Visualisation and evaluation of flood uncertainties based on ensemble modelling

Abstract

This study evaluates how users incorporate visualisation of flood uncertainty information in decision-making. An experiment was conducted where participants were given the task to decide building locations, taking into account homeowners' preferences as well as dilemmas imposed by flood risks at the site. Two general types of visualisations for presenting uncertainties from ensemble modelling were evaluated: (1) uncertainty maps, which used aggregated ensemble results; and (2) performance bars showing all individual simulation outputs from the ensemble. Both were supplemented with either two-dimensional (2D) or three-dimensional (3D) contextual information, to give an overview of the area.The results showed that the type of uncertainty visualisation was highly influential on users' decisions, whereas the representation of the contextual information (2D or 3D) was not. Visualisation with performance bars was more intuitive and effective for the task performed than the uncertainty map. It clearly affected users' decisions in avoiding certain-to-be-flooded areas. Patterns to which the distances were decided from the homeowners' preferred positions and the uncertainties were similar, when the 2D and 3D map models were used side by side with the uncertainty map. On the other hand, contextual information affected the time to solve the task. With the 3D map, it took the participants longer time to decide the locations, compared with the other combinations using the 2D model.Designing the visualisation so as to provide more detailed information made respondents avoid dangerous decisions. This has also led to less variation in their overall responses.

Color map design for visualization in flood risk assessment

“…A recent example of this approach is presented by Höllt et al (2015) who showed ensembles of ocean predictions in geographical context views, which are complemented with statistical graphs linked to marked geographical positions. Similarly, linked views have also been proposed by Lim et al (2016) to explicitly present individual results from flood simulation ensembles which form the base for aggregated flood uncertainty. …”

“…The varied parameters were: a) resolution of the digital elevation model (DEM); and, b) the Manning's roughness coefficient (n). For details concerning the modeling, we refer the reader to Lim et al (2016). Figure 5 shows binary flood maps of a few simulations from the ensemble.…”

“…To compare the usability of our new color map design, we used in this study the same flood scenario as used previously in a study by Lim et al (2016). The study area comprises a 2 km section of river Testebo located nearby Gävle, Sweden (see Figure 1 simulated with CAESAR-Lisflood (Coulthard et al 2013) to create an ensemble of 50 different simulations.…”

“…Examples of side-by side presentations of uncertain geographical data were presented e.g. in works by Aerts et al (2003) on urban growth, or in flood risk assessment as described by Lim et al (2016).…”

“…A comparison of three alternative flood visualizations in a specific study area (a). Visualization of probability of being flooded (b) (Mason et al 2009); visualization of uncertainty (entropy) in regarded to the predicted status of being flooded or not (c) (Horritt 2006); extrinsic visualization of results from ensemble studies for a selected position in the map (d) (Lim et al 2016).…”