Introduction

Debris flows are a special kind of torrent generated by liquid-phase water runoff, such as meltwater from glaciers and snow, entrained with a large number of solid-phase granular materials such as clay and boulders [1-4]. It often occurs in gullies or slopes with strong terrain incision. Currently, debris flow is one of the most harmful natural disasters with the largest threatened population in the world, which has caused serious human casualties and economic losses [5, 6]. Under the influence of global climate change, the frequency, intensity, duration, and influencing scope of various extreme climate events (such as heavy rainfall) are becoming increasingly severe. Coupled with frequent tectonic movements and human activities, debris flow disasters are showing an increasing trend [7-9]. Therefore, it is of great practical significance to research the risk recognition and spatial distribution law of debris flow for early prevention and control of debris flows.

At present, several methods have been developed for risk assessment of debris flows [10-13], which can mainly be summarized into three categories. The first category is the numerical simulation analysis method, commonly including FLO-2D [14, 15], Debris-2D [16, 17], PFC [18], and SPH [19, 20]. PFC and SPH are mainly used to study the initiation mechanism and motion characteristics of debris flow, while FLO-2D and Debris-2D are widely used in risk assessment and zoning of debris flow. Wang et al. accurately reconstructed the dynamics of debris flow during typhoons through comprehensive numerical calculations combining the hydrologic model (SCS-CN) with the FLO-2D model, and then predicted the vulnerability and risk values of buildings by considering factors such as debris flow intensity, building vulnerability, and building economic value [21]. Nocentini et al. carried out a dynamic model by using two dynamic codes, DAN-W and FLO-2D, to evaluate the potential debris flow hazard [22]. Cao et al. used the FLO-2D to simulate the flow situation of the main river before rainfall, assess the debris flow risk under different rainfall frequencies, identify the high-risk debris flow gully in the basin, and delimit the high-risk residential area [23]. Zhang et al. used the Rapid Mass Motion Simulation (RAMMS) to quantitatively predict the hazard of debris flows with different recurrence periods (30, 50, and 100 years) and conducted a vulnerability assessment of the Xigou debris flows in the Three Gorges Reservoir area [24]. Ding et al. conducted a scientific assessment of the risk probability of three debris flows in Yunlong County through field geological surveys and using the FLO-2D professional simulation system [25]. The second category is the mathematical models of risk assessment, such as the analytic hierarchy process (AHP) [26], fuzzy comprehensive evaluation method [27, 28], grey correlation model [29, 30], ideal point method [31, 32], neural network [33], cloud model [34]. The third category is the risk zoning of geographic information systems. Xiao et al. proposed a method for zoning the potential sudden hazards of debris flows based on deep neural networks and analyzed the risk degree of potential sudden debris flow in each region [35]. Choi et al. developed a framework for risk assessment of regional debris flow, which was applied to Chuncheon City in South Korea [36]. Zhou et al. established a risk assessment model and a vulnerability assessment model based on the extra tree model and constructed a risk assessment of a debris flow model using the contributing weight superposition method [9]. Cui and Zou established risk assessment and mapping methods based on the dynamic process of debris flows and flash floods, as well as vulnerability assessment of risk factors [11]. In addition, Chen et al. Constructed three risk management contribution models using GIS and entropy-weighted grey correlation analysis methods [13]. Chen et al. proposed a method combining a probability-based model with numerical simulations to quantify the expected and direct economic losses caused by debris flows to buildings under various rainfall intensity scenarios [37]. Li et al. constructed a road risk assessment of debris flow model that includes an improved key indicator screening and weight assigning method to assess the debris flow risk under future climate change [38]. Sun et al. proposed a comprehensive risk assessment framework of regional debris flow, which consists of three main procedures: debris flow risk analyses based on the weight-of-evidence method, physical vulnerability analyses based on 12 indices, and risk calculation considering the natural attributes of debris flow and social attributes of linear engineering [39]. Although these methods have made progress in the risk analysis of debris flow, there are some shortcomings and restricted application conditions. The numerical simulation method can obtain some information about the debris flow disaster, such as flow velocity, flow rate, and mud depth, but the mechanical parameters of the rock-soil mass are difficult to obtain, and the numerical modeling and mechanical model are simplified. It is difficult to determine the values of the evaluation index in mathematical models, and each model has its own application restrictions. For example, the AHP and fuzzy evaluation models are highly subjective, and their results are easily affected by the knowledge level or experience of experts. The grey theory and neural network need a large number of prior samples. The geographic information system is mainly used to assess the risk level in a certain area rather than a single unit, which has the disadvantages of poor accuracy and difficulty in preventing and controlling debris flows.

In the present study, a new method to evaluate the risk of debris flow gully is proposed based on attribute measure theory. 8 important factors are selected to construct the risk assessment index system, and the risk grading criteria of evaluation indices are established. The value of the evaluation index is quantified as a small range of the triangular fuzzy interval number, rather than a fixed value. The attribute measure functions and risk recognition analysis method are improved. The fuzzy theory-AHP weighting method is developed to determine the weight of the evaluation index. This method is applied to assess the risk of ten typical debris flow gullies in the Longnan region, analyze the risk levels of debris flow gullies, and compare the results calculated by the grey correlation model with the actual situations. This method can improve the accuracy of risk assessment results and provide a practical tool to assess the risk of debris flow gullies.

Material and Methods

Evaluation Index System of Debris Flow Risk

Selection of Risk Assessment Index

Whether the constructed evaluation index system is reasonable will directly affect the accuracy of risk assessment of debris flow. There are many factors affecting the risk of debris flow. If too many factors are considered, the assessment process will be complicated. Therefore, it is necessary to select important factors to construct the evaluation index system of debris flow. For this reason, we have carried out a large number of case statistical analyses and field investigations in the Longnan region [10, 27-29]. The main influencing factors and occurrence frequency of debris flow disaster are obtained, as shown in Fig. 1. In Fig. 1, I₁ is the maximum run-out volume at one time, I₂ is the watershed area, I₃ is the watershed incision density, I₄ is the main gully length, I₅ is the occurrence frequency of debris flow, I₆ is the maximum relative height difference of the watershed, I₇ is the proportion of unstable gully bed, I₈ is the maximum daily rainfall, I₉ is the bending coefficient of main gully, I₁₀ is the vegetation coverage rate, I₁₁ is the loose solid mass reserves, I₁₂ is the longitudinal slope drop of the main gully, I₁₃ is the ratio of length of sediment recharge section, I₁₄ is the population density of the watershed, I₁₅ is the slope degree, I₁₆ is the stratum lithology, I₁₇ is the distance from fault.

Based on the statistical results in Fig. 1, combined with the field investigation of debris flow gullies in the Longnan region, the maximum run-out volume at one time I₁, watershed area I₂, watershed incision density I₃, main gully length I₄, maximum relative height difference of the watershed I₆, maximum daily rainfall I₈, loose solid mass reserves I₁₁, and longitudinal slope drop of the main gully I₁₂ are selected to establish the risk assessment index system of debris flow gully, as shown in Fig. 2.

Grade Criteria for Risk Assessment Index

According to the grading criteria of geological hazards, the risk levels of debris flow are classified as slight risk (Ⅰ), medium risk (Ⅱ), high risk (Ⅲ), and extreme risk (Ⅳ). The risk level ranges from I to IV. The higher the risk level, the more serious the debris flow risk of the assessment object. On the basis of summarizing the existing achievements, the grade criteria for the risk assessment index of debris flow is proposed [10, 27-29], as shown in Table 1.

Fig. 2.

Framework structure for risk assessment of debris flow gully.

Fig. 1.

The usage frequency statistics of risk assessment factors of debris flow.

Table 1.

Grade criteria for risk assessment indices indexes of debris flow.

Evaluation index	Ⅰ (Slight risk)	Ⅱ (Medium risk)	Ⅲ (high risk)	Ⅳ (Extreme risk)
Maximum run-out volume at one time I1 (104 m3)	≤0.5	0.5~1	1~10	≥10
Watershed area I2 (km2)	<0.5	0.5~10	10~35	≥35
Watershed incision density I3 (km·km-2)	<5	5~10	10~20	≥20
Main gully length I4 (km)	<1	1~5	5~10	≥10
Maximum relative height difference of the watershed I6 (km)	≤0.2	0.2~0.5	0.5~1. 0	>1.0
Maximum daily rainfall I8 (mm)	≤25	25~50	50~100	≥100
Loose substance reserves I11 (104 m3)	≤10	10~100	100~200	≥200
Longitudinal slope drop of the main gully I12 (%)	≤5	5~15	15~30	≥30

Improved Attribute Interval Recognition Model

The improved attribute interval recognition model is enhanced from three aspects: (1) Considering the fuzziness of the assignment information of the evaluation index, the triangular fuzzy interval number [t_j^L, t_j^M, t_j^U] is used to assign the value of the evaluation index I_j instead of a fixed value. t_j^L, t_j^M, and t_j^U represent the lower limit, the most probable number, and the upper limit of the measure value of the index I_j, respectively. (2) Aiming at the complex nonlinear relationship between the value size of the evaluation index and the membership degree of risk level, the sine function is introduced to improve the traditional single-index attribute measure functions.

(3) A combination weighting method is established based on AHP and triangular fuzzy theory.

Improved Attribute Mathematical Theory

The traditional attribute synthetic assessment theory can be divided into three parts: single-index attribute measure analysis, multi-index synthetic attribute measure analysis, and attribute recognition analysis.

Let X = {x₁, x₂,…, x_n} be the evaluation object space. Any assessment object x_i has m evaluation index I_j (j = 1, 2, …, m), and its measure value is expressed by t_j, where I_j represents the jth evaluation index. Each evaluation index I_j (j = 1, 2,…, m) is divided into K continuous risk level C_k (k = 1, 2,…, K). The support probability that the measure value t_j of I_j has the attribute of risk level C_k is expressed by the single-index attribute measure function μ_ijk. The probability that the assessment object x_i has the attribute of risk level C_k is expressed by the synthetic attribute measure function μ_ik.

(1) Single-index attribute measure analysis

First of all, the classification standard of the evaluation index I_j is defined, as shown in Table 2. When a_jk satisfies a_j0<a_j1<…<a_jK or a_i0>a_i1>…>a_iK, let:

(1)

bjk=ajk−1+ajk2

Where a_jk-1 and a_jk are the lower limit and the upper limit of the evaluation index I_j (j = 1, 2,…, m) belonging to risk level C_k (k = 1, 2, …, K), respectively.

(2)

djk=min{|bjk−ajk|,|bjk+1−ajk|}

Where k = 1, 2, …, K-1. When a_j0<a_j1<…<a_jK or a_i0>a_i1>…>a_iK, the single-index attribute measure function is defined, and its expression is shown in Fig. 3. It can be seen from Fig. 3 that the single-index attribute measure value μ_ijk is linearly related to the measure value t_j of the evaluation index I_j. That is, the single-index attribute measure function is linear. The debris flow is a nonlinear complex dynamical system with multi-factors and multi-phase coupling, namely, the relationship between the attribute measure value and the measure value of the evaluation index is nonlinear. When the measure value t_j is small or big, its attribute measure value changes slowly. When the measure value is at a certain level, its attribute measure value changes rapidly. Therefore, the triangular fuzzy theory is introduced to nonlinearly improve the single-index attribute measure function.

When a_j0<a_j1<…<a_jK, the single-index attribute measure function μ_ijk(t_j) is as follows:

Table 2.

Classification standard of single evaluation index.

Evaluation index Ij	Risk level
	C1	C2	…	CK
I1	a10~a11	a11~a12	…	a1K-1~a1K
I2	a20~a21	a21~a22	…	a2K-1~a2K
…	…	…	…	…
Im	am0~am1	am1~am2	…	amK-1~amK

Fig. 3.

Traditional single-index attribute measure function.

(3)

μij1(tj)={1tj<aj1−dj112sin(32π−aj1+dj1−tj2dj1π)+12aj1−dj1≤tj≤aj1+dj10tj>aj1+dj1

(4)

μijk(tj)={0tj<ajk−1−djk−112sin(32π−tj−ajk−1+djk−12djk−1π)+12ajk−1−djk−1≤tj≤ajk−1+djk−11ajk−1+djk−1<tj<ajk−djk12sin(32π−ajk+djk−tj2djkπ)+12ajk−djk≤tj≤ajk+djk0tj>ajk+djk

(5)

μijK(tj)={0tj<ajK−1−djK−112sin(32π−tj−ajK−1+djK−12djK−1π)+12ajK−1−djK−1≤tj≤ajK−1+djK−11tj>ajK−1+djK−1

When a_j0<a_j1<…<a_jK, the single-index attribute measure function μ_ijk(t_j) is as follows:

(6)

μij1(tj)={0tj<aj1−dj112sin(32π−tj−aj1+dj12dj1π)+12aj1−dj1≤tj≤aj1+dj11tj>aj1+dj1

(7)

μijk(tj)={0tj<ajk−djk12sin(32π−tj−ajk+djk2djkπ)+12ajk−djk≤tj≤ajk+djk1ajk+djk<tj<ajk−1−djk−112sin(32π−ajk−1+djk−1−tj2djk−1π)+12ajk−1−djk−1≤tj≤ajk−1+djk−10tj>ajk−1+djk−1

(8)

μijK(tj)={1tj<ajK−1−djK−112sin(32π−ajK−1+djK−1−tj2djK−1π)+12ajK−1−djK−1≤tj≤ajK−1+djK−10tj>ajK−1+djK−1

Where j = 1, 2, …, m; k = 1, 2, …, K.

Considering the complexity of the hazard-pregnant environment and the uncertainty of the values of the evaluation index, a small range interval [t_j^L, t_j^M, t_j^U] is used to quantify the measure value t_j of the evaluation index I_j, which t_j^L, t_j^M, and t_j^U represent the lower limit, the most probable number, and the upper limit of the measure value of the index I_j, respectively.

The measured lower limit t_j^L, the most probable number t_j^M, and upper limit t_j^U are substituted into Eq. (3)~Eq. (8). Three single-index attribute measure vectors can be obtained. Their expressions are defined as follows:

(9)

{μjL=(μj1L,μj2L,⋯,μjKL)μjM=(μj1M,μj2M⋯,μjKM)μjU=(μj1U,μj2U,⋯,μjKU)

(2) Multi-index synthetic attribute measure analysis

Since each evaluation index has three single-index attribute measure values μ_j^L, μ_j^M, and μ_j^U. The multi-index synthetic attribute measure has 3^m combinations. The calculation formula of the multi-index synthetic attribute measure value is as follows:

(10)

μk=∑j=1m(ωjμjk)

Where w_j is the weight of the index I_j, which satisfies 0≤w_j≤1, and ∑i=1nwi=1.

(3) Attribute recognition analysis

The confidence criterion is introduced to identify the risk level of the assessment object, and the confidence degree λ is between 0.6 and 0.7. Assuming that the risk level C_k are arranged in order, the risk level of the object to be evaluated is determined as follows:

When C₁<C₂<…<C_K, if Eq.(11) is satisfied, the risk level of the assessment object is C_k0.

(11)

k0=max{k:∑l=kKuxl≥λ,1≤k≤K}

When C₁>C₂>…>C_K, if Eq.(12) is satisfied, the risk level of the assessment object is C_k0.

(12)

k0=min{k:∑l=1kuxl≥λ,1≤k≤K}

Aiming at the improved attribute interval risk recognition method, two calculation methods of risk level recognition are proposed based on the traditional attribute recognition analysis.

• a) Probability analysis method

Firstly, the synthetic attribute measure vector of each combination is obtained by using Eq. (10). Then, the risk level can be determined based on the confidence criteria. Each synthetic attribute measure vector corresponds to a risk level C_k0, and a total of 3^m C_k0 can be obtained. Finally, the number of risk levels C_k (k=1, 2,…, K) is counted, and its proportion is calculated. The proportion of the most probable risk level is the largest.

• b) Weighted average method

This method performs the weighted average on μ_jk^L, μ_jk^M, and μ_jk^U, to obtain the synthetic attribute measure vector. Then, the most probable risk level can be determined based on the attribute measure analysis. The calculation formula of the weighted average is as follows:

(13)

μjk=αμjkL+βμjkM+γμjkU3

In Eq. (13), α, β, and γ are the distribution coefficients of μ_jk^L, μ_jk^M, and μ_jk^U respectively, which can satisfy α + β + γ = 1, β>α>0, and β>γ>0. Their specific values can be determined by experts according to the actual situation.

According to the grade criteria in Table 2, the ingle-index attribute functions for risk assessment of debris flow are constructed, as shown in Table 3.

Triangular Fuzzy Theory-Analytic Hierarchy Process

The weights of the evaluation index have a direct impact on the accuracy and rationality of the assessment results, which are crucial to the risk level of the assessment object. To give full play to the subjective initiative of experts and the fuzziness and uncertainty of the influencing factors, the triangular fuzzy theory (TFT) is introduced to improve the analytic hierarchy process (AHP), and a new weighting method is established to determine the weight of the evaluation index.

A triangular fuzzy number R_ij = (r^l_ij, r^m_ij, r^u_ij) is used to quantify the relative importance between the index I_i (i = 1, 2, …, m) and the index I_j (j = 1, 2, …, m), where r^l, r^m, and r^u represent the lower limit, intermediate value, and upper limit of the relative importance between the indices, respectively. The values of these numbers are determined according to the 1~9 scale method as shown in Table 4.

An n-order judgment matrix can be formed from R_ij(i, j = 1, ..., n), namely R = (R_ij)_n×n. To test the consistency of the judgment matrix R, the triangular fuzzy opinions should be aggregated first.

(14)

mij=rijL+2rijM+rijU4 (i,j=1,2,⋯,n)

(15)

M=(mij)n×n

The weights of each evaluation index can be calculated based on the judgment matrix M. For the specific solution process, please refer to references [40] and [41]. The subjective weight vector W^s is expressed as:

(16)

W=(w1,w2,⋯,wn)

Table 3.

single-index attribute measure functions of debris flow assessment.

Function	Ⅰ	Ⅱ	Ⅲ	Ⅳ
I1	μ11(t1)={1t1<0.2512sin(2t1π)+120.25≤t1≤0.750t1>0.75	μ12(t1)={0t1<0.2512sin(2π−2t1π)+120.25≤t1≤0.7512sin(2t1−1)π+120.75≤t1≤1.250t1>1.25	μ13(t1)={0t1<0.7512sin(3π−2t1π)+120.75≤t1≤1.2511.25≤t1≤5.512sin(32π−14.5−t19π)π+125.5≤t1≤14.50t1>14.5	μ14(t1)={0t1<5.512sin(32π−t1−5.59π)+125.5≤t1≤14.51t1>14.5
I2	μ21(t2)={1t2<0.2512sin(2t2π)+120.25≤t2≤0.750t2>0.75	μ22(t2)={0t2<0.2512sin(2π−2t2π)+120.25≤t2≤0.7510.75<t2<5.2512sin(32π−14.75−t29.5π)+125.25≤t2≤14.750t2>14.75	μ23(t2)={0t2<5.2512sin(32π−t2−5.259.5π)+125.25≤tj≤14.75114.75<tj<27.512sin(32π−42.5−tj15π)+1227.5≤tj≤42.50tj>42.5	μ24(t2)={0t2<27.512sin(32π−t2−27.515π)+1227.5≤t2≤42.51t2>42.5
I3	μ31(t3)={1t3<512sin(t310π)+125≤t3≤150t3>15	μ32(t3)={0t3<512sin(2π−t310π)+125≤t3≤15115<t3<5512sin(32π−145−t390π)+1255≤t3≤1450t3>145	μ33(t3)={0t3<5512sin(32π−t3−5590π)+1255≤t3≤1451145<t3<15012sin(t3100π−π)+12150≤t3≤2500t3>250	μ34(t3)={0t3<15012sin(3π−t3100π)+12150≤t3≤2501t3>250
I4	μ41(t4)={1t4<12.512sin(t425π)+1212.5≤t4≤37.50t4>37.5	μ42(t4)={0t4<12.512sin(2π−t425π)+1212.5≤t4≤37.512sin(t425π−π)+1237.5≤t4≤62.50t4>62.5	μ43(t4)={0t4<37.512sin(3π−t425π)+1237.5≤t4≤62.5162.5<t4<7512sin(t450π−π)+1275≤t4≤1250t4>125	μ44(t4)={0t4<7512sin(3π−t450π)+1275≤t4≤1251t4>125
I5	μ51(t5)={1t5<2.512sin(t55π)+122.5≤t5≤7.50t5>7.5	μ52(t5)={0t5<2.512sin(2π−t55π)+122.5≤t5≤7.512sin(t55π−π)+127.5≤t5≤12.50t5>12.5	μ53(t5)={0t5<7.512sin(3π−t55π)+127.5≤t5≤12.5112.5<t5<1512sin(t510π−π)+1215≤t5≤250t5>25	μ54(t5)={0t5<1512sin(3π−t510π)+1215≤t5≤251t5>25
I6	μ61(t6)={1t6<0.512sin(t6π)+120.5≤t6≤1.50t6>1.5	μ62(t6)={0t6<0.512sin(2π−t6π)+120.5≤t6≤1.511.5<t6<312sin(t6−14π)+123≤t6≤70t6>7	μ63(t6)={0t6<312sin(9−t64π)+123≤t6≤717<t6<7.512sin(t65π−π)+127.5≤t6≤12.50t6>12.5	μ64(t6)={0t6<7.512sin(3π−t65π)+127.5≤t6≤12.51t6>12.5
I7	μ71(t7)={1t7<0.112sin(5t7π)+120.1≤t7≤0.30t7>0.3	μ72(t7)={0t7<0.112sin(2π−5t7π)+120.1≤t7≤0.310.3<t7<0.3512sin(32π−0.65−t70.3π)+120.35≤t7≤0.650t7>0.65	μ73(t7)={0t7<0.3512sin(32π−t7−0.350.3π)+120.35≤t7≤0.6510.65<t7<0.7512sin(2t7π−π)+120.75≤t7≤1.250t7>1.25	μ74(t7)={0t7<0.7512sin(3π−2t7π)+120.75≤t7≤1.251t7>1.25
I8	μ81(t8)={1t8<2.512sin(t85π)+122.5≤t8≤7.50t8>7.5	μ82(t8)={0t8<2.512sin(2π−t85π)+122.5≤t8≤7.517.5<t8<1012sin(t810π−12π)+1210≤t8≤200t8>20	μ83(t8)={0t8<1012sin(12π−t810π)+1210≤t8≤20120<t8<22.512sin(t815π−π)+1222.5≤t8≤37.50t8>37.5	μ84(t8)={0t8<22.512sin(3π−t815π)+1222.5≤t8≤37.51t8>37.5

Table 4.

The 1~9 scale method of factors comparison.

Scale value	Relative important explanation	Complementary value
1	Two factors contribute equally to the objective	1
3	One factor is slightly more important than the other	1/3
5	One factor is obviously more important than the other	1/5
7	One factor is strongly more important than the other	1/7
9	One factor is absolutely more important than the other	1/9
2, 4, 6, 8	Compromise value between two adjacent scales	1/2,1/4,1/6,1/8

Where w_i (i = 1, 2, …, n) is the weight value of the i-th evaluation index. Then, the consistency test of the initial weight of the evaluation index is carried out. The specific calculation steps refer to references [40] and [41].

Results and Discussion

Weighting Determination

Based on the 1~9 scale in Table 2, the triangular fuzzy judgment matrix is constructed, shown in Table 6. The weights of the evaluation index of debris flow are calculated.

Fig. 4.

Distribution map of debris flow gullies in the study region.

Conclusions

(1) According to the statistical analysis of the risk assessment factors, combined with the field investigation in Longnan region, the maximum outflow volume at one time I₁, watershed area I₂, watershed incision density I₃, main gully length I₄, maximum relative height difference of the watershed I₆, maximum daily rainfall I₈, loose solid mass reserves I₁₁, longitudinal slope drop of the main gully I₁₂ are selected to construct the evaluation index system of debris flow risk. The grade criteria of the debris flow evaluation index are proposed.

(2) A new method for comprehensive risk assessment of debris flow based on the fuzzy attribute interval measure model and TFT-AHP weighting method is proposed to predict the risk of debris flow. A triangular fuzzy number [t_j^L, t_j^M, t_j^U] is used to quantify the value of the evaluation index, including the lower limit t_j^L, the most probable value t_j^M, and the upper limit t_j^U. The traditional single-index attribute measure functions are improved to characterize the nonlinear relationship between disaster occurrence and influencing factors. The multi-index measure function and attribute recognition analysis method based on a random combination of multi-index attribute measures are developed. The combination weighting method based on the triangular fuzzy theory and AHP is established.

(3) The proposed method is applied to assess the risk of ten debris flow gullies in the Longnan region. The assessment results are compared with the results obtained by the grey correlation model and the actual field investigation situations. It is found that the proposed method is more in line with the actual situations, more applicable, and clearer in distinguishing the risk level, which provides an effective way to accurately evaluate the risk of debris flow.

The evaluation index of debris flow in this paper does not consider the differences between regions, and the evaluation index system will be revised in the future by collecting the influencing factors of debris flow in the Longnan region. In addition, it is difficult to determine the values of the evaluation index. In the next step, the values of evaluation indices can be determined using satellite remote sensing data.