KURAL TABANLI ALGORİTMA İLE HEYELAN DUYARLILIK HARİTALAMASI (BENİ AHMED BÖLGESİ, RİF DAĞLARI, FAS )
Özet
Landslides are still one of the most important natural hazards considering the economic losses and loss of life which they cause in all over the world. Even though numerous studies have been conducted over the last few decades, the researches related with reducing landslide hazard still keep up-to-date. The main reason for this issue the raise in complexity of the landslides observed in almost every geographical area of the world. Landslide susceptibility analysis is one of the first and the most important stage for the evaluation of landslide spatial probabilities; in other words, it is essential for landslide assessments performed in medium to regional scales to reduce landslide hazard. The purpose of this study is to produce a landslide susceptibility model by using a cascaded Mamdani fuzzy algorithm, a rule-based fuzzy algorithm based on expert opinion, in the Beni Ahmed Region in Rif Mountains in Morocco. The investigations were carried out in four stages. These are (i) producing of landslide inventory map for the study area, (ii) evaluation of the conditioning factors in the region, (iii) construction of the cascaded Mamdani fuzzy algorithms and rule-based structures of the expert models, and (iv) evaluation of the cascaded fuzzy inference systems, mapping of the landslide susceptibilities, and performance assessments. As a result, total 323 landslides were mapped in accordance with the field studies performed in the region, interpretations of satellite images which were acquired from the Google Earth, and recent papers which were already published in current literature. The landslide inventory produced in this study was used for verification of the expert models. The Receiver Operating Characteristics (ROC) curves were implemented in order to assess spatial prediction performances of the cascaded fuzzy inference systems. For this purpose, the Area Under the ROC Curve (AUC) statistics were evaluated. According to the AUC values, the best performance was calculated to be 0.67 for the model in which narrow membership functions and Defuzzification-Free Hierarchical Fuzzy System (DF-HFS) were implemented. As a consequence, in case necessary database is provided, expert-based landslide susceptibility maps can be produced for all Rif Mountains by using the expert models constructed in this study.