Improved grey clustering method in risk zonation of mountain flash flood disaster
Abstract
Flash floods are considered one of the worst weather-related natural disasters. Flash floods are dangerous because they are sudden and highly unpredictable. Identification of the locations of high-risk areas has a major effect on the improvement of flash flood disaster control and prevention. Earlier work conducted on flood disaster risk zonation was commonly based on Digital Elevation Mode (DEM) data and statistical yearbook data and used an index, such as rainfall, topography, slope, or river distribution, with the analytic hierarchy process (AHP) method to determine the weighting. In this method, the final regional risk map was created by using ArcGIS map algebra superposition. In the present study, an improved gray clustering method is put forward to improve the comprehensive evaluation of the risk of mountain flash flood disasters by constructing the exponential whitening function and by using the information entropy weight method, which produces results that are more accurate and more reliable than those of the traditional method. This improved method can make full use of the limited information available, improving not only the resolution but also the influence of the subjective method, and produces more objective and accurate evaluation results. We obtain the risk degree by combining the information entropy weight and improved whitening function approaches in a gray clustering methodology. Additionally, a method is applied to develop models for mapping the risk grade in zones of 1436 towns and counties in Hubei Province with remotely sensed (RS) data and the ArcGIS platform. The results show that the improved approach is useful for rapidly assessing flash flood hazard and vulnerability and for completing risk assessments in mountain areas.
Keywords
Download Options
Introduction
Flash flooding is one of the major natural disasters that may hamper human development in flash flood areas. A mountain flash flood disaster is one of the most serious natural disasters. Mountain flash flood disasters occur suddenly, are considerably destructive, have short durations and cause serious harmed in the form of many casualties and considerable property loss. China is one of the countries that seriously suffer from mountain flash floods. Jonkman (2005) studied flash flood data from 1975 to 2002 and found that flash flood mortality is higher than that for other natural hazards. The potential for flash flood casualties and damages is also increasing in many regions due to the social and economic development which imply pressure on land use. Consequently, the flash flood hazard is expected to increase infrequency and severity because of the impacts of global change on climate, severe weather in the form of heavy rains and river discharge conditions. Therefore, the management of flash flood risks is a critical component of public safety and quality of life.
As one of the important and fundamental steps in flood regionalization, flood risk evaluation has general public concern. Many achievements in flood risk assessment research have been realized. Currently, the main methods of evaluating and locating flood disaster risk include geomorphologic methods (Haruyama et al. 1996), hydrology-hydraulic models, system simulation methods (Solaimani et al. 2005; Elawad et al. 2004; Smemoe et al. 2004),methods based on historical disaster data(Liu and Shi 2001;Huang et.al) and ancient flood data(Bonito et al. 1998; Bonito et al. 2003; Bonito et al. 2004), methods based on remote sensing and GIS (Sanyal and Lu 2006), and machine system analysis methods (Tang and Zhu 2005;Li et al. 2005). However, fairly few research papers have focused on mountain flash flood zonation. Even rarer are integrated analyses of flood risk that comprehensively consider flood formation mechanisms, climate, geomorphology, river water systems, and historical flood data. The research conducted in this paper is focused on comprehensive risk assessment and zonation, considering the important factors affecting flash floods: precipitation, topography, water system, vegetation, GDP, population, cultivated land and flood control capacity.
In fact, the main characteristics of natural disaster systems are the uncertainty and complexity of the system. Determination of the weights of the indicators is a problem because of the wide range of both natural uncertainty and approaches. Some in myriad sample can be solved by probability and statistics ways, and some in kenning uncertainty can be dealt with by fuzzy mathematics. However, there also exists another category on uncertainty in less data and little sample, incomplete information and devoid of experience, which is suitable to be dealt with only by gray system theory (Deng,2005). In general terms, the uncertainty in less data and incomplete information is designated grayness. Thus, systems possessing grayness are said to be gray systems.
The gray theory provides the applications of clustering analysis, relational analysis, predication, and decision for the gray system (Deng, 1989). The so-called ‘‘gray” means that system information is incomplete, unclear, and uncertain. It is a useful method to address the problems of limited, deficient, and no rules available for data processing. Its analysis makes use of minor data and does not demand strict statistical procedures and inference rules. Recent studies have emphasized the importance of a comprehensive assessment of the flood risk using the gray system method (Liu, 2010). To address the problem of nonadjacent domain weighted superposition failure caused by the traditional gray clustering whitening function, this paper proposes a whitening function construction method based on exponential distribution, avoiding the condition of a zero weighting, and discusses on the steps of flood risk assessment in detail. In view of the complexity of the causes and the randomness of the occurring process of the flood disaster, we proposed a comprehensive assessment by introducing the concept of information entropy into the improved gray method and constructing atypical exponential whitening weight function. Based on the above characteristic, this method can effectively solve the zero-weight problem, make full use of the simple data and largely reserve the information implied in the clustering weight by modifying the clustering weight with the values reflected by the entropy. Finally, with data from Hubei Province, we graded the risk evaluation of 1436 towns in Hubei Province, which verifies the validity and objectivity of the method described in the paper. This illustrative example verifies that this method is simple and reasonable and can extend application range of the gray clustering in flash flood zonation.
Conclusion
In the risk assessment of mountain floods in Hubei Province, the villages and towns are taken as the evaluation units in the study area, and the geometric calculation function is used to calculate the area proportion of different risk grades in ArcGIS, as shown in Table 6.
TABLE 6 AREA PROPORTION OF DIFFERENT RISK GRADES risk grade area（km2） area proportion（%） Very low 43880.75 23.23% low 16089.42 8.52% medium 60416.60 31.98% high 36401.06 19.27% Very high 32123.65 17.00% Based on the hazard map in Fig. 4, some of the southeastern regions (such as Luotian County, Xishui County, Tuanfeng County, Daye City, Tongcheng County, and Tongshan County), western regions (such as Yunxi County, Yun County, Danjiangkou City, Gucheng County, Enshi and other areas) and northern regions (such as Zaoyang City and Suizhou County) of Hubei Province pose the highest degree of flash flood hazard.
However, based on the hazard map in Fig. 5, some of the southern regions (such as Zhijiang City, Yidu City, Songzi City and Wujiagang District), northern regions (such as Laohekou City and Danjiangkou City) and eastern regions (such as Anlu City, Huangpi District, Xinzhou District, Huarong District and Jingshan County) of Hubei Province are most vulnerable to flash flood.
Based on the risk map that considers both the hazard and the vulnerability, shown in Fig. 3, Zouma Town, Sha Road Town, Five Peak Town, Xintang County, Flower County and 209 other towns and counties are exposed to the highest flash flood risk level in Hubei Province. Among the highest risk areas, Ma Zhen occupies the largest area, 1417.61 km2, and accounts for 4.5% of the total very high-risk area.
Among 1436 towns and counties in Hubei Province, 196 towns are exposed to the low and very low flash flood risk. Juwan Town has the highest proportion of area in the low and very low flash flood risk. There are 407 towns, such as Tianjia Town, with a moderate flash flood risk, 624 towns with a high flash flood risk, and 209 towns with a very high flash flood risk. The method presented in this paper improves the traditional gray clustering method with the following changes: constructing an exponential whitening function, effectively overcoming the shortcomings of the traditional gray clustering method that considers the defects between only adjacent grades in the linear whitening function, broadening the coverage of the whitening function and greatly improving the utilization of the available information. Additionally, the entropy-based method for determining the clustering weight can avoid subjective influences, such as the AHP method, and produce more reasonable and objective weights and evaluations. Lastly, this research is useful for identifying the regions that are threatened by the highest risk and can be easily applied to flash flood zonation for disaster assessments.
In this paper, the issues of the existing gray clustering method are analyzed. To overcome the shortcoming of the existing gray clustering method and the method for determining weights, an improved gray clustering method that includes entropy is proposed. This improved method is used in a case study. The improved gray clustering method makes better use of characteristic information from a research database, and, compared to other methods, the improved gray clustering calculates risk levels of evaluation units more accurately and quickly. The results demonstrate that this method is simple and feasible, and the result is reasonable and accurate. It is reasonable to apply this method to the risk assessment and zonation of mountain flash floods and other disasters.
The results show that the improved approach is useful in rapidly assessing flash flood hazard and vulnerability as well as the risk assessment in mountain areas and could be adopted, with appropriate modification, elsewhere in areas with flash flooding.
Flood disaster risk assessment is very important for disaster prevention, decision-making and management, since reasonable planning and management in flood-prone areas not only reduces the flash floods loss and guarantees the safety of human lives in hilly areas but also provides disaster risk precaution information for local residents and promotes the sustained and stable development of asocial economy. Flood disaster risk assessment helps to quickly determine the prevention level and complete reinforcement measured in dangerous areas, to greatly reduce the workload and to improve work efficiency, which is important to promote flash flood relief work.