Managing disasters is challenging due to various kinds of hazards (e.g., natural and anthropogenic hazards) (De Silva and Kawasaki, 2018). Natural processes will become natural hazards, when they intersect and have a negative effect on people life (Gill and Malamud, 2017; Ward et al., 2020). Subsequently, natural hazards are the physical events that pose devastating impacts on communities, damaging critical infrastructure, affecting economic viability, and claiming lives (IPCC, 2007). Worldwide, an exponential increase in the human population leads to lateral expansion in development (urbanization, agricultural activities, infrastructures, and life lines), subsequently increasing the exposure and interaction with different types of natural hazards (Weinkle et al., 2018). Globally, several areas are vulnerable to events involving several types of hazards that are related to the same area and at the same time (Leonard et al., 2014). Natural hazards have impacted people and natural environments in many undeveloped countries more severely than developed countries, causing substantial loss of lives and enormous economic plunge (Iglesias et al., 2021). In a simple definition, a hazard refers to the probability of potential damage in certain areas (location or where?) and a specific period (time or when?) due to an event with specific magnitude (how large?) (Shi, 2019).
Arid and semi-arid areas around the world are impacted by different natural hazards events, ranging from insignificant influence events to extremely events (IPCC, 2012; Tabari 2020; Rutgersson et al., 2021). These natural hazards frequently occur, threatening everything (modern cities, ruler areas, infrastructures, and life lines) (De Silva and Kawasaki, 2018). These disasters can cause great economic damage, disruption of transportation systems, injury, and claiming lives. Mountainous provinces are highly likely among the most disaster-susceptible areas due to their various characteristics such as lithology, tectonics, climate, and hydrology (Baig et al., 2021). Risk related to natural hazards is high in middle East countries due to lack of disaster preparedness and public awareness, and inadequate funding support (AlQahtanya and Abubakarb, 2020). Saudi Arabia is one of these countries that experienced many natural hazards, yearly (Alyami et al., 2021). These events cost lives and bring the economy to a static state. Moreover, not much has been done to map different prone areas related to these natural hazards.
Mountainous regions that are inhabited by people and crossed by infrastructures are prone to not only one type of hazard, but they are frequently the scene of multiple disasters that interact together, such as earthquakes, avalanches, landslides, floods, mudslides, ground subsidence, soil erosion, and wildfires (Gill and Malamud, 2014; Shah et al., 2018). Duncan et al. (2016) defined multi-hazard as all potential and cascading hazards, in any particular area at a certain period. Most of the time, these hazards can cause severe damage to the various human activities located and intercept with the hazard prone areas by causing fatalities and injuries, damaging their urban and industrial areas, destroying infrastructures (roads, tunnels, and railways), and disrupt lifelines (power lines, water systems, and gas mains) (Bell and Glade, 2004).
Due to the complexity of natural hazards, many studies deal with individual hazard types as an independent approach (Wastl et al., 2011). Also, problems related to these problems (e.g., floods, landslides, and gully erosions) have been evacuated individually by applying various machine learning algorithms-MLAs (Sarkar and Mishra, 2018; Ghorbanzadeh et al., 2019; Hosseiny et al., 2020; Zhou et al., 2021). However, Earth system science approach indicates that a significant interaction between different types of hazard components due to the interaction between the component systems (e.g., the lithosphere, atmosphere, hydrosphere, and biosphere). Accordingly, a holistic approach to understand all different hazards in a certain area is a must and can prevent cascading hazards that can be formed by the interaction of different hazard types. A multi-hazard risk evaluation could be significant to control hazard interactions (Komendantova et al., 2014). To prevent these natural hazards and protect people and their properties and the country economy from long-term plunging, effective predictive models must develop (Bathrellos et al., 2017). These models should base on profound understanding of the most influential and triggering factors that have a significant contribution on these hazards (van Westen and Greiving, 2017).
Multi-hazard modeling becomes an essential tool in land use development at both regional and national scales (Saunders and Kilvington, 2016). These multi-hazard modeling approaches gain good attention recently worldwide based on their ability to consider different types of hazards that could impact the area (Schmidt et al., 2011; Skilodimou et al., 2019; Lombardo et al., 2020).
Different approaches were used to conduct multi-hazards modeling e.g., by using two decision-making tools including, sequential Monte Carlo method and decision-making tool (Komendantova et al., 2014); using a deterministic equation based (theoretical) and empirical understanding (Bout et al., 2018); and by using multi-criteria analysis and GIS (Skilodimou et al., 2019). Recently, MLAs have been widely used to predict various hazards based on random forest (RF), and support vector machine-SVM (Nachappa et al., 2020), boosted regression tree-BRT-, and generalized additive model-GAM (Ye et al., 2020).
In this study, no multi-hazard assessment has been done in Saudi Arabia before and the application of machine learning techniques in multi-hazard assessment will be a novel study in the area. In this study, developing a multi-hazard risk map using machine learning techniques for the most impacted natural hazards (landslides, floods, and gully erosion) in this area are crucial for effective landuse management. The current study evaluates machine learning models (MDA, GLM, FDA, BRT, and RF) as effective and accurate models to produce multi-hazard risk map for Hasher-Fayfa Basin that might be used by authorities, developers, and decision-makers.