Introduction
Material and methods
Study area and geological structure
Pixel value | Symbol | Age / Formation name | The content of the formation |
---|---|---|---|
1 | Qal | Quaternary / - | Alluvium |
2 | Qym | Quaternary / - | Slope debris |
3 | Tekçd | Middle-Upper Eocene / - | Diorite |
4 | Tekça | Middle-Upper Eocene / - | Dacite, rhyodacite |
5 | Tet | Middle Eocene / Taşpınar formation | Andesitic and dacitic volcanics, volcano-clastic turbiditic rocks |
6 | Tek | Middle Eocene / Kabaköy formation | Andesite, basaltic lava and pyroclastics, sandy limestone, sandstone, marl, tuff |
7 | Tee | Middle Eocene / Erenler formation | Mudstone, claystone, sandstone alternation |
8 | Tpeb | Paleocene-Lower Eocene / Bakırköy formation | Siltstone, claystone, sandstone, clayey limestone, marl |
9 | Kk1 | Upper Cretaceous-Paleocene / Kaçkar granitoid-I | Granite, granodiorite, quartz diorite, adamellite, gabbro, diabase |
10 | Kk1kd | Upper Cretaceous-Paleocene / - | Quartz diorite, diorite |
11 | KTc | Maastrichtian-Danian / Cankurtaran formation | Sandy limestone, mictiric limestone, tuff, marl, volcanics sandstone, agglomerate |
12 | KTct3 | Maastrichtian-Danian / - | Tuff, marl, limestone, sandstone |
13 | KTck | Maastrichtian-Danian / - | Limestone (gray-red colored) |
14 | KTa | Maastrichtian-Danian / Ağıllar formation | Reefal limestone, sandy limestone |
15 | Kçb | Maastrichtian / Çayırbağ formation | Dacite, rhyolite, rhyodacitic lava and pyroclastics |
16 | Kça | Campanian–Maastrichtian / Çağlayan formation | Basaltic, andesitic lava and pyroclastics, mudstone, sandstone |
17 | Kk | Santonian / Kızılkaya formation | Rhyodacitic, dacitic lava and pyroclastics |
18 | Kç | Turonian-Coniacian / Çatak formation | Basalt, andesitic lava and pyroclastics, clayey limestone, marl, siltstone, claystone |
Landslide inventory
Landslide conditioning factors
Factor | Min | Max | Sub-classes | Reference |
---|---|---|---|---|
Altitude (m) | 100 | 3370 | 1: 100–427, 2: 427–754, 3: 754–1081, 4: 1081–1408, 5: 1408–1735, 6: 1735–2062, 7: 2062–2389, 8: 2389–2716, 9: 2716–3043, 10: 3043–3370 | Chen et al. 2017 Kilicoglu 2021 Akinci 2022 Yavuz Ozalp et al. 2023 |
Aspect | - | - | 1: Flat, 2: North, 3: Northeast, 4: East, 5: Southeast, 6: South, 7: Southwest, 8: West, 9: Northwest | Sun et al. 2022 He et al. 2023 Vega et al. 2023 Yavuz Ozalp et al. 2023 |
Distance to drainage (m) | 0 | 928 | 1: 0–100, 2: 100–200, 3: 200–300, 4: 300–400, 5: 400–500, 6: 500–600, 7: 600–700, 8: 700–800, 9: 800–928 | Akinci 2022 Sun et al. 2022 Arabameri et al. 2020 Yavuz Ozalp et al. 2023 |
Distance to Faults (m) | 0 | 7723.34 | 1: 0–1000, 2: 1000–2000, 3: 2000–3000, 4: 3000–4000, 5: 4000–5000, 6: 5000–6000, 7: 6000–7000, 8: 7000–7723.34, | Feizizadeh et al. 2014 Akinci et al. 2020 Akinci 2022 Yavuz Ozalp et al. 2023 |
Distance to Roads (m) | 0 | 2197.84 | 1: 0–200, 2: 200–400, 3: 400–600, 4: 600–800, 5: 800–1000, 6: 1000–1200, 7: 1200–1400, 8: 1400–1600, 9: 1600–1800, 10: 1800–2197.84 | Akinci 2022 Zhang et al. 2022a He et al. 2023 Yavuz Ozalp et al. 2023 |
Lithology | - | - | Explained in Table 1 | |
Land cover | - | - | 1: Water, 2: Trees, 3: Grass, 5: Crops, 6: Scrub/shrub, 7: Built Area, 8: Bare ground, 9: Snow/Ice, 10: Clouds | Lv et al. 2022 Roy et al. 2023 Yavuz Ozalp et al. 2023 Yu et al. 2023 |
Plan curvature | -24.50 | 33.33 | 1: -24.50 – -0.001, 2: -0.001 – 0.001, 3: 0.001 – 33.33 | Kilicoglu 2021 Akinci 2022 Yavuz Ozalp et al. 2023 |
Profile curvature | -40.17 | 36.42 | 1: -40.17 – -0.001, 2: -0.001 – 0.001, 3: 0.001 – 36.42 | Kilicoglu 2021 Akinci 2022 Yavuz Ozalp et al. 2023 |
Slope (o) | 0 | 76.33 | 1: 0–5, 2: 5–10, 3: 10–15, 4: 15–20, 5: 20–25, 6: 25–30, 7: 30–35, 8: 35–40, 9: 40–45, 10: 45–76.33 | Kilicoglu 2021 Akinci 2022 Sun et al. 2022 Yavuz Ozalp et al. 2023 |
Slope length | 0 | 3577.64 | 1: 0–56.12, 2: 56.12–140.30, 3: 140.30–252.54, 4: 252.54–392.84, 5: 392.84–561.20, 6: 561.20–785.68, 7: 785.68–1080.31, 8: 1080.31–1487.18, 9: 1487.18–2090.47, 10: 2090.47–3577.64 | Hong et al. 2015 Akinci 2022 Ghasemian et al. 2022 Yavuz Ozalp et al. 2023 |
TPI | -70.88 | 79.58 | 1: -70.88 – -21.32, 2: -21.32 – -13.65, 3: -13.65 – -7.75, 4: -7.75 – -3.03, 5: -3.03 – 1.10, 6: 1.10 – 5.23, 7: 5.23 – 9.95, 8: 9.9.5 – 15.85, 9: 15.85 – 25.29, 10: 25.29 – 79.58 | Arabameri et al. 2020 Sahin 2020 Akinci 2022 Yavuz Ozalp et al. 2023 |
TWI | 2.26 | 26.18 | 1: 2.26–4.79, 2: 4.79–5.92, 3: 5.92–6.86, 4: 6.86–7.80, 5: 7.80–8.92, 6: 8.92–10.33, 7: 10.33–12.11, 8: 12.11–14.46, 9: 14.46–18.12, 10: 18.12–26.18 | Sahin 2020 Kilicoglu 2021 Akinci 2022 Yavuz Ozalp et al. 2023 |
Machine learning algorithms used in the study
Random forest (RF)
Extreme gradient boosting (XGBoost)
Light gradient boosting machine (LightGBM)
Categorical boosting (CatBoost)
Multicollinearity analysis
Performance assessment metrics
Metric | Equation | Description |
---|---|---|
Overall accuracy (OA) |
\(OA= \frac{TP+TN}{TP+TN+FP+FN}\)
| The ratio of landslides and non-landslides that are correctly classified. This shows how well the landslide model works (Ghasemian et al. 2022) |
Precision |
\(Precision= \frac{TP}{TP+FP}\)
| Precision, also called the positive predictive value, is the fraction of relevant instances (TP) amongst the retrieved instances (Azarafza et al. 2021) |
Recall |
\(Recall= \frac{TP}{TP+FN}\)
| Recall (or sensitivity) is the measure at which pixels corresponding to landslides are correctly classified as a landslide having occurred. Its value is ideal as long as it is closer to 1 (Bravo-López et al. 2022) |
F1-score |
\(F1-score=2 x\frac{Precision\;x\;Recall}{Precision\;+\;Recall}\)
| |
AUC-ROC |
\(True\;positive\;rate\;(TPR)= \frac{TP}{TP\;+\;FN}\)
\(False\;positive\;rate\;(FPR)= \frac{FP}{FP\;+\;TN}\)
| The ROC curve plots the FPR on the X axis and the TPR on the Y axis. It shows the trade‐off between the two rates (Pourghasemi et al. 2012). The area under the ROC curve (AUC) is an indicator to check the prediction performance of the model (Yilmaz 2009) |
RMSE |
\(RMSE= \sqrt{\frac{1}{N}} \sum\limits_{i=1}^{N}{({y}_{obs}-{y}_{pred})}^{2}\)
|
SHapley additive exPlanation (SHAP)
Results
Multicollinearity analysis
Landslide Conditioning Factors | VIF | TOL |
---|---|---|
Slope | 1.694949 | 0.589988 |
Aspect | 1.080109 | 0.925832 |
TWI | 3.918890 | 0.255174 |
TPI | 1.483154 | 0.674239 |
Altitude | 1.410740 | 0.708848 |
Plan Curvature | 1.301666 | 0.768246 |
Profile Curvature | 1.140360 | 0.876916 |
Distance to Drainage | 1.139817 | 0.877334 |
Distance to Faults | 1.076383 | 0.929037 |
Distance to Roads | 1.229125 | 0.813587 |
Lithology | 1.294081 | 0.772749 |
Land Cover | 1.031502 | 0.969460 |
Slope Length | 3.059941 | 0.326804 |
Landslide susceptibility mapping
Model | Susceptibility class | Landslide probability | PoA (%) | NLP | PLP (%) | FR |
---|---|---|---|---|---|---|
RF | Very low | 0 – 0.08 | 76.50 | 753 | 0.815 | 0.0107 |
Low | 0.08 – 0.267 | 9.39 | 2303 | 2.491 | 0.2653 | |
Moderate | 0.267 – 0.506 | 5.37 | 3566 | 3.857 | 0.7182 | |
High | 0.506 – 0.773 | 4.21 | 10,829 | 11.714 | 2.7824 | |
Very high | 0.773 – 1.000 | 4.53 | 74,995 | 81.123 | 17.9079 | |
CatBoost | Very low | -0.816 – -0.057 | 15.95 | 6 | 0.007 | 0.0004 |
Low | -0.057 – 0.101 | 55.91 | 29 | 0.031 | 0.0006 | |
Moderate | 0.101 – 0.335 | 15.29 | 320 | 0.346 | 0.0226 | |
High | 0.335 – 0.677 | 6.90 | 4309 | 4.661 | 0.6755 | |
Very high | 0.677 – 1.311 | 5.95 | 87,782 | 94.955 | 15.9588 | |
XGBoost | Very low | -1.214 – -0.093 | 8.77 | 127 | 0.138 | 0.0157 |
Low | -0.093 – 0.059 | 61.39 | 490 | 0.530 | 0.0086 | |
Moderate | 0.059 – 0.288 | 18.47 | 2165 | 2.342 | 0.1268 | |
High | 0.288 – 0.658 | 6.44 | 10,517 | 11.376 | 1.7665 | |
Very high | 0.658 – 1.561 | 4.93 | 79,147 | 85.614 | 17.3659 | |
LightGBM | Very low | -0.811 – -0.063 | 13.64 | 158 | 0.171 | 0.0125 |
Low | -0.063 – 0.099 | 63.67 | 968 | 1.047 | 0.0164 | |
Moderate | 0.099 – 0.361 | 12.26 | 2060 | 2.228 | 0.1817 | |
High | 0.361 – 0.721 | 5.18 | 7966 | 8.617 | 1.6635 | |
Very high | 0.721 – 1.488 | 5.25 | 81,294 | 87.937 | 16.7499 |
Performance assessment and comparison
Stage | Metrics | RF | CatBoost | XGBoost | LightGBM |
---|---|---|---|---|---|
Training | Accuracy (%) | 82.556 | 91.948 | 92.885 | 92.992 |
Precision | 0.82656 | 0.92143 | 0.92917 | 0.93124 | |
Recall | 0.82553 | 0.91942 | 0.92884 | 0.92996 | |
F1-score | 0.82541 | 0.91941 | 0.92883 | 0.92988 | |
RMSE | 0.3572 | 0.2598 | 0.2411 | 0.2437 | |
Validation | Accuracy (%) | 82.338 | 91.831 | 92.641 | 92.765 |
Precision | 0.82433 | 0.92024 | 0.92667 | 0.92913 | |
Recall | 0.82344 | 0.91845 | 0.92643 | 0.92755 | |
F1-score | 0.82326 | 0.91828 | 0.92640 | 0.92759 | |
RMSE | 0.3576 | 0.2628 | 0.2469 | 0.2481 |