Multiobjective Optimization of Landslide Rheological Parameters Based on Random Forest and Nondominated Sorting Genetic Algorithm II

Abstract Rheological parameters (e.g., pore pressure coefficient <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" alttext="r Subscript u" display="inline" overflow="scroll"> <mml:msub> <mml:mrow> <mml:mi>r</mml:mi> </mml:mrow> <mml:mrow> <mml:mi>u</mml:mi> </mml:mrow> </mml:msub> </mml:math> and dynamic friction angle <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" alttext="phi" display="inline" overflow="scroll"> <mml:mi>ϕ</mml:mi> </mml:math> ) are necessary input for landslide runout evaluation. However, due to high randomness of initiation, propagation, and accumulation, it is impossible to directly measure rheological properties. This study proposes an optimization method for calibrating landslide rheological parameters based on random forest (RF) and Nondominated Sorting Genetic Algorithm II (NSGA-II). Firstly, a dynamic numerical model of landslide is established using three-dimensional model DAN (DAN3D). Then, the lateral accumulation width and the longitudinal runout distance at different spatial locations across the deposit are integrated to calibrate the rheological parameters. Next, the NSGA-II algorithm is employed to obtain the Pareto solutions of rheological parameters by minimizing the prediction errors of lateral and longitudinal runout characteristics at each location. To avoid extensive repetitive DAN3D simulations within NSGA-II, RF-based surrogate models are established to estimate the lateral and longitudinal runout characteristics at different spatial locations. The final optimal rheological parameters are objectively selected from Pareto solutions by the entropy weight method (EWM). Finally, the proposed approaches are illustrated and validated using two landslides that occurred in Gansu Province, China. Results indicated that rheological parameters optimized by the proposed approaches not only can accurately characterize the endangered area of historical events, but also provide reasonable forward prediction of runout behavior of similar landslide events with desired accuracy. RF surrogate models can remove the computational burden within NSGA-II. Integrating both the lateral accumulation width and longitudinal runout distance into the calibration of rheological parameters can improve the prediction accuracy of landslide endangered area.