5.1. Causative factors analysis
During recent years, the application of statistical bivariate models (i.e. FR, SE, WofE, EBF, LR, IV…etc.) to create landslides susceptibility maps (LSM) have become one of the compelling methods and are thus, widely used for landslides hazard assessment, mitigation risks and good management of territories (Guzzetti 2006; Pham et al.2018). Likely, in the current study area and its surroundings, several models have been tested effectively in numerous studies. The previous studies so far not suggested in favor of any one model better than another (Guzzetti et al. 1999; Van Westen et al.2006). The variable and better results are always attributable to the structures of the models used and the quality of the input data (Pham et al. 2018).
The process and the spatial distribution of landslides, generally cannot be understood and explained by a single factor, but is attributed to a combination of more than one factor (Varnes1978; Crozier 1984). Therefore to have better understanding and quantifying the relationship between the landslides occurrence and the eleven causative factors (i.e. altitude, slope, aspect, curvature, shaded relief, proximity to streams, proximity to lineaments, proximity to road network, land use/cover, lithology and rainfall), and to determine the important causative factors, the current study has been carried out... This study and the steps involved are very crucial for better guiding, and promoting the prevention plans against landslides incidences by targeting the predominant factors. Thus, in this study based on Shannon entropy modeling in the region of Tetouan-Bou Ahmed and its hinterland, northernmost of Morocco, it was found that rainfall followed by elevation, slope and lithology are the most effective factors in landslides trigging. This finding is consistent with our field observations and others studies carried out in the Rif area (Millies-Lacroix1965; Fares 1994 ; Al Gharbaoui1980; Chaouni1999 ;El Kharim2002 ; Moutchou2014; Brahim et al.2018; El moulat and Barhim2018). The other factors such as proximity to stream, landuse, proximity to faults, shaded relief, aspect, proximity to roads and curvature are respectively less involved in landslides occurrence in the area. A quantitative comparison of the results between models, it is found suitable to use this dataset in building these models (Iqbal 2020). Es-smairi et al. (2021), using AHP and WofEmethods, found approximately similar result; in thisregion for building models with a main difference of priority rankings of major and secondary causative factors in landslides occurrence. As mentioned before, having an idea of the weight of the impact of the factor in the triggering of mass movements represents a very important step in the studies of prevention and mitigation of risks. This could still perform by giving more details on the vulnerable areas exposed to the action of a given factor. For this purpose, we were used FR model for highlighting the critical areas to landsliding. The analyses of FR results showed that, as expected, a linear correlation between elevation and landslidingis found. This correlation further showed that the more we advance towards the high countries ( JbelKelti,JbelEfrane,HafatJeltane, JbelTazout,Jbeldarsa), the frequency of terrain instabilities increases. However, a decrease in frequency is observed in the high ridges of the limestone and dolomite due to its resistance against water erosion and weathering processes. In addition, the area has abundant cover of vegetation and forest, which helps in stabilization of the hill slopes (Es-smairi al al. 2021).
The moderate elevations (100-600m) is the most vulnerable to slope failure under study area. It is because; these elevations overlap with Ghomarides nappes, the Sebtides, flysch nappes and the tangier units, which are very predisposed to landsliding. It is due to occurrence of predominantly of flyschs deposits that are very deformed, altered and poorly resistant. The very low and low elevation area, which superpose with flat areas corresponding to the furrows of the main rivers, the alluvio-marine and plains, and the sandy coasts are the most stable areas to landslide occurrences. The same is observed for slope factors that is the susceptibility to landslide risk increases as the slope angle is increasing., We note that the linear correlation of slope and landslide occurrences in the area is interrupted for two last classes which can be however, explained by the superposition of these slopes at the crests and cliff of Calcareous Dorsaland sandstone bars that are rheologically competent and resistant. Similarly, the north direction aspect followed by West and East aspect had the most impact on landsliding process respectively. These directions are less exposed to solar radiation and more wetness as observed is due to the wet currents of the Mediterranean Sea and the Atlantic Ocean, whereas south-facing hill slopes more exposed to solar radiation show a weak impacton slope movements. Among the curvature classes, the concave hillslope class has the greatest probability for slope failure. It is probably due to the capacity of these topographic forms to collect rainwater that after percolation leads to leaching and alterations of the rocky material and consequently the reduction of the mechanical resistance of these slopes (Es-smairiet al.2021). Further, the results of proximity to streams showed that it closely exerted an important impact to undermining of the base of hillslopes, especially after intense rainfall. The debris flow and sliding are common under study area. The disconnection in soil and rocks related to the faults and the major contact between geological formations constituting an important conduit for water circulation, which are the main causes responsible for landslide trigging. Under study area, many water sources flow along its geological contacts constituting sources of drinking water supply for several villages (i.e. source of Bni Salah, source of BeniHarchan, source of Bouaanan, source of Ezzarka etc.), In this study, the results show that the distance between 90 and 500m from fault constitute the areas more prone to landsliding. Our field observation show that, in addition to the impact of the proximity to the sea (undermining of the continental shelf by dynamic and erosion of water sea) the installation of the coast national route 16 (N16) that connects Tangier on the northwest coast of Morocco to Saïdia on the northeast coast passing through study area, was the major cause of the local activation and reactivation of some hot spot landslide occurrences, .However, the results show a weak relation between landslides occurrence and proximity to roads this probably is due to the moderate number of landslides mapping along roadside and a large part of roads network pass under flat zones.
In term of landuse and as expected, the barren lands deprived of cover vegetal are more exposed to water erosion and sliding. Lithology is one of the most determining factors in the occurrence of landsliding in Rif domain; since this factor has undergone a very intense tectonic stresses (e.g., shear zones, foliation, fractures, faults, folds). The presence of a very contrasted climate with very intense rainy periods, these processes accentuate the alteration process of lithology (dissolution, karstification, disintegration, soil formation etc) that lead to mass wasting and weakens their cohesion and resistance to landsliding and erosion .Among the lithological units that show a very strong susceptibility to mass movements are sandstones-marlstones and marlstones-limestones respectively., These later overall are calcareous clays moderately plastic and with a weak mechanical strength. Besides this, the annual rainfall factor is relevant in landslides trigging especially after heavy rainy days. The water saturation decreases the cohesion forces and the shear strength which gives rise to triggering instabilities (Song et al. 2012; Hamed et al. 2014). The results show a linear relationship between rainfalls and spatial distribution of landslides except for the last three classes that show a less influence in landsliding process. This disagreement is probably due to the fact that these classes superpose with high elevation that are constituted by limestone and dolomite of calcareous dorsal that are less sensitive to landslides and water erosion. These results of FR model are in agreement with results of WofE model using by Es-smairi et al (2021).
Overall, in light of these results obtained from FR and SE model generally, we can note that on the one hand, the models function almost similarly; on the other hand, the most important factors of instability of the hill slope and consequently on landslides occurrences are elevation, rainfall, slope and lithology. The analyses of the maps as shown in (Fig. 6a,b) and the results of (Fig. 7a,b), it is found to be highly in agreement with the terrain condition reality and the spatial distribution of landslides in the area. Furthermore, approximately 14% of the study area does not constitute a real risk by landslides occurrences as it corresponds to flat zones (alluvium plains and sandy coasts). The area that is classified as moderately susceptible to slopes failure is represented by 24% of the study area. It coincides with high altitudes of the calcareous dorsal resists to erosion and weathering. But, this area, still and always represents a tendency towards collapse (rocks falls, debris slide) following the intense fracturing especially as a result of anthropogenic intervention (e.g., quarrying activity). Besides this, large area (60%) is classified as most dangerous zone that is highly exposed to landsliding. This area extends over the outcrops of the Ghomarides nappes, the Sebtides, the Flyschs nappes and the tangier unit, crossed by several river notches. These areas have undergone a very intense and complex kinematic course along its formation, recognised by significant metamorphism (e.g., folds, foliations schistosity, fracture, Faults etc). It is accompanied with a strong supergene weathering area that makes them very fragile to instability. Just a tiny unearthing of the foot of a hillslopes, is enough to initiate the triggering process of landsliding. In summary, these results with quantitative methods, allowed to highlight the important and the close relationship between geological, geomorphological, hydrogeological, climatic and anthropogenic activities in the context of landslide occurrence in the coastlines between Tetouan- Bou Ahmed and its hinterlands.
5.2. Validation and comparison
Among the most crucial steps in any risk prediction modelling is the validation of the results obtained. The reliability of the results is also closely dependent on the quality of the data and the model used (Phamet al. 2018). In this study, verification and validation were done on the one hand, by a direct confrontation with the reality on the terrain (the inventoried instabilities), and on the other hand, by using the AUC curves of the ROC method (Yilmaz 2010). The ROC curve plots in the Y-axis shows the rank of the landslide susceptibility index in descending order whereas; the X-axis shows the cumulative percent of landslide occurrence corresponding to success and prediction rate, respectively (Trigila et al. 2015). A curve of a model with a largest AUC, which varies from 50–100%, can be considered the best model (Hong et al2018). The Analysis of the AUC results of the current study, shows that for the FR Model, the success rate using training data was 79.023% and for the prediction rate using validation data was 78.974%, (Fig. 8a,b).Whereas, for SE model, the success rate, and prediction rate were 76.689% and 76.016% respectively (Fig. 8a,b). It appears that FR model is very interesting for exploring areas prone to slopes failure compared to the SE model. In another study, carried by Es-smairi et al. (2021) using WofE and AHP, it was found that the prediction accuracy for both models is 74,653% and 71,394%, respectively. We can depict from this comparison that the prediction precision given by both models (FR and SE), is reasonable and consistent in term of production of landslide susceptibility maps under our study area, with clear advantage of FR model. This finding of accuracy and precision of FR modelling landslides susceptibility mapping is similar and appreciated with many studies in several regions of different context suffered by landslides occurrence, e.g., India (Kannan et al. 2013), Himalaya (Regmi etl.2014), Malaysia (Lee and Pradhan 2007), Indonesia( Rasyid et al.2016), Nepal (Thapa and Bhandari2019), Iran (Nohani et al. 2019), Turkey (Yalmiz2009), Algeria (Karim et al. 2019)and other context.