Land Subsidence Monitoring and Building Risk Assessment Using InSAR and Machine Learning in a Loess Plateau City—A Case Study of Lanzhou, China

Abstract

As a representative city located in the Loess Plateau region of China, Lanzhou is affected by various environmental and engineering factors, such as precipitation, earthquake subsidence, and building construction, which all lead to frequent geological disasters. Obtaining information on land subsidence over a long time series helps us grasp the patterns of change in various types of ground hazard. In this paper, we present the results of using Interferometric Synthetic Aperture Radar (InSAR) to monitor land subsidence in the main urban area of Lanzhou from 26 October 2014 to 12 December 2021. The main influential factors leading to subsidence were analyzed and combined via machine learning simulation to assess the land subsidence risk grade distribution of a building unit. The results show that the annual average deformation rate in Lanzhou ranged from −18.74 to 12.78 mm/yr. Linear subsidence dominated most subsidence areas in Lanzhou during the monitoring period. The subsidence areas were mainly distributed along the Yellow River, the railway, and villages and towns on the edges of urban areas. The main areas where subsidence occurred were the eastern part of Chengguan District, the railway line in Anning District, and the southern parts of Xigu District and Qilihe urban area, accounting for 38.8, 43.5, 32.5, and 51.8% of the area of their respective administrative districts, respectively. The random forest model analysis results show that the factors influencing surface subsidence in Lanzhou were, in order of importance, precipitation, the distribution of faults, the lithology of strata, high-rise buildings, and the distance to the river and railway. Lanzhou experienced excessive groundwater drainage in some areas from 2015 to 2017, with a 1 m drop in groundwater and 14.61 mm surface subsidence in the most critical areas. At the same time, extensive subsidence occurred in areas with highly compressible loess ground and most railway sections, reaching a maximum of −11.68 mm/yr. More than half of the super-tall building areas also showed settlement funnels. The area at a very high risk of future subsidence in Lanzhou covers 22.02 km2, while the high-subsidence-risk area covers 54.47 km2. The areas at greatest risk of future subsidence are Chengguan District and Qilihe District. The city contains a total of 51,163 buildings in the very high-risk area, including about 44.57% of brick-and-timber houses, 51.36% of old housing, and 52.78% of super-tall buildings, which are at especially high risk of subsidence, threatening the lives and properties of the population. The deformation results reveal poor building safety in Lanzhou, providing an essential basis for future urban development and construction.

Land subsidence risk assessment along metro lines in Shanghai using InSAR monitoring data and optical images

“…Land subsidence has emerged as a significant global geo-environmental issue intensified by human activities (Pacheco-Martínez et al 2015;Bhattarai and Kondoh 2017;Bagheri et al 2021;Buffardi and Ruberti 2023;Xu et al 2023;Sadeghi et al 2023). Compared to violent geological activities such as earthquakes and volcanic eruptions, land subsidence progresses slowly, exhibiting characteristics such as proneness to occurrence, irreversibility, and gradual accumulation (Hu et al 2009;Nduji et al 2023;Zhang et al 2023c).…”

“…Hazard assesses the probability of damage resulting from land subsidence, with larger settlements generally indicating a higher likelihood of damage. Vulnerability denotes the potential losses that urban entities may face due to disasters, with more developed socio-economic areas typically experiencing greater losses from the resulting damages (Wang et al 2014;Bhattarai and Kondoh 2017;Xu et al 2023).…”

“…Regarding research methods, the analytic hierarchy process (AHP) stands out as the most common modeling method (Hu et al 2009;Huang et al 2012;Wang et al 2014;Lyu et al 2020;Zhang et al 2023c). Due to the development of geographic information system (GIS) and computer technology, machine learning methods such as random forest model (Xu et al 2023), boosted regression tree (Zhang et al 2023b), support vector machine (Mehrnoor et al 2023), artificial neural network (Ku and Liu 2023), and ensemble deep learning (Mohammadifar et al, 2023) are progressively being applied to risk assessment. Furthermore, methods including decision matrices (Akcin 2021;Zhao et al 2023), fuzzy logic (Mohebbi et al 2021Faryabi 2023), full consistency decision-making (Sadeghi et al 2023), evidence-weighted models (Oh and Lee 2010), and Dempster-Shafer (D-S) evidence theory (Chen et al 2014) have been effectively employed.…”

“…However, there is a growing consensus on the necessity and importance of conducting risk assessment focused specifically on urban infrastructure, as our understanding of the hazards of urban land subsidence deepens. (Liu et al 2015;Lyu et al 2020;Xu et al 2023). Concurrently, the index system has undergone further enrichment and refinement.…”

“…Regarding settlement data, it typically comprises cumulative settlement (Zhang et al 2023c;Nduji et al 2023) and settlement rates (Hu et al 2009;Lyu et al 2020); geological information encompasses aquifer thickness (Huang et al 2012;Faryabi 2023), deposition layer thickness (Mehrnoor et al 2023;Pacheco-Martínez et al 2015), water level (Rahmati et al 2019;Zhang et al 2023b), and surface elevation (Kim et al 2006;Faryabi 2023); socio-economic data include population density (Hu et al 2009;Bhattarai and Kondoh 2017), gross domestic product (GDP) (Zhang et al 2023c;Wang et al 2014), land use type (Hu et al 2009;Ku and Liu 2023;Mehrnoor et al 2023), and infrastructure density (Lyu et al 2020;Wang et al 2014). Moreover, factors such as climate (Mehrnoor et al 2023), precipitation (Xu et al 2023;Sadeghi et al 2023), distances to rivers, roads, and faults (Rahmati et al 2019;Mohammadifar et al 2023;Faryabi 2023), and investments in disaster prevention and reduction have been integrated into the index system due to their relevance to land subsidence. In the application of these factors, it is essential to screen and select appropriate evaluation indices based on the specific research object and foundation (Bhattarai and Kondoh 2017;Mohebbi et al 2021;Zhao et al 2023;Nduji et al 2023).…”