KR100982447B1 - Landslide occurrence prediction system and predicting method using the same - Google Patents

Abstract

PURPOSE: A landslide occurrence prediction system and a predicting method using the same are provided to accurately occurrence estimation of a landslide through more than one geospatial correlation. CONSTITUTION: A landslide geological information database(100) provides the geological information of a site where landslide is generated. An integrated analysis unit(200) analyzes the geospatial correlation based on the geological information of the landslide geological information database. A landslide occurrence estimation rate setting unit(300) estimates the generation of a landslide based on the geospatial correlation.

Description

Landslide occurrence prediction system using integrated geospatial correlation technique and landslide occurrence prediction method using the same {Landslide Occurrence Prediction System and Predicting Method using the same}

The present invention relates to a prediction system for predicting landslide occurrence in a specific region and a landslide prediction method using the same as a result of the research project "GIS-based national geographic information system practical technology development" hosted by the Ministry of Knowledge Economy.

Due to global climate change, typhoons and local heavy rains are increasing in Korea. Landslides and forest damage increased sharply due to the impact of typhoon Jeni in 1998, lusa in 2002, and cicada in 2003. In 2006, numerous landslides were caused by typhoons Ewinia, Billis and ants. In 2006, more than twice the amount of rainfall (758 mm) was recorded, compared to 1,500 billion won in property damage nationwide in July (Water Journal, 2007). In Gangwon-do, 675 mm of heavy rain fell over 2 days, accounting for 95.4% of the total damage, and numerous landslides occurred in Inje-gun and Pyeongchang-gun.

As such, landslides are one of the geological phenomena that damage human life and infrastructure. Because of their irregularities within the Earth system, predictive research is needed.

With the development of advanced technologies and equipment, real-world geospatial data associated with geological phenomena are becoming increasingly large and complex. Therefore, Geographic Information System (GIS) is needed to effectively find information related to geological phenomena and correlate with geospatial elements with complex characteristics.

  1 is a conceptual diagram illustrating a general geographic information system (GIS). Referring to this, the GIS is a powerful tool for efficient storage, management, and integrated analysis of massive geospatial data, and serves to support expert decision making. However, GIS lacks the ability to predict the spatial relevance related to the occurrence of events or to predict future occurrences inherent in large multidimensional data. Therefore, various fields related to geospatial use probabilities, statistics, and pattern recognition techniques using GIS to efficiently find potential information, spatial correlations and patterns in the spatial database, and predict future occurrences of events.

  In other words, the existing various analytical methods were to calculate correlation coefficients or weights through the primary analysis between dependent variables (event data) and independent variables (input data), and to prepare event occurrence predictions using them. Verification is performed to determine the predictive accuracy of the results obtained through each technique. However, even if the result shows the maximum prediction accuracy, there is a limit to the accuracy improvement depending on the type and characteristic of the technique.

Therefore, there is a limit in increasing the accuracy of the landslide geological phenomenon, which leads to problems such as preventive measures or development plans based on accurate predictions of the geological phenomenon in mountainous areas.

The present invention has been made to solve the above problems, the object of the present invention is to use likelihood ratio analysis, WOE analysis, logistic regression using the location information of landslide occurrence and data such as terrain, geology, clinical, soil, land use The present invention provides a prediction system capable of performing accurate landslide occurrence prediction analysis by allowing geospatial correlation analysis to be performed at least once using an analysis unit capable of performing analysis and neural network analysis.

As a means for performing the above-described problems, the present invention is a landslide occurrence prediction system, landslide geological information database for providing geological information such as landslide occurrence location, topography, geology, clinical, soil, land use; A geospatial correlation integrated analysis unit for integrated analysis of geospatial correlations based on factors provided by the landslide geological information database; A landslide occurrence prediction map generation unit providing a landslide occurrence prediction map and an integrated prediction map based on the information analyzed by the geospatial correlation integration analysis unit; A verification unit quantitatively verifying the accuracy of the landslide occurrence prediction map using the landslide position; It is possible to provide a landslide occurrence prediction system using a geospatial correlation integrated technique, characterized in that it comprises a ;;

In particular, the landslide location information extracted from aerial photographs is a geological disaster database of a landslide location as a geological disaster of a research site; A terrain database in which spatial factors such as slope, slope, curvature, Topographic Wetness Index (TWI), Stream Power Index (SPI), etc. extracted from the topographic map of the region to be analyzed are spatialized; A geological database in which spatial factors such as geology and faults extracted from geological maps of the target area are spatialized; Clinical database of spatial database of factors such as salary, clinical, small density, and salary, extracted from the clinical map of the analysis area; A soil database in which spatial factors, such as topography, drainage, base material, effective soil, and soil, extracted from soil maps of the analysis area are databased; Soil use database extracted from the land use of the analysis area; It may comprise any one or more of.

In addition, in the system according to the present invention, the geospatial correlation integrated analysis unit, landslide occurrence location in the landslide geological information database that provides all the factors related to the landslide as a dependent variable, and topography, soil, clinical, geology and land A likelihood ratio analysis unit for predicting the probability of landslides by analyzing correlations between factors related to the observed landslide using data such as utilization as independent variables; A weight of evidence (WOE) analysis unit for analyzing weights of grades between landslides and related factors; A logistic regression analysis unit for analyzing the logistic regression correlation coefficients of the independent data and the dependent data as input data; The landslide occurrence area and the landslide-free area may include any one or more of the artificial neural network analysis unit for analyzing the weight of each factor through the backpropagation algorithm.

In particular, the geospatial correlation integrated analysis unit, the first to n-th order prediction including any one or more of the likelihood ratio analysis unit, WOE (weight of evidence) analysis unit, logistic regression analysis unit, artificial neural network analysis unit The n-th predictive analyzing unit has the same configuration as the (n-1) th predictive analyzing unit, and includes the n-th order result value provided by the (n-1) th predictive analyzing unit. As an independent variable of the predictive analysis unit, a landslide occurrence prediction system using a geospatial correlation integrated technique, which is re-analyzed with the landslide occurrence dependent variable of the analysis region, can be constructed (where n is a natural number of 2 or more). .

In addition, by using the landslide occurrence prediction system according to the present invention it is possible to perform a prediction method for predicting the occurrence of landslides.

Specifically, using the landslide occurrence prediction system according to the present invention, the terrain, soil, clinical, geological and land use factors, which are factors related to the landslide occurrence in the analysis target region, which performs the landslide prediction prediction analysis, are constructed as a spatial database. Step 1 to do; Integrating and analyzing geospatial correlations between factors related to landslide positions constructed in the spatial database; A three step of generating a landslide prediction and integrated prediction diagram using the two steps of analysis information; A fourth step of verifying quantitative accuracy of landslide occurrence prediction maps and integrated prediction maps derived in steps 2 and 3; Landslide occurrence prediction method using the geospatial correlation integrated technique, including; 5 steps for all results derivation.

In particular, the second step, the likelihood of occurrence of landslides as a dependent variable, terrain, soil, clinical, geological and land-use data as independent variables, likelihood ratio analysis, WOE (weight of evidence) analysis, logistic regression analysis, It may be formed by analyzing any one or more of the artificial neural network analysis to predict the landslide occurrence prediction.

In addition, using the landslide prediction map derived in step 2 as an independent variable, and using the dependent variable input in step 2 again, likelihood ratio analysis, WOE (weight of evidence) analysis, logistic regression analysis, and artificial neural network analysis It is also possible to implement a landslide occurrence integrated prediction method using a geospatial correlation integrated technique characterized in that the process of analyzing the landslide occurrence prediction at least two times by performing any one or more analysis.

According to the present invention, an analysis using an analysis unit capable of performing likelihood ratio analysis, WOE analysis, logistic regression analysis, and neural network analysis using information related to landslides can be performed at least two times. It has the effect of performing predictive analysis.

1 is an exemplary diagram conceptually showing a configuration of a conventional GIS.
Figure 2a is a block diagram of a landslide occurrence prediction system using the integrated geospatial correlation method according to the present invention, Figures 2b and 2c is a configuration example of all factors related to the landslide, Figure 2d is a landslide occurrence according to the present invention It is an example of the topographic map of the area.
Figure 2e shows the result of building all the factors related to landslides into a spatial database.
2f is an image illustrating a method of extracting an existing landslide occurrence region and determining a location of landslides for landslide prediction analysis according to the present invention.
3A and 3B illustrate the configuration of the geospatial correlation integrated analysis unit of the present system. Figure 3c shows the flow of landslide prediction analysis using integrated geospatial correlation analysis, Figure 3d is a prediction of landslide occurrence using the likelihood ratio, Figures 3e-3g is likelihood ratio and weight of landslides and related factors Table showing the evidence.
FIG. 3h illustrates a landslide occurrence prediction using weight of evidence, and FIG. 3i illustrates a logistic regression correlation coefficient of landslides and related factors. Figure 3j shows a landslide occurrence prediction using logistic regression analysis.
FIG. 3K is a table illustrating artificial neural network weights of landslides and related factors, and FIG. 3L illustrates a landslide occurrence prediction map using artificial neural networks.
3m shows the combined likelihood ratio and integrated weight of evidence of landslide related factors.
Figure 3n shows the integrated neural network weight of landslide related factors (primary analysis like land likelihood ratio, WOE, logistic regression, artificial neural network analysis predicted).
Figure 4a shows a result of integrated landslide occurrence prediction (secondary analysis) using a likelihood ratio, Figure 4b shows a result of landslide occurrence integrated prediction (secondary analysis) using a weight of evidence, Figure 4c is a logistic The result of integrated landslide occurrence prediction (secondary analysis) using regression analysis is shown. 4d shows a result of integrated landslide occurrence prediction (secondary analysis) using an artificial neural network.
Fig. 5A shows the accuracy of the landslide occurrence prediction map of the first order single analysis, and Fig. 5B shows the accuracy of the landslide correlation prediction map of the geospatial correlation integration analysis, and Fig. 5C shows the first single analysis and the second order. This study compares the accuracy of landslide occurrence prediction and integrated prediction in the integrated analysis.

Hereinafter, with reference to the accompanying drawings will be described in detail the configuration and operation according to the present invention. In the description with reference to the accompanying drawings, the same components are given the same reference numerals regardless of the reference numerals, and duplicate description thereof will be omitted. The terms first, second, etc. may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another.

The present invention provides a landslide occurrence prediction system using an integrated geospatial correlation technique, and in particular, predicts a landslide occurrence by performing one or more of a likelihood ratio analysis, a WOE (weight of evidence) analysis, a logistic regression analysis, and an artificial neural network analysis. The main point is that the accuracy of the prediction can be improved by analyzing the degree or repeating the first analysis result again.

Referring to Figure 2a, which shows the configuration of a landslide occurrence prediction system (hereinafter referred to as the 'main system') using the geospatial correlation integrated technique according to the present invention.

The present system according to the present invention integrates and analyzes the geospatial correlation based on the landslide geological information database 100 which provides information on the landslide occurrence location and related factors of the analysis region and the factors provided in the landslide geological information database. Geospatial correlation integrated analysis unit 200, and the landslide occurrence prediction map analyzed by the geospatial correlation integrated analysis unit 300 to create an integrated prediction map, landslide occurrence prediction map using the landslide occurrence position It may be configured to include a verification unit 400 for quantitatively verifying the accuracy of the integrated prediction diagram and a result derivation unit 500 for comparing the results thereof.

In particular, the landslide geological information database 100 may be formed as a database group that classifies landslide related factors as shown in FIG. 2B. Specifically, the landslide geological information database 100 may include a geological disaster database 110 in which a geological disaster and related factors of an analysis target area are made into a spatial database, and further, the slope, slope direction, and curvature of the analysis target area. Geological database (130), which is a spatial database of geographic data such as TWI (Topographic Wetness Index) and SPI (Stream Power Index) or geological and tomographic information extracted from geological maps of the target area. It may be configured to include more.

In addition, the land use database 140 that provides information on land use of the analysis region, the clinical database 150 that provides data of emergency, clinical, small density, and salary extracted from the clinical degree of the analysis region, the analysis region It may be configured to further include a soil data database 160 and the like, which provides soil data such as topography, drainage, base material, effective soil, soil.

An example of the formation of a database for landslide prediction for a specific region will be described using this system described above with reference to FIGS. 2C and 2D.

Figure 2c is a classification of the factors constituting the landslide geological information database formed in Figure 2b, Figure 2d is an example according to the present invention, showing a topographical map of the true land surface where the most landslides occur in Pyeongchang-gun, Gangwon-do, The system of the present invention will be described through an analytical embodiment of the present invention (the true surface is latitude 37˚ 35 '2˝ ~ 37˚ 40' 26˝N, longitude 128˚ 29 '49˝ ~ 128˚ 35' 36 Between ˝E).

The data collected for landslide prediction analysis in Jinbu-myeon area, which is the analysis area, are landslide distribution, topographic map, soil map, clinical map, land use and geological map. The location of landslides was identified by aerial photographs and field surveys before and after landslides. Landslide locations were categorized into training (70%, 1,261 grids) and verification (30%, 542 grids) and used for predictive analysis and prediction. Slope, slope, curvature, Topographic Wetness Index (TWI) and Stream Power Index (SPI) on topographic maps, topography, drainage, base material, effective soil and soil on soil maps, light, clinical, density and light on clinical maps. Lipids and monolayers were extracted. All factors related to landslides were constructed as a spatial database as shown in Figure 2e, and the grid size of 10m × 10m was set in consideration of the scale of input data, and the number of grids in the study area was 923 × 1,053. The number is 597,884.

2f is an image illustrating a method of extracting an existing landslide occurrence region and determining a location of landslides for landslide prediction analysis according to the present invention.

In landslide prediction analysis, existing landslide positions play an important role in understanding the environment in which landslides can occur in the future. Therefore, accurate landslide positioning can lead to optimal analysis results. The 50cm aerial photograph provided by Daum (www.daum.net) has been taken by the UltraCam-X digital camera from Sama Aerial Survey Co., Ltd; www.samah.com since 2007. The camera's image has more than 12-bit radio resolution, which provides better resolution of terrain changes than conventional analog video (8-bit) and provides high-resolution images without image distortion (Schneider and Gruber, 2008). Therefore, digital aerial photographs are more useful than satellite image and analog aerial photographs in determining the exact landslide position regardless of the landslide size.

  Therefore, this study selected the exact landslide location by comparing and detecting aerial photographs before and after landslide occurred in 2006. Before landslides, aerial photographs were issued by the National Geographic Information Institute (NGII), 1 / 20,000 scale aerial photographs (shooting date: April 4, 2005). Photographs were taken on March 3, 2008. The landslide location read by comparing the orthodontic aerial photographs was exactly matched by the actual landslide location in the field survey (FIG. 2F). 1,803 landslide coordinates read out through this work procedure were extracted and constructed as point data in ArcGIS 9.0 environment (Figure 2d). In Figure 2d, a) is an analog aerial photo before a landslide occurs, b) a digital aerial photo after a landslide occurs, and c) shows a site survey photograph.

A configuration of the present system and a method for implementing a landslide prediction map using the same will be described with reference to FIGS. 3A to 3N.

3A and 3B show the configuration of the geospatial correlation integrated analysis unit of the present system (FIG. 2A).

The geospatial correlation integrated analysis unit 200 according to the present invention specifically refers to the occurrence position of the landslide provided by the landslide geological information database as a dependent variable, and the terrain, soil, clinical, geological and land-use data as independent variables. Likelihood ratio analysis unit for predicting the probability of landslide occurrence by analyzing the correlation between the factors related to the observed landslide using A weight of evidence (WOE) analysis unit for analyzing weights of grades between landslides and related factors; A logistic regression analysis unit for analyzing the logistic regression correlation coefficients of the independent data and the dependent data as input data; The landslide occurrence area and the landslide-free area may include one or more of the artificial neural network analysis part that analyzes the weight of each factor through the backpropagation algorithm, or may further include various other model analysis parts.

Alternatively, as shown in FIG. 3B, in another preferred embodiment according to the present invention, in addition to the above-described one geospatial correlation integrated analysis unit 200, an n-th predictive analysis of the result value analyzed by the first prediction unit is performed. It may be configured as an independent variable of, and further comprises an n-th prediction unit for performing the analysis by inputting the dependent variable used in the first prediction unit as a dependent variable again. The predictive analysis unit of this structure may be formed at least twice (where n is a natural number of two or more).

For example, the first to nth order predictive analysis unit including any one or more of the likelihood ratio analysis unit, the WOE (weight of evidence) analysis unit, logistic regression analysis unit, artificial neural network analysis unit, other model analysis unit When formed, the n-th predictive analyzing unit has the same configuration as the (n-1) th predictive analyzing unit, and the result value provided by the (n-1) th predictive analyzing unit is independent of the n-th order analyzing. It can be formed into a structure in which the dependent variable input from the (n-1) prediction part is input again as the dependent variable and the analysis is performed again (n is a natural number of two or more).

Figure 3c shows a flow chart for performing a landslide occurrence predictive analysis using the air pipe correlation integrated analysis using the above-described system.

As shown, the landslide predictive analysis using this system, geospatial correlation integration uses the first analysis of landslide occurrence predicted using a variety of single techniques as a new independent variable (input data) and the existing dependent variable. (Event data) and secondary analysis. This technique can be used to predict various geological phenomena (geological, natural disasters, geological resources, environmental pollution) as well as GIS, and can be applied to predicting all events occurring in geospatial space. The primary analysis method for the prediction of lipid phenomena may be various techniques based on probability, statistics, and data mining (see FIG. 1). In the present invention, a landslide occurrence integrated predictive analysis using GIS, likelihood ratio, weight of evidence, logistic regression, artificial neural network, etc. was performed in the same procedure as shown.

In other words, first, select a research field and a research area (eg, Jinbu-myeon, Gangwon-do), and then build a spatial database using GIS. In this case, the landslide occurrence position is a dependent variable among the values provided from the spatial database, and the terrain, soil, clinical, geological, and land use data are input as independent variables to the geospatial correlation integrated analysis unit.

When the geospatial correlation integrated analysis unit is formed of a single primary predictive analysis unit as shown in FIG. 3A or an nth integrated predictive unit formed in a complex form as shown in FIG. 3B, a single predictive analysis or repetitive analysis is performed as described above. Predictive analysis will be performed, and then the accuracy of landslide prediction and integrated predictions analyzed by the likelihood ratio, weight of evidence, logistic regression, artificial neural network, and various model analysis units will be verified. .

Then, the verified landslide prediction map can be derived more accurate results through the accuracy comparison.

Hereinafter, a detailed analysis method of the geospatial correlation integrated analysis unit will be described in detail.

1) Likelihood Ratio Analysis Part-Likelihood Ratio Analysis

The application of the likelihood ratio technique reveals the correlation between the factors related to the observed landslide occurrence and predicts the landslide occurrence area by the likelihood ratio of each factor. The likelihood ratio is the ratio of landslide occurrence area by grade for each factor based on the conditional probability principle. If the likelihood ratio is greater than 1, the probability of landslide is high, and if it is less than 1, it is low. To calculate the likelihood ratio, the distances from the input data slope, TWI, SPI, and line structure were classified into 10 grades with the uniform area by grade. The calculated likelihood ratios (Figs. 3e to 3g) were assigned to the grades of each factor and then the landslide susceptibility index by likelihood ratio (LSILR) was calculated using GIS superposition analysis as shown in {Formula 1}.

{Equation 1}

Landslide prediction maps prepared using the fragility index were graded into the top 5%, 10%, 15% and 70% for visual interpretation (Figures 3d and 3d show the landslide prediction map using the likelihood ratio) First round analysis: 84% of landslides within the top 30% of the Vulnerability Index (Very high to Medium) showed a minimum of 2.82, a maximum of 25.19, an average of 17.00, and a standard deviation of 5.28.

2) WOE analysis part-weight of evidence analysis method

The weight of evidence technique takes a natural log of the likelihood ratio and presents the correlation between landslide occurrence and factor ratings with positive and negative weights. Studentized value, C / S (C), is an optimal truncation value that divides the values of factors related to landslide occurrence.It is positively weighted and (-) based on the grade having the maximum C / S (C) value. Divide by weight. If the weight has a value of zero, there is no correlation. If the weight has a value of zero, it has a negative correlation, and if it has a positive value, it shows a positive correlation. The calculated weight for each factor is as shown in FIGS. 3E to 3G.

The dichotomous weights were assigned to the grades of each factor, followed by overlapping analysis, and the Landslide Susceptibility Index by weight of evidence (LSIWOE) was obtained as shown in {Equation 2}.

{Equation 2}

Landslide prediction maps prepared using the fragility index were ranked in the top 5%, 10%, 15% and 70% for visual interpretation (FIG. 3H). Landslides in the top 30% of the Vulnerability Index (Very high Medium) were 80%, and the minimum Vulnerability Index was -8.19, the Maximum was 5.89, the Average was -1.38, and the Standard Deviation was 3.64.

3) Logistic Regression Analysis Unit-Logistic Regression Analysis

In order to accurately grasp the relationship between the independent variable (input data) and the dependent variable (event data), a regression analysis is required to obtain a regression or prediction equation representing the regularity between the two variables. Logistic regression is appropriate because landslides are classified as 0 and 1. Logistic correlation coefficient is a statistic that shows how much the dependent variable changes according to the change of the independent variable.If the value of (+) is high, the probability of landslide increases as the value of the independent variable is high, and if it is (-) The larger the value of the independent variable, the greater the probability that no landslide will occur.

The landslide occurrence regression equation was derived using the logistic regression coefficient in FIG. 3i as shown in {Formula 3}, and the probability of landslide occurrence was calculated as {Formula 4}. Coefficients multiplied by each factor are weights of factors, and landslide susceptibility index was calculated using {Equation 3} and {Equation 4}.

{Equation 3}

{Equation 4}

Landslide prediction maps prepared using the fragility index were ranked in the top 5%, 10%, 15% and 70% for visual interpretation (FIG. 3J). Landslides in the top 30% of the Vulnerability Index (Very high to Medium) were found to be 90%. The minimum Vulnerability Index was 0, the Maximum was 0.030090, the Average was 0.001956, and the Standard Deviation was 0.002272.

4) Artificial Neural Network Analysis Part-Landslide occurrence prediction using artificial neural network

The backpropagation algorithm used in the artificial neural network precisely recognizes landslides and unoccupied areas in the neural network and trains landslides based on the input data of these locations. Through this, the neural network calculates the result of the output layer, that is, the weight of landslide occurrence. In order to calculate the weight, the artificial neural network structure was set to 17 × 34 × 1, and the relative weight of the factors was calculated by setting the maximum number of repetitions to 5,000 and the learning rate to 0.01 before reaching the target error (FIG. 3k). The landslide susceptibility index for the entire study area was calculated by assigning the calculated weight to each factor.

Landslide prediction maps prepared using fragility indexes were ranked in the top 5%, 10%, 15% and 70% for visual interpretation (FIG. 3L). Landslides in the top 30% of the Vulnerability Index (Very high to Medium) were 79%, and the minimum Vulnerability Index was 0.557, the maximum was 0.9793, the average was 0.7625, and the Standard Deviation was 0.2205.

5) Geospatial Correlation Integration Method-Second Analysis

If the first analysis using the respective analysis units of 1) to 4) is defined as the first analysis, thereafter, as another embodiment according to the present invention, the prediction is repeatedly performed using the above-described first analysis result. We will explain how to analyze (this is called 'secondary analysis' for convenience).

Landslide prediction analysis using the existing single technique is carried out by drawing up prediction maps through primary analysis and verifying and comparing their accuracy. In the present embodiment, in order to integrate such geospatial correlation, the second analysis is performed by applying the predictive value derived through the first analysis as a new independent variable (input data).

를, 기법 fm에 적용하여 예측도

을 도출한다.In addition, the predictive value of the secondary analysis can be applied as a new independent variable (input data) for the third analysis. By repeating this, n-th analysis is performed using n new independent variables (input data). The geospatial correlation analysis analyzes the landslide occurrence forecasting process by (i) inputting raw data, ii) generating intermediate results or generating new input data, and v) deriving high-level results. {Equation 5} is a general formula for predictive mapping using a single method of primary analysis, which is a raw independent variable related to landslides (raw input data).

Is applied to the technique fm

To derive

{Equation 5}

은 지공간 상관관계 통합분석에서 종속변수(산사태 위치)의 새로운 독립변수(입력자료)로 사용되어, 통합기법의 {식 6}을 이용하여 통합 예측도를 작성하게 된다.

은 2차 분석 {식 6}의 독립변수(입력자료)

로 설정하고 기법 fm에 적용하여 통합예측도

을 도출한다.Prediction diagram of {Equation 5}

Is used as a new independent variable (input data) of the dependent variable (landslide location) in the geospatial correlation integration analysis, and the integrated prediction map is made using the equation (6) of the integrated technique.

Is an independent variable (input) of the secondary analysis {Equation 6}

And predictive integration using technique fm

To derive

{Equation 6}

은 n차 분석의 독립변수(입력자료)

로 설정하고, 기법 fm에 적용하여 n차 통합예측도

을 도출한다.

Is an independent variable (input data) of n-th analysis

N-th integrated prediction by applying to technique fm

To derive

{Equation 7}

The new geospatial correlation analysis analyzes the relationship between the likelihood ratio, weight of evidence, logistic regression, landslide occurrence prediction derived from artificial neural network and landslide location in the same way. The landslide integration weakness index was calculated by deriving the weight of evidence (Figure 3m), logistic regression equation (Equation 8), and the weight of artificial neural network (Figure 3n).

{Equation 8}

Landslide incidence predictions, prepared using the integrated fragility index, were ranked in the top 5%, 10%, 15% and 70% for visual interpretation (see FIG. 4).

Figure 4a shows the result of the landslide occurrence prediction (secondary analysis) using the likelihood ratio, Figure 4b shows the result of the landslide occurrence prediction (secondary analysis) using the weight of evidence, Figure 4c is a logistic Integrated landslide prediction map (regression analysis) using regression analysis is shown. 4D shows a landslide occurrence integrated prediction diagram (secondary analysis) using an artificial neural network.

In particular, the minimum value of the integrated weakness index calculated by the second analysis likelihood ratio technique is 0.01, the maximum value is 10.59, the average value is 4.00, the standard deviation is 2.92, and the integrated weakness calculated by the second analysis weight of evidence technique. The minimum value of the index is -0.80, the maximum value is 3.74, the average value is -0.35, and the standard deviation is 0.90.The minimum value of the integrated weakness index calculated by the secondary analysis logistic regression technique is 0.000004, the maximum value is 0.015317, and the average value is 0.000210. The standard deviation was 0.000252, and the minimum value of the integrated weakness index calculated by the secondary analysis artificial neural network method was 0.0391, the maximum value was 0.9473, the average value was 0.6809, and the standard deviation was 0.2698.

6) Result delivery department

Landslide susceptibility indices calculated using the integrated single- and geospatial correlation techniques are estimated values and need to be verified to know the quantitative accuracy of the prediction. For this, SRC (Success Rate Curve) and AUC (Area Under the Curve) methods were used (see FIGS. 5A and 5B). FIG. 5A shows the accuracy of the landslide prediction map of the first DAIL analysis, FIG. 5B shows the accuracy of the geospatial correlation integrated prediction map, and FIG. 5C shows the landslide occurrence of the first single analysis and the second integrated analysis. It is a comparison of the accuracy of the predictions.

The X-axis of the SRC is a graded value of the fragility index in the upper percentage, and the Y-axis is a graded value in the cumulative percentage of landslide occurrences. If the value of the X-axis is 1% and the value of the Y-axis is 100%, it means that all landslides occur within the top 1% of the landslide susceptibility index and that their predictions are correct. The AUC method was used for quantitative verification rather than SRC. This method is to calculate the area under the SRC. By multiplying the X-axis and the Y-axis by 1: 1, the area under the SRC can be obtained. The larger the area, the higher the accuracy of the prediction.

  Therefore, the accuracy of integrated predictiveness increased from 0.07% to 3.13% as a result of comparing the accuracy of the predicted and integrated predicted data between the first single analysis and the second geospatial correlation integrated analysis. Therefore, it can be seen that the secondary geospatial correlation integrated analysis method has better predictive performance than the first single analysis method.

In the foregoing detailed description of the present invention, specific examples have been described. However, various modifications are possible within the scope of the present invention. The technical idea of the present invention should not be limited to the embodiments of the present invention but should be determined by the equivalents of the claims and the claims.

100: landslide information database
110: Geological Disaster Database
120: topographic database
130: Geological Data Database
140: land use database
150: clinical data base
160: soil data database
200: geospatial correlation interleaver
210: likelihood ratio analysis unit
220: weight of evidence analysis unit
230: logistic regression analysis unit
240: artificial neural network analysis unit
300: landslide occurrence prediction drawing
400: landslide occurrence prediction map verification unit
500: result comparison section

Claims (7)

In the landslide occurrence prediction system,
A landslide geological information database providing geological information of landslide occurrence areas; A geospatial correlation integrated analysis unit for integrated analysis of geospatial correlations based on geological information provided from the landslide geological information database; A landslide occurrence prediction map generator based on the information analyzed by the geospatial correlation integrated analysis unit; A verification unit which verifies the quantitative accuracy of landslide occurrence using the landslide prediction map; And a result derivation unit for comparing all the result diagrams analyzed in the prediction system.
The landslide geological information database,
Geological disaster database which is a spatial database of the landslide distribution as geological disaster of the analysis area; and geographic data such as slope, slope direction, curvature, topographic wetness index (TWI), and stream power index (SPI) of the analysis area. Database of topographic data; Geological data database which is a spatial database of geological and fault information extracted from geological maps of the target area; A clinical data database providing data on urgent, clinical, small density, and salary derived from the clinical map of the region to be analyzed; A soil data database that provides soil data such as topography, drainage, base material, effective soil, and soil of the analyzed area; A land use database that provides information on land use in the area of analysis; Landslide occurrence prediction system using a geospatial correlation integrated method characterized in that it comprises any one or more of. delete The method according to claim 1,
The geospatial correlation integrated analysis unit,
Using the landslide occurrence location provided from the landslide geological information database as a dependent variable, and using terrain, soil, clinical, geological and land use data as independent variables,
A likelihood ratio analysis unit for predicting the probability of landslide occurrence by analyzing correlations between the factors related to the observed landslides;
A weight of evidence (WOE) analysis unit that analyzes weights of grades related to landslides;
A logistic regression analysis unit for analyzing the logistic regression correlation coefficients of the independent data and the dependent data as input data;
Artificial neural network analysis unit for analyzing the weighting factor by landslide occurrence area and non-landslide area through backpropagation algorithm;
Landslide occurrence prediction system using a geospatial correlation integrated technique comprising any one or more of the above.
The method according to claim 3,
The geospatial correlation integrated analysis unit,
Consisting of the first to n-th predictive analysis unit comprising any one or more of the likelihood ratio analysis unit, WOE (weight of evidence) analysis unit, logistic regression analysis unit, artificial neural network analysis unit;
The nth predictive analyzing unit has the same configuration as the (n-1) th predictive analyzing unit, and the resultant value provided by the (n-1) th predictive analyzing unit is an independent variable of the nth order analyzing unit, and (n) -1) Input the dependent variable input from the prediction unit as the dependent variable
Landslide occurrence prediction system using the integrated geospatial correlation method characterized in that the analysis again (n is a natural number of two or more).
A first step of constructing a spatial database of geological information of an analysis target region for performing predictive analysis of landslide occurrence;
Analyze the integration of geospatial correlations by inputting the geological information constructed in the spatial database, and analyze the likelihood ratio using the location of landslide as a dependent variable and the topographic, soil, clinical, geological and land use data as independent variables Analyzing the predictive degree of landslide occurrence by first performing at least one of a weight of evidence (WOE) analysis, a logistic regression analysis, and an artificial neural network analysis;
A third step of creating a landslide prediction map and integrated prediction map using the analysis information of the two steps; and a four step of quantitatively verifying the accuracy of the landslide prediction map and integrated prediction map of the three landslides; And 5 steps of deriving a result for comparing all the results in steps 3 and 4;
The third step,
The landslide prediction predicted by performing one or more of the likelihood ratio analysis, the weight of evidence analysis, the logistic regression analysis, and the neural network analysis are the independent variables of the second analysis. At least two processes of analyzing landslide occurrence predictions are performed by performing one or more of the likelihood ratio analysis, the WOE (weight of evidence) analysis, the logistic regression analysis, and the neural network analysis. Landslide occurrence prediction method using a geospatial correlation integrated method characterized in that to be performed more than once. delete delete

Images