Jong-Tae Kim

This study proposes a novel predictive approach for constructing landslide susceptibility maps by combining boosting models with hyperparameters optimized using Optuna. We used a well-defined spatial database comprising 128 landslides and 22 factors affecting landslides in the Eunsan village, Buyeo City, Chungcheong Province, Korea. Optuna is an open-source hyperparameter optimization framework for automating the determination of the best hyperparameters for machine learning (ML) models. The results demonstrate the excellent performance of three ML models optimized using Optuna. The LightGBM model exhibited the highest accuracy (a receiver operating characteristic value of 0.932) among the three models, followed by the XGBoost model (0.926) and the random forest model (0.917). These ML models highlight the significance of hyperparameter tuning and selection using Optuna. The efficiency, ease of use, and flexibility of Optuna make it a powerful tool for optimizing ML models and accelerating the research and development process in the field of ML and artificial intelligence. This study emphasizes the potential of these approaches as optimized tools for landslide susceptibility mapping.

Here, the results of Soil and Water Assessment Tool (SWAT) simulation are calibrated and validated using SWAT-CUP Premium, and the R² (coefficient of determination) values of simulated and observed flows are compared. According to the SWAT analysis of the Cheongsong Yongjeon stream basin, low accuracy of 0.40 was the R² value for simulated and observed flows, and the baseflow was calculated at an annual average of 46.659 m³/s. During the same period, after calibration and validation using SWAT-CUP Premium, the R² value of simulated and observed flows was improved to 0.71, and the evaluation indexes showed 0.51 for Nash–Sutcliffe efficiency (NS), 37.1 for percent bias (PBIAS), 0.73 for P-factor, and 0.33 for R-factor, indicating that the relationship between observed and simulated flows was improved. In addition, after calibration and validation, the baseflow was 56.951 m³/s, which was a more accurate value than the one before calibration and validation.

Here, SWAT simulation results are calibrated and validated using SWAT-CUP Premium, and R² (coefficient of determination) values of simulated and observed flows are compared. According to the SWAT analysis of Cheongsong Yongjeon-stream basin, a low accuracy of 0.40 was the R² (coefficient of determination) value for simulated and observed flows, and the base flow was calculated at an annual average of 46.659 m³/s. During the same period, after calibration and validation using SWAT-CUP Premium, the R² value of simulated and observed flows was improved to 0.71, and the evaluation indexes showed that NS was 0.51, PBIAS was 37.1, P-factor was 0.73, and R-factor was 0.33, indicating that the relationship between observed and simulated flows was improved. In addition, after calibration and validation, the base flow was 56.951 m³/s, which was a more accurate value compared to the one before calibration and validation.