Very low	33,778.97	21.56	22,658.55	11.01	5360.03	3.21
	Low	52,316.41	33.39	61,378.43	29.82	23,819.26	14.26
	Medium	42,085.88	26.86	56,437.62	27.42	47,458.54	28.42
	High	23,003.19	14.68	43,459.82	21.12	52,733.37	31.58
	Very high	5515.56	3.52	21,865.58	10.62	37,628.81	22.53

.

The evaluation results indicate that high-risk areas in Henan, Shaanxi, and Shanxi Provinces account for 54.11%, 31.74%, and 18.2%, respectively. The proportion of high-hazard areas in Henan Province is 75.72% (high 63.69% and very high 12.03%), significantly higher than the other two provinces. Typical cities such as Zhengzhou, Xinyang are concentrated in very high-hazard areas due to the high precipitation variability and low-lying terrain. In Shaanxi Province, high areas account for 41.26%, primarily distributed in the Loess Plateau and the Yellow River Basin, which is closely related to soil erosion and high topographic relief. In Shanxi Province, high-hazard areas account for only 15.58%, but medium-risk areas account for 48.58%, indicating that while the overall risk is relatively low, localized areas still require attention. The proportions of high-exposure areas in the three provinces are 38.31%, 37.59%, and 39.96%, respectively. The high vulnerability areas in Henan, Shaanxi, and Shanxi account for 38.28%, 7.07%, and 8.04%, respectively.

According to the evaluation results, areas with low or very low coping capacity in Shanxi, Shaanxi, and Henan account for 37.74%, 63.45%, and 19.78%, respectively. Both Shanxi and Shaanxi have large areas of the Loess Plateau, characterized by significant terrain variations and numerous ravines, making transportation construction challenging, with limited options for route selection. Furthermore, the ecological environment in the Loess Plateau is fragile, and production and construction activities in these areas should prioritize ecological protection to prevent soil erosion and the destruction of ecological balance.

Shaanxi Province is located in the middle reaches of the Yellow River, with terrain following a north–south high, central low pattern. About 40% of the province’s area is on the Loess Plateau. Due to severe soil erosion, loose soil, and sparse vegetation, coupled with a temperate monsoon climate and uneven rainfall distribution concentrated in the summer, the flood risk in this province is relatively high during extreme rainfall events. High and relatively high-risk areas account for 31.74% of the province’s total area, primarily concentrated in cities such as Xi’an, Baoji, Hanzhong, and Xianyang. Shanxi Province is located on the left bank of the Yellow River in the middle reaches, with terrain dominated by the Loess Plateau, characterized by rugged terrain and intersecting valleys. Precipitation is greatly influenced by the terrain, with mountainous areas and limited plains. During heavy rainfall, water collects in the valleys and basins, leading to higher flood risks in cities along the rivers. High and relatively high-risk areas account for 18.2% of the province’s total area, mainly in cities such as Xinzhou, Changzhi, Jincheng, and Taiyuan. Henan Province has the largest proportion of high and relatively high-risk areas, reaching 54.11%. These areas are primarily distributed in economically developed and densely road-networked cities such as Zhengzhou, Xinxiang, Jiaozuo, Luoyang, and Kaifeng. As coping capacity improves, the increasing impervious areas exacerbate the urban “heat island effect,” leading to higher surface temperatures, increased surface runoff, and reduced infiltration. This also results in a decrease in vegetation coverage, further intensifying the risk to people, the economy, and infrastructure during floods. Low-risk areas are mainly located in high-altitude, steep mountainous and hilly regions with higher forest coverage and lower precipitation, where risks and vulnerabilities are generally lower. Most of Henan Province lies in high and relatively high-risk areas, which should receive increased attention in terms of flood control infrastructure and emergency rescue in the future.

4.4. Validation of Flood Risk Assessment Models

The experiment employs precision, recall, F1 score, and accuracy metrics to assess the performance of the large-scale disaster risk assessment proposed in this study. In this section, TP is the number of true positives, TN is the number of true negatives, FP is the number of false positives, and FN is the number of false negatives.

Accuracy is defined as the ratio of correctly classified samples to the total number of samples, serving as a statistical metric applicable to the entire sample set. The mathematical expression is presented in Equation (21):

$$Accuracy={\displaystyle \frac{TP+TN}{TP+FP+TN+FN}}$$

(21)

Precision, also referred to as positive predictive value, serves as an evaluative metric for predicting results. The ratio of accurately classified positive samples to the total number of samples identified as positive by the classifier. Precision is a statistical metric that pertains to certain samples, emphasizing the statistical aspects of data classified as positive by the classifier. It is defined by Equation (22):

$$Precision={\displaystyle \frac{TP}{TP+FP}}$$

(22)

The recall rate is defined as the ratio of correctly classified positive samples to the total number of true positive samples. The recall rate serves as a statistical measure for certain samples, emphasizing the statistical characteristics of actual positive samples. It is expressed as follows (23):

$$Recall={\displaystyle \frac{TP}{TP+FN}}$$

(23)

F1 score is the harmonic mean of accuracy and recall, providing a balanced evaluation that takes into account both the accuracy and recall of classification models. The statistical indicator serves as a measure for assessing the accuracy of classification models. The variable has a maximum value of one and a minimum value of 0. The model’s performance improves with higher values, which is determined through Equation (24):

$$F_1=2\times {\displaystyle \frac{Precision\times Recall}{Precision+Recall}}$$

(24)

The experiment employed the flood inundation range extraction results generated by U-Net++ [60]. Additionally, flood and non-flood points were randomly chosen using ArcGIS. A total of 642 non-flood points and 648 flood points were chosen in Henan. According to the equation provided above, accuracy was determined based on the spatial distribution of the entities within the risk zoning. The specific results are illustrated in Figure 7.

Additionally, according to the “2021 National Ten Natural Disasters Report” released by the Ministry of Emergency Management, the assessment confirmed the significant losses incurred in the disaster events impacting Henan, Shaanxi, and Shanxi [61]. The loss statistics are presented in Figure 8, indicating that the risk level zoning results align closely with the disaster situation. Based on the list of the top ten natural disasters in China in 2021 released by the Ministry of Emergency Management, this study verifies the results by focusing on disaster events that caused significant losses in Henan, Shaanxi, and Shanxi provinces. According to Section 6 of the Flood Disaster Assessment Standard (SL 579-2012) [62], five key indicators—direct economic losses, number of collapsed houses, affected population, disaster-related fatalities, and affected crop area—were selected for statistical analysis. The results show that Henan Province experienced the most severe losses among the three provinces. The distribution of disaster risk levels corresponds well with the actual disaster impacts, demonstrating the reliability of the risk zoning outcomes.

5. Discussion

This study focuses on the Loess Plateau region and establishes a large-scale flood disaster risk assessment system driven by multi-source data. A practical indicator framework is constructed from four dimensions: hazard, exposure, vulnerability, and coping capacity. Based on a hybrid weighting approach that combines subjective and objective methods, the LS-FRAM assessment model is developed, achieving flood risk zoning at a spatial resolution of 1 km. Compared with previous studies [63,64], this research differs from those based solely on remote sensing data. At the data level, it integrates multi-source heterogeneous data, including remote sensing imagery, topographic and geomorphological data, meteorological data, land use, and socioeconomic information. This approach enhances the model’s comprehensiveness and spatiotemporal adaptability, making it particularly suitable for large-scale and topographically complex areas for risk monitoring and early warning.

Secondly, unlike data-driven machine learning models [65,66], this study introduces a hybrid weighting strategy combining the Analytic Hierarchy Process (AHP), the entropy weight method, and fuzzy comprehensive evaluation. This approach reduces dependence on large volumes of historical flood samples, thereby improving the model’s applicability and generalization ability in data-scarce areas. Furthermore, the introduction of fuzzy methods better captures the uncertainties and ambiguities in multi-factor weight assignment, enhancing the credibility of the assessment results.

Additionally, this study introduces the “soil erosion modulus” as a critical disaster-forming factor for large-scale flood risk in the Loess Plateau. It emphasizes the significant impact of geomorphic evolution and slope disturbances on the disaster-forming capacity of short-term heavy rainfall events. Experimental results demonstrate that this indicator shows strong explanatory power in the Loess region. Moreover, the assessment results reveal that the proportions of high- and medium-risk zones in Henan, Shaanxi, and Shanxi provinces are 54.11%, 31.74%, and 18.2%, respectively, showing high spatial consistency with actual events such as the catastrophic rainstorm in Henan Province in 2021. This consistency validates the effectiveness and applicability of the model.

Despite establishing a relatively systematic large-scale flood risk assessment model, this study still has certain limitations. The current precipitation data have a spatial resolution of 1 km, which may not adequately capture the influence of micro-topographic features and local micro-environmental factors on flood accumulation. Future research could incorporate advanced techniques such as high-resolution remote sensing, radar-precipitation fusion, and spatiotemporal interpolation to optimize data downscaling. Additionally, future efforts will explore the use of advanced machine learning methods, such as graph neural networks and Bayesian frameworks, to adaptively model weight structures and nonlinear factor relationships, thereby further improving assessment accuracy and automation levels.

In conclusion, the proposed LS-FRAM model demonstrates notable innovations in both multi-source data integration and indicator selection. It provides a valuable technical basis for the refined identification of large-scale flood disaster risks. Nonetheless, further research is needed to enhance the model’s generalizability, standardization of indicators, and predictive capabilities for extreme events, aiming to strengthen pre-disaster risk identification and post-disaster rapid response capabilities.

6. Conclusions

This study takes the Yellow River’s middle and lower reaches as an example and conducts a comprehensive analysis based on data such as precipitation variability, soil erosion degree, population, and topography. Using a 1 km × 1 km grid cell as the risk assessment scale, the spatial distribution of rainstorm-type flood disaster risk in the middle of the Yellow River and the lower reaches was obtained. The experimental results show that the large-scale flood disaster risk assessment model driven by multi-source data has an accuracy of 88.49%. The areas of high and relatively high risk in Henan, Shaanxi, and Shanxi provinces account for 54.11%, 31.74%, and 18.2%, respectively. The high-risk areas in the study area are relatively concentrated, with Henan Province having the highest proportion of high and relatively high-risk zones, accounting for more than 50% of the province’s total area. In subsequent disaster reduction efforts, the construction of emergency facilities in medium- and high-risk areas should be prioritized to reduce environmental hazards, as well as the vulnerability of populations and the economy, and their exposure to disasters, thereby enhancing the region’s disaster reduction capabilities.

Due to global climate change and the increasing frequency of extreme weather events, global warming has led to higher water vapor content in the atmosphere, which exacerbates the risks of heavy rainfall and flood disasters. It has also caused more frequent and prolonged heat waves and droughts, leading to the melting of polar glaciers and ice caps, rising sea levels, and an increased risk of large-scale flood disasters. Therefore, this study conducts a spatial division of the risk of large-scale rainstorm-type floods, providing data support for future flood control construction and emergency rescue efforts in key areas.