Geodata-based Machine Learning for real-time urban risk reduction systems

The project GEObyIT addresses the common need of two departments of the Oslo municipality (the Agency for Emergency Planning and the Water and Sewage Department) to assess natural (e.g., earthquakes, landslides) and man-made events (e.g., explosions, accidents) in an urban environment. We deployed seismic sensors at different locations within the city of Oslo. The sensors were connected for real-time data transmission and remote maintenance. The sensors were placed in the basement of private businesses, private houses, and public-school buildings. The first batch of 4 sensors was deployed with a dense layout in an area with quick clay in the sub-surface in the Eastern part of Oslo (Alna) to allow for near-surface structural monitoring using ambient noise and detecting of possible movements. More sensors were deployed closer to the city centre and in the western part of Oslo, close to a sub-surface blasting area for the construction area of an underwater cavity of water storage. Data analysis was done using three approaches: (1) automatic detection of metro trains, (2) automatic identification of outlier events such as construction and mining blasts, and (3) noise interferometry to monitor the near sub-surface in an area with quick clay. We use a supervised machine learning method trained with visually identified seismic signals on three sensors distributed along a busy metro track (1). Application to continuous data allowed us to reliably detect trains and their direction, without false alarms. Further development of this approach will be useful to sort out known repeating seismic signals or to monitor traffic in an urban environment. In approach (2) we aim to detect unusual seismic events using an outlier detection method. A convolutional autoencoder was trained to reconstruct the signal and detect anomalies. The outlier detection method is followed by locating these events if they are observed on multiple sensors. From several construction blast areas, we then select a cluster with most activity in the western part of Oslo and train a supervised classification method to be applied to detect similar events in continuous data. Finally, we apply seismic noise interferometry to close-by sensor pairs to measure temporal variations in the shallow ground (3). We observe clear seismic velocity variations during periods of strong frost in winter 2021/2022 and clear change associated to the strong rain periods during "Hans" in summer 2023. This opens up for the potential to detect also non-seasonal changes in the ground, for example related to instabilities in quick clay deposits located within the city of Oslo. The 2022 activities of WP2 were focused on fieldwork-based data collection from selected areas in the Oslo region that are adjacent to existing or proposed rail and road tunnels. Fault and fracture data were collected during two excursions. One goal was to assess the predictive potential of topographic lineaments with respect to “good” and “bad” rock, e.g., can lineament analysis help improve the siting and planning of future tunnelling projects. Some of these field data will be incorporated into a MS thesis focussing on lineament analysis and data that is being completed by Karolina Arctander at NTNU. A second goal was to study areas where known fault structures exist to better constrain their surface locations and surface expressions. A third goal was to begin testing a new structural database that is being developed at NGU. The database is intended to link topographic and geophysical (e.g., magnetic) lineament and field data within one database structure, to enable a better understanding of how topography is (or is not) controlled by fracturing and faulting of the bedrock. In WP3 we are interested to compute the seismic risk for Oslo and to do so we need information on hazard, vulnerability, and exposure. During 2021 and 2022, the exposure model has been defined for Oslo, identifying the main building typologies in the city. Fieldwork in specific areas has been carried out in winter 2020-2021 where pictures of façade were taken manually with the mobile phone and information on structural system and material data have been collected. After that, 5 building typologies have been identified: timber, unreinforced masonry, reinforced concrete, composite (steel-reinforced concrete) and steel. To recognize the building typologies for all Oslo’s buildings (around 135.000), we used a Convolutional Neural Network to automatically identify the different building typologies in the city of Oslo based on facade images taken from in-situ fieldwork and from Google Street View. In relation to the hazard component and soil amplification, fieldwork and horizontal to vertical spectral ration measurements have been conducted in some areas of Oslo during spring and summer 2021 to estimate average shear wave velocity in the upper 30 meters of the ground and depth to basement values.

We propose to develop and implement an autonomous, intelligent system for real-time monitoring and multi-risk assessment in urban areas using seismic, acoustic, and remote sensing data. The project GEObyIT addresses needs of two departments of the Oslo municipality. The Emergency Department requires improved capabilities for infrastructure monitoring and detection of unusual events such as accidents, explosions, and crimes based on data collected in line with the General Data Protection Regulation that prohibits usage of “sensitive data”. The Water and Sewage Department is currently involved in building a tunnel for securing the freshwater supply in Oslo. This requires improved mapping of surface features and monitoring for small earthquakes or other events which might be able to affect water flow paths. GEObyIT will address both needs by producing new layers of information based on seismic, acoustic, and remote sensing data, which will help to quickly locate incidents and other urban events, ready to be integrated into the existing alertness plan of the city, and to identify and classify potentially hazardous features on the Earth’s surface. While the employed data sets are different in nature, each serves the purpose of risk assessment and the proposed automatic processing pipelines will use a common methodological approach based on Machine Learning. The final product of GEObyIT will be tools for urban real-time alert and risk assessment, including hazard maps for predictive and operational purposes and a dashboard-like webpage including real-time location maps of events. This concept of providing novel data layers for future city management is extremely ambitious, but there is a huge potential to yield major gains and radical breakthroughs, including knowledge transfer and applicability on the global scale. The societal impact of these new layers is broad, for example by improving emergency response of the city authorities in case of accidents and crimes.