Based On Spatial Analysis Of Debris Flow Disaster Forecast Model Research Of Sichuan Province

Prediction of debris flow is still a problem in the discipline of disaster controlalthough it has been investigated for almost forty years. The main reasons of this factare due to that environmental factors are distinct in different areas on the one hand, onthe other hand, the insufficiency on obtaining macroscopic regional information is alsoresponsible in this regard. Fortunately, with the recent development of related scientifictechnology, in particular the Geographic Information System (GIS) and Remote Sensing(RS) can both be used for quantitative prediction to overcome this difficulty mentionedabove. In view of the aforementioned reason, this dissertation will combine theconventional debris flow prediction theory together with the above mention twotechnologies (GIS&RS) aimed at enabling us to increase the accuracy on debris flowprediction in the recent years.In this thesis, Sichuan Province was chose as researching area, spatial interpolationmethod and prediction theory of debris flow were chose as research base. Firstly, theprediction models were established by using the data of debris flow during the period of1981-2002, precipitation data and environmental data. Then, the observational data ofdisaster during the period of2003-2004were used to examine the validity of the modelswe built. The main methods we used and corresponding conclusions are given below:(1) The precipitation data which came from meteorology stations are used forcomparing the interpolation method: IDW and OK. The results we obtaining indicatethat the OK interpolation method is more suitable for Sichuan province. The methodparameters are like this: the Exponential model is selected as the theoretical model forsemivariogram function, the number15is used as the number of Neighbors to Includeand the spatial resolution is1km*1km.(2)The logistic regression model is used to comparatively establish the predictionmodel of two kinds of rainfall combination: i) intraday rainfall and10-day previousrainfall, ii) intraday rainfall and two types of effective antecedent rainfall which areshort-time-heavy rainfall and long-time-light rainfall. The results indicate: Firstly, thelocation of debris flows and the distribution of rainfall are factors interrelated. Secondly, the contribution rate of intraday rainfall is the dominated one. Thirdly, the second formrainfall combination has a higher prediction accuracy,2.3%in short-time-heavy rainfalland2.1%in long-time-light rainfall, these data suggests that a moderate improvementachieved by the rainfall classification.(3)A grid unit based information model is selected to calculate the informationvalues of the7factors, i.e. elevation, slope, aspect, flow accumulation, vegetationcoverage, soil texture and type of land use. Then the information value was divided into5part by natural break method, they are: light dangerous area (-9.79~-2.578)、less lightdangerous area(-2.578~-0.286)、moderate dangerous area(-0.286~3.037)、less highdangerous area(3.037~6.631) and high dangerous area(6.631~16.426).(4)The prediction model is established by considering the environmental factorsbase on the relationship between debris flow and rainfall and five sub-areas. The firststep is to determine the weight of influential factors by Grey Relational Analysis indifferent sub-areas. The second step is to build the prediction model. The predictionaccuracy was90.5%in moderate dangerous area,81.1%in less high dangerous area and89.6%in high dangerous area. By comparing the results came from those models whichonly considering precipitation, the models which were built by adding theenvironmental factors have higher accuracy, i.e.22%in moderate dangerous area,9.7%in less high dangerous area and14.3%in high dangerous area.