Landslides are a significant natural hazard within the geographical boundaries of Türkiye. Several contributing factors, such as channel incision, seismic activity, heavy precipitation, and anthropogenic influences, collectively underscore their importance as recurrent events. For the purpose of conducting AHP-based studies, as elaborated in previous sections, a careful selection of causal and triggering factors has been undertaken. This study has assessed various factors deemed important for initiating landslides, including lithology, land cover, internal relief, slope, aspect, classified landforms, classified curvature, and topographic wetness index (TWI). At the same time, rainfall and seismic activity were identified as critical triggering factors, providing the basis for the extensive analysis conducted herein.
The rationale for choosing a total of ten factors in the study was two-folds. First, these factors were chosen because of their widespread availability in the public domain and their established utility in landslide susceptibility research efforts. Second, to allow for a detailed and nuanced analysis, the selected factors were categorized into two distinct groups: one set of ten factors and another set of eight factors.
Due to their minimal impact on landslide development in areas of moderate susceptibility, aspect and rainfall were deliberately excluded from the ten-factor analysis. In particular, aspect was found to play a relatively minor role in influencing landslide susceptibility throughout the study basins. In addition, the precipitation factor, which denotes the annual mean total precipitation, represents the arithmetic mean of the annual recorded precipitation. While it provides insight into the annual precipitation received, it lacks the granularity necessary to measure rainfall intensity relative to the threshold for potential landslide occurrence. Extreme rainfall events with the potential to trigger landslides may not be accurately captured by the arithmetic averaging of annual rainfall data.
In this context, a careful landslide hazard assessment was carried out for each drainage basin using the AHP methodology. Two different approaches, namely the 8-factor and 10-factor models, were used to comprehensively investigate the susceptibility within each basin. Each basin was thoroughly examined and individual results are presented in Online Resource 3, as previously detailed by Okalp (2013) and Okalp and Akgün (2022).
Landslide inventory maps, crucial for validating the study's findings, were sourced from 1:500,000 scale maps published by the General Directorate of Mineral Research and Exploration of Türkiye (MTA) within the past decade. These authoritative maps were digitized to delineate historical landslide polygons. The polygons were then systematically overlain on the generated landslide susceptibility maps, which incorporated either eight or ten factors. Finally, for each susceptibility map, a comprehensive analysis of pixel counts, both within and outside the landslide areas, was performed to facilitate consistent evaluation.
Figure 3a illustrates the distribution of landslide susceptibility in the western Mediterranean basin under two scenarios: using eight-factor and ten-factor based maps. The figure shows pixel counts categorized as inside and outside historical landslide polygons. The vertical axes represent these pixel count distributions. A key challenge is the disparity in the range of pixel count values between the inner and outer landslide areas. This makes it difficult to visualize both distributions in a single figure. To address this issue, the peak locations for each scenario (eight and ten factors) were highlighted on a unified horizontal line labeled ''E-line'' in Fig. 3a, as suggested by Okalp (2013) and Okalp and Akgün (2022). While the peaks themselves may differ, this approach facilitated a comparative analysis of the overall distribution patterns.
After generating two landslide susceptibility maps - one based on eight factors and the other based on ten factors - a critical challenge emerged in the selection of an optimal map for the specific basin studied. Ideally, the most appropriate map would be reflected by the corresponding histogram curves. The ideal histogram for pixel values within historical landslide polygons would have a pronounced positive skew, with its peak (on the x-axis) approaching 1. Conversely, the optimal histogram for pixel values outside of the landslide polygons would exhibit a negative skew, with its peak value approaching 0. This established criterion, based on the distribution tails of the histograms, served as the primary basis for the selection of the eight-factor and ten-factor maps (Okalp, 2013).
Inspection of Fig. 3a reveals that the peak values of the histograms from both scenarios, derived from pixel counts within historical landslide polygons, were closely aligned. Moreover, the peak value of the eight-factor histogram for areas outside historical landslide polygons showed a slightly stronger negative skew, nearing a value of 0. However, these subtle differences in the histogram curves were deemed insufficient to decisively select the optimal map (eight-factor vs. ten-factor) for the basin under study. To resolve this ambiguity and facilitate data-driven selection, Receiver Operator Characteristic (ROC) curve analysis was incorporated into the evaluation process as a subsequent step (Aditian et al. 2018).
To objectively select the optimal landslide susceptibility map, a rigorous evaluation using ROC curves was performed. ROC curves provide a comprehensive assessment of model performance across all classification thresholds. The area under the ROC curve (AUC) is utilized as a quantitative indicator of the model's overall accuracy. As shown in Fig. 3b, a comparative analysis of the ROC curves revealed superior predictive power for the ten-factor model. This was evidenced by the larger AUC associated with the ten-factor ROC curve. In addition, the ten-factor ROC curve showed a trajectory closer to the upper left corner of the graph, indicating a stronger ability to discriminate between landslide and non-landslide pixels (Fawcett, 2006). Therefore, based on the robust performance metrics provided by the ROC analysis, the ten-factor model was selected for this particular basin.
Landslide susceptibility maps, along with landslide hazard maps, are often reclassified into a manageable number of classes (typically three to five) to facilitate interpretation. However, a significant obstacle for this process in this study was the substantial variability observed in the susceptibility maps produced for each individual basin. This variability precluded the application of generic classification thresholds across all maps. As a result, each map required independent determination of appropriate thresholds and classification schemes. Although several synthetic classification methods were explored, none provided entirely satisfactory results. In particular, the application of popular techniques such as Jenks' natural breaks, the quantile method, and the geometric interval method resulted in inconsistent thresholds, especially for the "very high" susceptibility subclass. Interestingly, the application of these methods identified the western Mediterranean basin as "very high landslide prone", a finding that contradicts the documented moderate landslide activity in the region. This discrepancy underscores the challenges in effectively using standard procedures to manage the synthetic classification of susceptibility maps produced in this research.
An innovative and highly subjective methodology was employed to synthetically classify the maps generated in this study (Okalp 2013, Okalp and Akgün 2022). Initially, the peaks from both the inner and outer landslide polygons were aligned on a singular axis (line E) in Fig. 3a, independent of their actual magnitudes. The peak value (point A) on the histogram curve, representing pixel counts from areas outside the landslide zones, was used as the initial threshold to delineate between the "no" and "low" susceptibility classes.
Subsequently, the intersection (point B) of the histogram curves for the outer and inner landslide polygons was established as the secondary threshold, differentiating the "Low" and "Moderate" susceptibility classes. Furthermore, the peak (point C) on the histogram curve for the inner landslide pixels was designated as the tertiary threshold, distinguishing between the "Moderate" and "High" classes. Lastly, the midpoint (point D) between point C and a fixed value of 1.0 was set as the quaternary threshold to separate the "High" and "Very High" susceptibility classes. This complex classification process was systematically applied to categorize the landslide susceptibility of the Western Mediterranean basin, as illustrated in Fig. 3c. Other basins were evaluated independently, with their corresponding results detailed in tables and graphs presented in the Supplementary Material Section as Online Resource 4.
Evaluation of the generated landslide susceptibility maps
The Analytic Hierarchy Process (AHP) enables the translation of qualitative concerns into quantifiable metrics, providing an invaluable tool for multi-criteria decision analysis problems. It skillfully transforms subjective judgments into objective data, thereby increasing the robustness of decision processes. In the context of landslide susceptibility analysis, the lack of established or fixed values for weights and ratings corresponding to different factors is further emphasized (Okalp 2013). This lack of standardization necessitates the use of objective analysis to determine these values, rather than relying on subjective expert opinion.
The Analytic Hierarchy Process (AHP) uses historical landslide footprints to determine rating values and factors influencing landslide occurrence. Overlaying these footprints with unnormalized values for each factor supports a detailed and unbiased assessment, thereby improving landslide hazard analysis.
The study revealed several key factors influencing landslide susceptibility across various basins. Slope exhibited a non-linear relationship, with the highest landslide frequencies concentrated between 5° and 15°, indicating a critical range for susceptibility assessments. Moderate internal relief (200–250 m/km²) emerged as a prominent factor associated with increased landslide activity.
Notably, historic landslides in the study areas clustered within a specific range of Topographic Wetness Index (TWI) values. A significant proportion occurred within a TWI layer where values of 12 and 13 were extracted from a DEM with a 90 m resolution. Because topography affects water movement in sloping terrain, the TWI effectively quantifies the effect of local topography on hydrologic processes. This provides insight into soil moisture distribution and surface saturation. Incorporation into the TOPMODEL which is a distributed hydrological model that aids in defining hydrological similarity, highlights the importance of TWI in modeling topography-driven processes at hillslope and basin scales.
Further analysis showed that about half of the basins possessed TWI values of 12–13, which fell at the midpoint of the TWI distribution. This suggested a potential threshold of TWI 12 for landslide initiation within the study basins at this resolution.
As expected, historic landslides occurred predominantly in areas with open and planar slope (S/S) curvature types. However, the lack of significant aggregation across precipitation and aspect distributions within the study regions necessitated the use of the Analytic Hierarchy Process (AHP). This involved a two-pronged analysis for each basin, employing both 8-factor and 10-factor layers.
Historical data analysis within the study regions showed that agricultural areas and forests were more susceptible to landslides, potentially exacerbated by land-use changes like deforestation. Additionally, a significant portion of historical landslides have occurred within Earthquake Zone 1, highlighting its role as a triggering factor. Regarding lithology, clastic and carbonate rock formations have been identified as being most prone to landslides due to their extensive history of such events and their inherent geological properties.
Rainfall intensity, particularly in specific regions, emerged as a primary trigger for landslides, underlining the crucial impact of precipitation on slope stability. Aspect had a minimal influence on susceptibility across the study basins. Specific landforms like U-shaped valleys and open slopes exhibited a clear association with increased landslide frequency.
After producing unclassified landslide susceptibility maps for the basins using both 8-factor and 10-factor approaches, a selection process was carried out as previously described. Comparative analyses showed that the 10-factor approach performed better in 9 out of 26 basins, namely Marmara, Buyuk Menderes, Western Mediterranean, Akarcay Endorheic, Western Black Sea, Yesilirmak, Kizilirmak, Upper Euphrates and Eastern Black Sea basins.
The decision between the 8-factor and 10-factor methods was made by analyzing pixel distributions within historical landslide zones (inner and outer) and examining Receiver Operating Characteristic (ROC) curves. The main criterion was to prioritize peaks of the outer landslide histogram closest to 0 and the inner landslide histogram closest to 1, using ROC curve values as a secondary criterion. When histogram data were inconclusive, the area under the ROC curve was reviewed.
The results revealed significant variations in ROC curves across the basins. Several basins achieved impressive AUC values, exceeding 0.7, indicating strong performance in predicting landslides. For example, the Lower Maritsa-Ergene Basin exhibited a remarkable AUC of 0.8709, highlighting its exceptional predictive capability. Similarly, the Akarcay Endorheic Basin reached a high AUC of 0.8421, showcasing its effectiveness in landslide susceptibility assessment. Conversely, some basins presented lower AUC values. The Upper Tigris Basin, for instance, yielded an AUC of 0.606, suggesting a need for further investigation or refinement for this basin. Similarly, the Sakarya Basin displayed a lower AUC of 0.5672, indicating that additional data or adjustments could enhance its predictive performance for this basin. The selected unclassified landslide susceptibility maps, developed using the AHP with either 8 or 10 factors, consistently classified the basins into five distinct groups, as outlined earlier. Table 5 summarizes these group distributions.
The Analytic Hierarchy Process (AHP) is a recognized technique for assessing landslide susceptibility and serves as a benchmark in current research, particularly when evaluating against machine learning algorithms. A study by Huang et al. (2020) in Shicheng County, China, assessed various models including heuristic AHP, statistical approaches, and machine learning techniques like Binary Logistic Regression, Multilayer Perceptron, Backpropagation Neural Network, Support Vector Machine, and C5.0 Decision Tree. The results demonstrated the effectiveness of all models, with the C5.0 Decision Tree achieving the highest accuracy with an AUC of 0.868, suggesting a potential for refining AHP-based results (AUC of 0.773) through integrating machine learning, neural networks, fuzzy logic, and other soft computing techniques.
Analysis of the "very high" landslide susceptibility zone, as shown in Fig. 4 and summarized in Table 5, revealed that in most basins the spatial extent of this zone exceeded that of the corresponding historical landslide footprint, highlighting the ability of the method to identify potentially landslide-prone areas beyond known locations.
The focus was on the combined area classified as "high" and "very high" landslide susceptibility within each basin (last columns of Table 5), as compared to the historical landslide distribution. Ideally, the "very high" susceptibility area should exceed the documented historical landslide footprint for the corresponding basin.
While most basins met this expectation, six basins, notably Akarcay Endorheic, Western Black Sea, and Upper Euphrates, showed a smaller "very high" susceptibility zone compared to the historical landslide area. This discrepancy is likely due to the threshold values used for classifying landslide susceptibility maps, which may have underestimated susceptibility in these basins.
Though a synthetic reclassification approach could address this underestimation, it falls outside the scope of this study as all basins were classified using the standardized procedure. Table 5 has been expanded to include an additional column displaying the combined area percentage of "high" and "very high" landslide susceptibility zones. Except for the Akarcay Endorheic Basin, the combined area of these susceptibility classes exceeded the historical landslide area for each basin. The minimal difference in the Akarcay Endorheic Basin may be considered negligible for landslide susceptibility assessment at a 1:500,000 scale.
Table 5
Synthetically classified landslide susceptibility zone distributions of basins and comparison of synthetically classified highly landslide susceptible zone areas with historical landslides
Basin | Selected factors | Pixel counts | Classified landslide susceptibility zones vs Historical LS |
No | Low | Moderate | High | Very high | High + Very high |
Lower Maritsa-Ergene | 8 | 1783136 | 64.35% | 23.22% | 10.26% | 2.10% | 0.07% > 0.04% | 2.17% > 0.04% |
Marmara | 10 | 2764674 | 31.42% | 18.53% | 41.49% | 6.38% | 2.18% > 1.89% | 8.56% > 1.89% |
Susurluk | 8 | 2938688 | 55.49% | 21.84% | 10.68% | 9.01% | 2.98% > 0.59% | 11.99% > 0.59% |
North Aegean | 8 | 1223660 | 45.55% | 23.62% | 9.75% | 16.98% | 4.10% > 0.54% | 21.08% > 0.54% |
Gediz | 8 | 2050888 | 51.17% | 26.51% | 16.51% | 4.52% | 1.28% > 0.50% | 5.80% > 0.50% |
Kucuk Menderes | 8 | 847415 | 57.04% | 31.62% | 10.48% | 0.73% | 0.12% > 0.08% | 0.85% > 0.08% |
Buyuk Menderes | 10 | 3155909 | 57.03% | 18.65% | 15.50% | 6.56% | 2.26% > 0.80% | 8.82% > 0.80% |
Western Mediterranean | 10 | 2552371 | 55.82% | 18.29% | 7.95% | 14.68% | 3.26% > 1.42% | 17.94% > 1.42% |
Antalya | 8 | 2492362 | 62.63% | 16.80% | 9.36% | 9.50% | 1.71% > 0.70% | 11.21% > 0.70% |
Burdur Endorheic | 8 | 724849 | 45.87% | 24.79% | 14.49% | 10.18% | 4.66% > 0.15% | 14.84% > 0.15% |
Akarcay Endorheic | 10 | 932025 | 60.37% | 28.87% | 8.78% | 1.76% | 0.21% < 2.81% | 1.97% < 2.81% |
Sakarya | 8 | 7412864 | 49.23% | 20.46% | 15.87% | 11.46% | 2.98% > 1.70% | 14.44% > 1.70% |
Western Black Sea | 10 | 3568596 | 46.39% | 22.41% | 12.24% | 14.37% | 4.59% < 10.56% | 18.96% > 10.56% |
Yesilirmak | 10 | 4863497 | 42.89% | 26.02% | 14.92% | 10.23% | 5.94% > 4.38% | 16.17% > 4.38% |
Kizilirmak | 10 | 9863497 | 45.84% | 21.97% | 11.46% | 16.62% | 4.11% > 2.01% | 20.73% > 2.01% |
Konya Endorheic | 8 | 5928017 | 64.77% | 20.05% | 9.97% | 4.61% | 0.59% > 0.05% | 5.21% > 0.05% |
Eastern Mediterranean | 8 | 2656410 | 52.36% | 22.79% | 10.87% | 12.15% | 1.82% < 2.24% | 13.98% > 2.24% |
Seyhan | 8 | 2655827 | 55.39% | 23.53% | 10.44% | 8.93% | 1.70% > 0.26% | 10.63% > 0.26% |
Lower Asi | 8 | 947267 | 71.06% | 5.98% | 8.25% | 10.95% | 3.76% > 0.35% | 14.71% > 0.35% |
Ceyhan | 8 | 2595952 | 73.98% | 8.68% | 9.29% | 6.21% | 1.84% > 0.77% | 8.05% > 0.77% |
Upper Euphrates | 10 | 14484190 | 47.05% | 20.87% | 16.12% | 13.13% | 2.84% < 4.98% | 15.97% > 4.98% |
Eastern Black Sea | 10 | 2813106 | 54.85% | 23.17% | 8.89% | 11.31% | 1.77% < 2.91% | 13.08% > 2.91% |
Upper Coruh | 8 | 2495302 | 51.02% | 25.96% | 6.46% | 13.34% | 3.22% < 5.01% | 16.56% > 5.01% |
Upper Aras | 8 | 3420905 | 53.81% | 12.91% | 12.74% | 15.24% | 5.30% > 4.54% | 20.54% > 4.54% |
Lake Van Endorheic | 8 | 2210789 | 41.25% | 30.64% | 14.17% | 9.81% | 4.13% > 2.08% | 13.94% > 2.08% |
Upper Tigris | 8 | 6426892 | 54.19% | 14.68% | 12.08% | 14.87% | 4.19% > 2.40% | 19.05% > 2.40% |
Conclusions and recommendations
This study has conducted a comprehensive analysis of landslide susceptibility mapping in mid-sized regions using publicly available datasets and a GIS-based semi-quantitative approach. The study systematically outlined an established framework for mapping landslide susceptibility. Using the Analytical Hierarchy Process (AHP), which is an accepted semi-quantitative method, the research applied this technique to the study basins by using both eight- and ten-variable models.
The study included rigorous validation and evaluation processes, including histogram and ROC curve analyses, resulting in factor-based maps for each basin. As a result, the research produced accurately constructed 1:500,000 scale landslide susceptibility maps for the designated regions. This work stands as a pioneering effort in systematically assessing landslide susceptibility across the major drainage basins of Türkiye. By encompassing all these basins, the study effectively provides the first comprehensive assessment of landslide susceptibility for the entire country. This basin-wide approach offers a more nuanced understanding of the factors influencing landslide occurrence compared to previous national-scale studies that might rely on less detailed data. The findings not only improved our knowledge of landslide susceptibility in Türkiye but also established an invaluable framework for future susceptibility assessments in other geologically diverse regions.
The methodology developed for semi-quantitative landslide susceptibility zoning provided important insights for decision-makers involved in regional-scale assessment of landslide susceptibility, hazards and risks, and is potentially applicable at both national and continental scales.
Scale and pixel size are recognized as critical elements in landslide susceptibility analysis. Their selection must prioritize practicality and ensure alignment with the capabilities of available hardware and software. This is particularly important when processing high-resolution datasets covering large areas. Given these constraints, the study identified a 90-meter pixel resolution and a 1:500,000 scale as optimal for investigating landslide susceptibility over large regions. Results indicated that the 8-factor approach produced superior results in 17 of the 26 basins analyzed.
A significant number of historical landslides were noted within the Topographic Wetness Index (TWI) layer, particularly at TWI values of 12 and 13, calculated from a 90-meter resolution digital elevation model (DEM). TWI is highly regarded as an important metric for modeling topographic influences at the hillslope or drainage basin scale. Detailed analysis of the distribution of TWI across the basins identified a value of 12 as the critical threshold for landslide initiation at the 90-meter pixel resolution used.
In the basin-specific assessments, curvature, landform, and seismic activity emerged as the primary controlling factors, collectively accounting for approximately 50% of the influence on landslide susceptibility (Table 6). The curvature and landform layers, which are terrain derivatives, were extracted directly from the DEM. These factors are significantly influenced by lithology, climatic conditions, and seismic activity, and serve as critical indicators of landslide potential. This demonstrates a strong alignment between digital modeling and field-based analytical methods for understanding and predicting landslide activity.
Table 6
Weights of factors obtained for basins (governing factors are highlighted in grey color)
Basin | Slope | Int. relief | Rainfall | Lithology | Land cov. | Earthq. | Aspect | TWI | Landform | Curvatur. |
Lo. Mar-Erg.8 | 12.24% | 9.24% | - | 9.97% | 11.11% | 9.31% | - | 11.11% | 18.47% | 18.55% |
Lo. Mar-Erg.10 | 8.09% | 8.09% | 14.85% | 8.09% | 8.09% | 8.09% | 4.43% | 8.82% | 15.72% | 15.72% |
Marmara8 | 13.17% | 7.35% | - | 8.51% | 13.17% | 16.91% | - | 9.29% | 15.35% | 16.24% |
Marmara10 | 10.67% | 6.87% | 7.66% | 7.66% | 12.92% | 15.04% | 2.83% | 8.31% | 14.02% | 14.02% |
Susurluk8 | 12.03% | 5.86% | - | 12.01% | 12.99% | 16.39% | - | 10.13% | 14.92% | 15.66% |
Susurluk10 | 10.92% | 4.91% | 8.23% | 9.50% | 11.99% | 14.54% | 2.90% | 8.84% | 13.62% | 14.54% |
Nort. Aegean8 | 9.04% | 7.25% | - | 11.67% | 11.67% | 18.08% | - | 10.08% | 15.40% | 16.83% |
Nort. Aegean10 | 7.63% | 6.30% | 8.81% | 9.47% | 9.47% | 15.81% | 5.64% | 8.18% | 13.44% | 15.25% |
Gediz8 | 10.45% | 7.74% | - | 11.42% | 12.02% | 18.27% | - | 9.13% | 15.48% | 15.48% |
Gediz10 | 8.38% | 6.53% | 10.32% | 10.32% | 10.32% | 16.44% | 3.17% | 8.38% | 13.07% | 13.07% |
K. Menderes8 | 12.17% | 8.54% | - | 9.28% | 11.13% | 18.56% | - | 10.17% | 14.45% | 15.70% |
K. Menderes10 | 9.72% | 8.09% | 9.72% | 8.09% | 9.72% | 16.52% | 3.48% | 8.09% | 13.28% | 13.28% |
B. Menderes8 | 9.10% | 8.20% | - | 14.81% | 9.10% | 17.30% | - | 9.99% | 15.98% | 15.52% |
B. Menderes10 | 7.44% | 7.44% | 6.75% | 12.02% | 8.66% | 16.67% | 4.06% | 8.01% | 14.47% | 14.47% |
West. Med.8 | 11.56% | 6.43% | - | 6.43% | 11.53% | 16.95% | - | 11.56% | 16.69% | 18.86% |
West. Med.10 | 10.88% | 5.76% | 6.48% | 5.76% | 9.77% | 13.44% | 4.50% | 10.45% | 15.89% | 17.06% |
Antalya8 | 12.01% | 6.16% | - | 12.01% | 12.01% | 12.92% | - | 12.92% | 14.75% | 17.22% |
Antalya10 | 9.37% | 4.93% | 14.52% | 9.37% | 9.37% | 10.82% | 4.74% | 10.07% | 11.29% | 15.52% |
Burdur End.8 | 10.00% | 5.00% | - | 6.01% | 10.00% | 22.21% | - | 10.00% | 17.52% | 19.27% |
Burdur End.10 | 9.07% | 4.62% | 9.39% | 4.62% | 8.70% | 17.78% | 3.82% | 9.96% | 15.57% | 16.46% |
Akarcay End 8 | 7.95% | 6.72% | - | 10.64% | 14.58% | 15.15% | - | 11.71% | 16.18% | 17.07% |
Akarcay End 10 | 7.02% | 6.18% | 11.97% | 8.23% | 12.74% | 12.74% | 3.94% | 9.50% | 13.64% | 14.04% |
Sakarya8 | 9.94% | 6.28% | - | 7.57% | 13.95% | 15.14% | - | 11.86% | 19.14% | 16.14% |
Sakarya10 | 8.94% | 5.59% | 7.42% | 6.88% | 12.54% | 13.39% | 3.60% | 11.02% | 15.31% | 15.31% |
West. Bl. Sea8 | 9.12% | 5.46% | - | 19.10% | 12.89% | 11.80% | - | 10.81% | 14.75% | 16.07% |
West. Bl. Sea10 | 8.41% | 4.89% | 5.94% | 14.61% | 12.88% | 11.57% | 3.97% | 9.35% | 13.77% | 14.61% |
Yesilirmak8 | 9.95% | 5.37% | - | 4.64% | 10.52% | 20.47% | - | 9.95% | 17.84% | 21.26% |
Yesilirmak10 | 8.93% | 4.57% | 8.47% | 4.05% | 9.33% | 17.91% | 3.62% | 9.75% | 16.68% | 16.68% |
Kizilirmak8 | 10.76% | 6.84% | - | 6.18% | 11.59% | 12.70% | - | 13.91% | 19.79% | 18.23% |
Kizilirmak10 | 8.71% | 5.79% | 12.68% | 5.18% | 9.87% | 10.65% | 3.20% | 11.03% | 17.15% | 15.75% |
Konya End.8 | 10.40% | 9.32% | - | 9.92% | 11.11% | 11.05% | - | 9.92% | 18.45% | 19.84% |
Konya End. | 8.77% | 8.77% | 8.77% | 8.77% | 8.77% | 9.56% | 4.66% | 8.77% | 16.11% | 17.06% |
East. Med.8 | 12.06% | 6.73% | - | 8.48% | 14.17% | 14.17% | - | 11.10% | 15.49% | 17.79% |
East. Med.10 | 10.40% | 5.48% | 11.85% | 7.09% | 11.09% | 11.85% | 3.89% | 9.76% | 12.77% | 15.80% |
Seyhan8 | 10.94% | 6.14% | - | 10.94% | 11.90% | 15.70% | - | 11.90% | 13.07% | 19.42% |
Seyhan10 | 9.28% | 5.12% | 8.15% | 9.28% | 11.17% | 13.81% | 4.38% | 9.97% | 11.99% | 16.84% |
Lower Asi8 | 7.77% | 7.22% | - | 9.06% | 15.13% | 19.98% | - | 8.17% | 16.34% | 16.34% |
Lower Asi10 | 6.95% | 6.56% | 7.56% | 8.20% | 13.61% | 17.74% | 3.22% | 7.23% | 14.47% | 14.47% |
Ceyhan8 | 8.76% | 3.93% | - | 12.13% | 17.52% | 14.81% | - | 10.18% | 14.93% | 17.74% |
Ceyhan10 | 8.02% | 3.68% | 4.25% | 11.33% | 16.04% | 14.07% | 3.68% | 9.05% | 14.93% | 14.93% |
Up. Euphrates8 | 9.69% | 6.18% | - | 6.78% | 10.52% | 16.42% | - | 11.56% | 19.42% | 19.42% |
Up Euphrates10 | 9.17% | 5.72% | 5.47% | 6.17% | 9.77% | 14.97% | 3.21% | 10.53% | 17.49% | 17.49% |
East. Bl. Sea8 | 9.93% | 6.10% | - | 11.81% | 19.08% | 9.93% | - | 11.81% | 14.32% | 17.03% |
East. Bl. Sea10 | 8.73% | 5.49% | 6.41% | 10.70% | 17.00% | 9.35% | 5.26% | 10.07% | 13.24% | 13.76% |
Upper Coruh8 | 11.84% | 6.12% | - | 9.23% | 11.97% | 15.49% | - | 10.88% | 15.42% | 19.06% |
Upper Coruh10 | 10.07% | 5.16% | 10.82% | 8.31% | 10.84% | 12.07% | 4.75% | 9.44% | 12.98% | 15.56% |
Upper Aras8 | 10.19% | 6.10% | - | 5.29% | 10.19% | 20.73% | - | 9.76% | 19.59% | 18.15% |
Upper Aras10 | 9.06% | 5.40% | 8.60% | 4.50% | 8.72% | 17.24% | 3.48% | 9.87% | 17.24% | 15.90% |
Lk. Van End.8 | 10.81% | 6.09% | - | 5.08% | 10.10% | 19.48% | - | 10.10% | 20.85% | 17.48% |
Lk.Van End. 10 | 8.88% | 4.84% | 10.89% | 4.53% | 8.88% | 16.78% | 4.67% | 9.86% | 15.34% | 15.34% |
Upper Tigris8 | 9.10% | 4.99% | - | 11.72% | 13.43% | 19.79% | - | 11.72% | 14.63% | 14.63% |
Upper Tigris10 | 8.23% | 4.26% | 8.78% | 10.10% | 11.31% | 17.22% | 3.35% | 10.10% | 13.33% | 13.33% |
8 Basin results for 8-factor based analysis |
10 Basin results for 10-factor based analysis |
The methodology used in this study to evaluate mid-scale landslide susceptibility mapping in mid-sized regions provides a valuable framework for future efforts focused on the development of comprehensive nationwide landslide susceptibility maps. However, direct implementation of sophisticated modeling techniques for national or continental-scale landslide susceptibility assessment, which require larger data sets, finer pixel resolution, and greater computational power, remains impractical due to current hardware and software limitations.
These factors were incorporated into a comprehensive analysis that resulted in the development of landslide susceptibility maps for the study regions. The semi-quantitative model employed facilitated the generation of histogram and Receiver Operating Characteristic (ROC) curves. These curves were then subjected to rigorous evaluation and reclassification using a novel methodological approach. This process culminated in the careful construction and robust validation of several landslide susceptibility maps within the study areas. Guided by our findings, we propose the following recommendations for future landslide susceptibility research:
-
Prioritize open data accessibility: This research utilized open-source or publicly accessible datasets. Hard copies of landslide inventory maps, geological maps, and rainfall data were obtained from relevant government agencies and subsequently digitized into vector formats. Additional datasets were sourced from CGIAR-CSI and the European Environment Agency.
-
Utilization of causal factors from DEM: The study utilized the Digital Elevation Model (DEM), from which most of the causal factors, including internal relief, slope, aspect, classified landforms, classified curvature, and topographic wetness index (TWI), were derived. This highlights the critical role of the DEM in landslide susceptibility assessment.
-
Application of Multi-Factor Approaches: The research involved the application of both eight-factor and ten-factor based approaches within a semi-quantitative framework.
-
Innovation in Landslide Susceptibility Maps: A groundbreaking technique for synthetic classification of the generated landslide susceptibility maps within the semi-quantitative model was implemented.
-
Future evaluation of the TWI variable: Future studies are intended to comprehensively evaluate the Topographic Wetness Index (TWI) at various pixel resolutions and scales to determine whether a TWI value of 12 is a viable threshold for assessing landslide susceptibility.
-
Applicability of Study Methodology: The study's comprehensive methodology demonstrates the potential for transferability to landslide susceptibility mapping over large geographic regions, including countries or continents.
-
Impact of DEM resolution: The increase in DEM resolution plays a critical role, potentially revealing hidden relationships or rules within coarse-resolution DEMs.
-
Strategic partitioning for area analysis: For extensive landslide susceptibility analyses, strategic subdivision of the study area into smaller sub-basin boundaries is recommended. This approach allows for a discrete and more accurate analysis of each basin and promotes a deeper understanding of the unique basin characteristics that influence landslides.
-
Expanded Subdivision for Advanced Study: To facilitate more detailed analysis in the future, it is suggested that the study basins be subdivided into sub-basins. This, combined with finer pixel resolution and larger scales, will result in a more accurate and comprehensive assessment of landslide susceptibility.
This study presents the first comprehensive assessment of landslide susceptibility for Türkiye, analyzing various factors such as slope, topography, and land cover across all major drainage basins. The research provided valuable insights for future susceptibility assessments. Successful implementation of such detailed studies requires robust interagency coordination. This coordination facilitates the strategic allocation of qualified personnel, time, financial resources, and reliable data sets, thereby fostering significant progress in the field of landslide hazard assessment.