Implementation of Citizens’ Observations in Urban Pluvial Flood Modelling

Sammanfattning: Damages caused by urban pluvial floods are believed to increase due to climate change and urbanization as more citizens are impacted in densely populated cities and extreme rainfalls occur more frequently with higher intensities. To prepare cities for these calamities, urban pluvial flood models are created to provide knowledge about how an extreme rainfall event could inundate the studied city. However, due to the scarcity of observation data from these rainfall events, flood models are seldom calibrated which is necessary to ensure their accuracy.  To improve the feasibility of calibrations an emerging data source was tested, crowdsourced images from citizens. Citizens’ observations have become increasingly available due to the increase of mobile phones and the development of social media enabling citizens to document and upload their observations to the public. Researchers could use these observations as an unconventional data source to calibrate models and reduce the knowledge gap regarding urban floods. The aim of this study was to explore and increase our understanding of how citizen’s observations can be used to calibrate an urban pluvial flood model. A case study about the cloudburst event in Malmö was conducted to study this topic. During that event, more than 100 mm of rain fell over a period of 6 hours in the city and caused 60 million euros of damages.  A total of 297 images depicting the flood caused by the cloudburst event were gathered from social media platforms, newspapers archives, and by inquiring citizens. Images were screened and analysed: water levels were estimated in 66 images and were then used to calibrate a 2D flood model. Furthermore, a sensitivity analysis of the calibrated results was conducted by calculating the RMSE for different subsets and compare it with the RMSE for the full dataset of citizens’ observations. This was done to study how different characteristics, such as timestamp and source as well as sample size and location of the images influences the calibration procedure. After the model was calibrated, the importance of spatial variability in the rainfall input was tested by comparing the flood model output between the spatially varied observed rainfall and a Chicago Design Storm rainfall, which lacks spatial variability.  It was concluded that images from citizens can be used to calibrate an urban pluvial flood model, but the procedure is time-consuming. However, it was also evident that images directly inquired from citizens reduced the time needed as their local knowledge could be integrated. The calibration procedure was also sensitive to the quality of the observations, especially when the images were photographed in relation to the rainfall event. Even though the study had limitations it demonstrates new possibilities to calibrate urban pluvial flood models.