Ubiquitous data are here to improve flood management

We engineers have always been able to predict urban flooding, but we were never really sure how good our predictions were. Mostly, because observed flows or water levels for calibration and validation were lacking. Where was the water flowing? How high was the peak exactly? When and how fast did it arrive? Answers to such questions have been very challenging, but now this is changing. Tapping into new data sources and computation techniques, such as deep learning and computer vision methods, will open up new opportunities. An overview on recent developments.

Impacts of urban flooding will increase

Recent studies (e.g. 2014 UN World Urbanization Prospects) highlight for significant disruption to city activities that can be caused by natural hazards, such as flooding. Unless there are significant changes in how urban drainage infrastructures are managed and planned in the future, the impacts of these events are expected to increase due to the predicted changes in climate (e.g. more frequent and heavier storms) and urban migration that may contribute to an increase in impervious areas from having to construct new dwellings and urban facilities.


Urban Flooding in Portugal (Photo: Jornal i)


Urban flood modelling and the need for up-to-date and accurate data

Urban flood models are computer tools that can be used to assess flood impacts (e.g. generate flood risk maps) and to support flood rescue and protection actions in (quasi) real-time. Key to the implementation of these models for potential disaster management is the quality, specifically, the accuracy and “up-to-dateness” of the data:

  • Quality of rainfall data is crucial because rainfall is the main driver of flooding; 
  • High-resolution topography data are essential for urban flood modelling since overland flow is driven by gravity;
  • Land-use data play an important role in flood modelling as they affect the amount of infiltration and consequently the amount of overland flow, and
  • Although not regarded as input data, flood related data (e.g. water depth and flow velocity) are also crucial for model calibration, i.e. to adjust model parameters in such a way that the model produces accurate results. However, these data are difficult to obtain using traditional flow sensors.

Data are becoming available “everywhere” and in large amounts: an opportunity to improve urban flood modelling!

Urban flood modelling, and subsequently flood risk management, can strongly benefit from recent technological advancements in new sensors, data communication and computational algorithms. As we have demonstrated in recent studies, new technologies, such as unmanned aerial vehicles photogrammetry can be used for frequent acquisition of sub-metre urban topography (Leitão et al., 2016) and land-use maps (Tokarczyk et al., 2015), which might have a significant impact on how urban drainage systems and floods will be managed in future. The Internet of Things is also contributing for easier integration of countless traditional flow sensors with data transmission and storage platforms that can provide real-time information about the hydraulic condition of the systems – see, for example, the Urban Water Observatory project led by Dr. Frank Blumensaat from our Department.

Also extremely interesting and revolutionary is the vast amount of data that are continuously (24/7) being collected in many places (everywhere!?) for different purposes that can be transformed into useful information to improve urban flood modelling. In this category, Dr. Jörg Rieckermann (Department of Urban Water Management, Eawag) is investigating the use of mobile phone signals to estimate rainfall fields in urban areas. Also in this category, I have been exploring the value of surveillance and road traffic control cameras as well as images and videos uploaded on social media platforms (e.g. Instagram, Twitter and Flickr) to estimate flood velocity and flood depth. As demonstrated in Chaudhary (2019), by implementing new deep learning tools we could derive reasonable accurate water depth estimations from social media images. Based on a fruitful collaboration with Photrack AG, we have already found that it is possible to obtain overland flow velocity estimates from surveillance cameras using an advanced image velocimetry tools with errors smaller than 10% – details of this study can be found in Leitão et al. (2018).


Flood depth estimation from social media images (Chaudhary, 2019). Flood depth is estimated using deep learning methods that look into objects (e.g. persons, cars and buildings) in the water to assign a water depth class. The resolution of the water depth classes is approximately 15 cm.


Active citizen science is another way of collecting useful data to improve our understanding of urban floods. Examples include platforms to share personal weather station data and to collect flood depths (e.g. www.petabencana.id).



Locations of personal weather stations registered to the Weather Underground community (March 2020). Reproduced with permission from The Weather Company, LLC.


There are still challenges that need to be addressed...

The real potential to influence how urban flooding will be managed in future emerges from the combination of the new data and computation techniques, such as deep learning and computer vision methods. Nevertheless, there are still challenges that need to be considered. One is related to the accuracy of these newly available data. Quality control of multiple data sources will need to be accounted for. Accurate and high-resolution topography data can now be easily obtained but acquiring accurate high-temporal and spatial resolution rainfall data is still an unresolved challenge. The incorporation of all this variability of data sources and data quality into urban flood modelling will require the development of new or improvement of existing methods. Another important point that needs to be considered and which might also hinder the ample use of these new data sets is related to the data privacy issue. It is known that the use of image-based data sources, such as surveillance camera and social media platform videos and images, raise strong concerns among the public. In the case of social media images, geolocating them represents an extra challenge as, again due to the data privacy issue, most of the location data (e.g. geographic coordinates) are often deleted from the images metadata or never recorded in the first place. This will most likely require a shift in societal worldview that might take a few decades…

In our opinion, newly available data sources coupled with new computational algorithms will definitely open up new opportunities to modelling urban floods more accurately and contribute to being better prepared for urban floods, consequently protecting people and assets from these natural hazards.



Chaudhary, P., D’Aronco, S., Moy de Vitry, M., Leitão, J.P., Wegner, J.D. (2019). Flood-water level estimation from social media images. In ISPRS Annals Photo-grammetry, Remote Sensing and Spatial Information Sciences, IV-2/W5, 5-12. doi: 10.5194/isprs-annals-IV-2-W5-5-2019

Leitão, J.P., Moy de Vitry, M., Scheidegger, A., Rieckermann, J. (2016). Assessing the quality of Digital Elevation Models obtained from mini-Unmanned Aerial Vehicles for overland flow modelling in urban areas. Hydrology and Earth Systems Science, 20, 1637-1653. Doi:10.5194/hess-20-1637-2016

Leitão, J.P., Peña-Haro, S., Lüthi, B., Scheidegger, A., Moy de Vitry, M. (2018). Urban overland runoff velocity measurement with consumer-grade surveillance cameras and surface structure image velocimetry. Journal of Hydrology, 565, 791-804. Doi:10.1016/j.jhydrol.2018.09.001

Tokarczyk, P., Leitão, J.P., Rieckermann, J., Schindler, K., Blumensaat, F. (2015). High-quality observation of surface imperviousness for urban runoff modelling using UAV imagery. Hydrology and Earth System Sciences, 19, 4215–4228. Doi:10.5194/hess-19-4215-2015



Joao Paulo Leitao
Senior scientist (Group Leader)
view profile

Write new comment

Please make sure your comment is in accordance with our netiquette.


No comments found!