EGMS+: A Machine Learning Framework for Projecting Future Land Subsidence and the Associated Structural-Damage Risk of Shallow-Foundation Buildings: A case study of Gouda and Krimpenerwaard municipalities, the Netherlands

Sustainable water and land management strategies to address land subsidence in Dutch built-up areas, as outlined in the backcasting approach of the Living on Soft Soil (NWA-LOSS) research programme, require robust projections of future subsidence and associated risks in the built-up area, such as the structural-damage risk of shallow foundation buildings, under different intervention water and land management strategies. While InSAR data, such as the ortho vertical displacement of ground surface from the European Ground Motion Service (EGMS), provides excellent records of recent vertical ground-surface deformation with millimeter accuracy, it is not a standalone tool that can be used to forecast future subsidence under varying future conditions. To bridge this gap, we introduce EGMS+, a machine learning framework that integrates the EGMS data with a Random Forest (RF) regressor to project future subsidence under various future conditions. The Random Forest algorithm was employed to learn the complex, non-linear relationships between EGMS mean annual velocity rates and a suite of relevant spatial predictors. These predictors consist of the mean lowest groundwater level, percentage of built-up area, percentage of old buildings, ground-surface elevation, Holocene soft-soil thickness, and percentage of grass cover within each 100 m grid cell across the built-up areas of Gouda and Krimpenerwaard municipalities in the Netherlands. The model achieved high predictive accuracy (R² = 0.73, Out-of-Bag score OOB = 0.73, Mean Absolute Error MAE = 0.095 mm/year) on five years of data (2019–2023). For the structural-damage risk assessment, we use a fragility curve developed by the NWA-LOSS team at TU Delft, which defines the probability of slight structural damage as a function of 5 mm crack width. This curve was used to compute building-specific structural-damage probabilities by integrating differential settlement with the short-side dimension of each building unit in the study area. Essentially, the EGMS+ framework enables future scenario projection by simulating how changes in these predictors affect future subsidence. This capability can be demonstrated by projecting future subsidence and associated risks, such as the structural-damage risk of shallow foundation buildings under several NWA-LOSS targeted future states, such as those involving raised water tables and intervention targeting shallow-foundation building units. This EGMS+ framework provides quantitative estimates of the effectiveness of various mitigation strategies, offering a powerful, dynamic decision-support and spatial planning tool that can evaluate and prioritize sustainable pathways of addressing land subsidence in the Dutch built-up areas.