An operational IoT-based slope stability forecast using a digital twin

The paper investigates the combined use of real-time hydrological monitoring, publicly available meteorological data and hydrological and geotechnical numerical modelling, to develop data-driven models to forecast the stability of a slope. This study showcases a first attempt to integrate these critical aspects into a fully automatic Internet of Thing (IoT)-based local landslide early warning system (Lo-LEWS).

The paper uses a validated hydrological numerical model, back-calculated over real monitored conditions, to evaluate the slope stability. The factor of safety ($FoS$) was computed coupling the commercial package GeoStudio, using transient SEEP/W and Slope. The analyses were conducted for 5 different 1-year datasets encompassing both historical (2019–2020, 2021–2022, 2022–2023) and future projections (2064–2065, 2095–2096) of meteorological variables. Daily variation of hydrological and meteorological variables, along with vegetation indicators were used as inputs to train data-driven models, using polynomial regression (PR) and Random Forest (RF), to forecast daily $FoS$ values. The trained models proved to be effective and were employed to forecast slope stability for the rolling three days. To accurately forecast the $FoS$, it was essential to incorporate forecasted hydrological, meteorological and vegetation variables into the analysis. The hydrological variables used as inputs for the data-driven models are forecasted using an open-source Python package for the analysis of hydrogeological time series, called Pastas (Collenteur et al., 2019). This model uses historical and forecasted meteorological and vegetation conditions to, specifically, replicate and forecast the time series of volumetric water content (VWC) and pore water pressure (PWP). The forecasted hydrological variables from Pastas, the forecasted meteorological variables as well as Leaf Area Index ($LAI$) are used as inputs for the trained data-driven models to forecast the $FoS$ values.

Finally, a web-based platform (WBP) has been created that automatically runs once a day and perform the following actions: 1) fetches measured and forecasted data using APIs, 2) runs rolling three days forecast based on collected hydrological, meteorological and vegetation variables, and 3) sends the forecasted values back to the Norwegian Geotechnical Institute (NGI) data platform, NGI Live, making them available for real-time visualization in online dashboards. If $FoS$, VWC or PWP threshold values are exceeded, text messages and emails are sent to the system managers, enabling them to take appropriate actions. The successful implementation of this framework is the result of a collaborative effort across diverse expertise areas, including geotechnics, hydrology, meteorology, instrumentation, and informatics.