Paper No. 208-20
Presentation Time: 9:00 AM-6:30 PM
LOGISTIC REGRESSION MODELING OF LANDSLIDE SUSCEPTIBILITY IN A REGION OF MUSKINGUM COUNTY, OHIO, USING GIS AND SPSS
Mapping landslide susceptibility is very important for civil engineering projects. Knowing the locations of current or potential landslides can be very beneficial for planning future projects in an area. Muskingum County is located in the southeastern part of Ohio. The local geology consists of primarily Pennsylvanian-aged rocks from the Conemaugh, Monongahela, and Allegheny groups. The area is part of the unglaciated Allegheny Plateau with many hills and valleys. A landslide susceptibility map was developed using logistic regression modeling of known landslide locations. Known locations were obtained from ODOT, personal observation, and aerial photographs. Overall, we mapped 186 known locations and 93 of these locations were chosen randomly for use in creating the landslide susceptibility map. The remaining 93 known locations were used to test the accuracy of the model. In addition to the known landslides, the model incorporated 186 random locations which were generated in GIS for use in the model. Factors included in the modeling were: slope, aspect, plan curvature, altitude, stream power index, topographic wetness index, land use, and lithology.
Maps were then created in GIS to display each factor individually for the study area. All maps were based on a 2.5 foot-grid DEM of the area except for land use and geology. The logistic regression modeling was done using SPSS. The maps and the logistic regression factors were combined using the raster calculator to produce the Landslide Susceptibility map. Based on the beta values from the logistic regression model, it appears that geology plays the biggest role in landslide development in Muskingum County. The Monongahela Group had the highest value at 3.548. Other factors which had higher positive values that affect landslide susceptibility are land use, aspect, slope, and SPI. The accuracy for the logistic regression model was around 86%. Of the 93 events that were selected for this study, 93.55% (87) occurred in either High or Very High areas of susceptibility. The unused known events had 89.25% (83) occurrence in High or Very High areas. This model appears to have very good accuracy for predicting landslide susceptibility in Muskingum County.