Akademik Veri Yönetim Sistemi
3rd International conference on Mediterranean Geosciences Union, MedGU 2023, İstanbul, Türkiye, 26 - 30 Kasım 2023, cilt.Part and, ss.125-130, (Tam Metin Bildiri)
One of the primary Göksu River sub-watersheds, the Ermenek River watershed, has a surface area of 4020 km2. In Miocene clastic and carbonate units that crop out in the deeply incised valleys, landslides are frequently seen. In the watershed, 354 deep-seated slide-type landslides totaling 184 km2 in size were found. It was decided to take into account seventeen landslide conditioning parameters. The landslides were randomly split into 80% and 20%, respectively. The one-dimensional convolutional neural network method was used to assess the landslide susceptibility assessments. The convolutional layers with two hidden layers, 50 and 25, were activated using the rectified linear activation function. The landslide susceptibility output was calculated using the sigmoid algorithm. Using natural breaks optimization, the landslide susceptibility map was categorized into five classes. The prediction-success rates and the area under the receiver operating characteristic curve were used to evaluate its performance. The receiver operating characteristic curve’s area under it was 0.82, which indicates outstanding discrimination. The resulting susceptibility map, whose high and very high sensitive zones equate to 10.14% of the watershed and include 88.66% of the reported landslides, provided a high degree of forecast accuracy.