Land subsidence simulation based on Extremely Randomized Trees combined with Monte Carlo algorithm

Abstract

Land subsidence is a threat to the living safety of urban residents. Land subsidence simulation helps make prevention and conservation efforts more effective. This work introduces an innovative model for land subsidence simulation which combines Extremely Randomized Trees (ET) with the Monte Carlo algorithm to conduct quantitative analysis on the importance of different factors in land subsidence. Long-time series land surface deformation results in the eastern part of Beijing Plain were obtained by time series Interferometric Synthetic Aperture Radar (InSAR) monitoring technology based on 91 scenes of RADARSAT-2 data collected from Sept. 2009 to Jan. 2020. In addition, multi-source data, such as layered compressible layer and layered groundwater level were introduced to analysis the evolutionary mechanism of land subsidence. The results showed that the average annual rates of land subsidence at the following three typical subsidence bowls in the Beijing Plain, Chaoyang-Jinzhan, Dong Balizhuang-Da Jiaoting and Tongzhou-Song Gezhuang were −66.4 mm/a, −64.2 mm/a and −53.2 mm/a from 2010 to 2019, indicating a rather alarming subsidence issue. The most important feature affecting land subsidence is the groundwater level which contributed 67.6% to 81.8% to land surface deformation. The Monte Carlo simulation results indicated that land subsidence occurs at Dong Balizhuang-Da Jiaoting Subsidence Bowl is estimated to be the worst, with a 95% confidence interval differing by 95.7 mm/a between the upper and lower limits. Such work contributes to better understand the complex relationship between the land subsidence and influencing factors. And the simulation results provide different scientific means to facilitate efficient management and control of three land subsidence bowls.