The dynamic internal structure of the Earth has been through a continuous change, which is influenced by both natural and anthropogenic factors, since its creation. These natural factors commonly cause a variety of natural phenomena such as earthquakes, volcanic eruptions, floods, landslides, extreme weather events etc., while anthropogenic activities usually disrupt natural balances and turn natural events into disasters causing deaths and destructions (Alcantara-Ayala 2002). From this point, comprehensive disaster analyses and determination of natural hazards and risks are of critical significance to mitigate the potential losses especially in areas intensely inhabitated.
Landslides and rockfalls are amongst the main natural events that may pose high risks to the humans and settelements. A diversity of factors such as geological, geomorphological, climatic and meteorological influences, as well as the human activities initiate landslides. There are also triggering factors which cause the gravity-driven downslope movement of the large masses (soil, rock, debris etc.) (Ercanoğlu et al. 2008). Landslides occur in a wide range of different geographies in the world and significantly affect the landscapes (Gariano and Guzetti 2016). Yet, while changing the physical structure of the environment, landslides can also end up with severe economic losses, infrastructure damages, injuries and fatalities (Prakash et al. 2020). According to UNISDR, 4.8 million people were affected from landslides worldwide and 18414 fatalities were recorded between 1998-2017 (United Nations Office for Disaster Risk Recudtion [UNISDR] 2017).
A number of studies about natural disasters covering different research periods in Turkey also point out that landslides take the first place in terms of event number in the country. According to the evaluation based on the number of the influenced housing, landslides rank second after earthquakes in terms of the losses they caused (Çan et al. 2013). Landslides are very commonly observed in the Black Sea, Central and Eastern Anatolia regions in the country and result in severe physical and economic impacts as well as deaths and injuries. According to the Disaster and Emergency Management Presidency (AFAD), 13494 lanslides and 2596 rock fall events were reported in the country between 1965 and 2015, while 151 events happened in 2018 (AFAD 2015). The 2019 disaster statistics, on the other hand revelaed that 245 landslide/rock fall events took place in Turkey (AFAD 2020).
Within this context, development of landslide susceptibility maps, which deals with the spatial likelihood of the mass movements associated with their occurrence in a particular area (Nsengiyumva and Valentino, 2020), is vital to make efficient physical planning, manage potential risks for the existing settlements and structures, construction works (dams, roads, etc.), and delicate landscapes. Thus, loss and damage risks are properly and timely mitigated (Highland and Bobrowsky 2008). For this reason, researches and studies focusing on landslides hazard and risk modelling via different approaches and techniques have been an important research area. Statistical methods (Mersha and Meten 2020; Pasang and Kubíček 2020; Thanh et al. 2020; Zhang et al. 2020), machine learning algorithms (Bui et al. 2020; Fang et al. 2021; Merghadi et al. 2020; Sahin 2020; Wang et al. 2020) and hybrid models (Chen and Chen 2021; Chen and Li 2020) are commonly utilized by a good number of researchers to develop susceptibility maps, make predictions for potential flows and compare the efficiency and accuracy of different techniques.
For example, Kirshbaum et al. (2020) examined the landslides in the High Mountain Asia region mostly initiated by extreme precipitation, and used satellite and Global Climate Model data and applied Landslide Hazard Assessment for Situational Awareness (LHASA) model to determine the potential landslide hazard in the future within the study area. Slope, lithology, land cover change, distance to road networks, and distance to fault zones data were utilized for the development of the landslide susceptibility map. Lui et al. (2021) utilized three machine learning techniques to model the landslide susceptibility triggered by rainfall in Veikledalen Valley, Norway. The authors used slope angle, aspect, plan curvature, profile curvature, flow accumulation, flow direction, distance to rivers, total water content, saturation, rainfall and distance to roads data as the triggering factors. Lacroix et al. (2020) aimed to determine the relation between the irrigation and the landslide activities on the southwestern coats of Peru. The authors used Hexagon spy satellite and SPOT 6/7 images to detect the land use and morphological changes (elevation change patterns) between 1978 and 2016, and benefited from KH9, Landsat 5 and Landsat 8 imagery to determine the horizontal displacements in the study area. The results showed that large slow-moving landslides occurred within the irrigated areas. Lee and Pradhan (2007) used parameters such as slope, aspect, curvature, precipitation distribution, lithology, vegetation index, land cover, distance from drainage, etc. and applied both frequency ratio (FR) and logistic regression (LR) methods to develop landslide hazard map in Selangor, Malaysia, which is prone to severe landslide activities especially triggered by intense rainfall. In their study, the authors both used satellite images and conducted field surveys to detect the landslide locations, and after the analyses concluded that FR method provided more accurate results compared to LG in their study area.
Besides the techniques adopted, the parameters used for assessing the landslide hazards are of vital importance and may vary according to the aim of the study and the characteristics of the geographic context. Gökçeoğlu and Ercanoğlu (2001) focused on the commonly used geological, geomorpholological, hydrological, and antropogenic data sets and evaluated 21 studies conducted for the development of landslide susceptibility maps. The authors found that slope data was used in all 21 studies, lithology in 20, distance to main faults in 11, curvature and elevation in 10, and drainage network, vegetation and land use potential datasets in 8 studies. In a nother similar study, 117 studies in the literature were examined to determine the parameters used to develop landslide susceptibility maps. According to the results, it was detected that 94.02% of the studies have used curvature, 67.52% lithology, 63.25% aspect, 51.28% drainage characteristics, and 50.43% elevation (AFAD 2015).
Within this context, the aim of this study is to develop a GIS-supported landslide susceptibility maps in Karaburun Peninsula, İzmir, using FR method. As Thanh et al. (2020) also underlined, FR is a very popular bivariate statistical method for the assessment of the landslide susceptibility, since it is easy to use and provide good results. Reciever Operating Characteristic (ROC) analysis was performed for the determination of accuracy of the landslide susceptibility map.