Landslide Spatial Prediction Based on SlopeUnits and Support Vector Machines

Landslides are major natural geological disasters in China,and large-scale engi-neering activities induce and aggravate the occurrence of catastrophic landslides.Traditionalspatial analytical techniques cannot easily discover patterns,trends,and relationships thatcan be hidden deep within complicated landslide hazard systems due to limited data sourceand long update cycle.Focusing on the Three Gorges,a variety of environment and trigge-ring factors for landslide occurrence were calculated or extracted from the multi-source spa-tial data.Secondly,the study area was partitioned into slope units derived semi-automatical-ly from a digital elevation model to resample the conditioning factors.Finally,a two-classSVM was trained and then used to map landslide susceptibility with the best accuracy of 98.21%.To evaluate the models,the susceptibility maps were validated by comparing themwith the existing landslide locations according to success rate curve and error rates.