Bristol Traffic Model

Unlike the Barcelona case study, Bristol did not have a pre-existing traffic model available for testing. Due to this, we looked to develop a micro-scale traffic model using the Open Source "Simulating Urban Mobility" (SUMO) software [20]. In contrast to the meso-scale model, the micro-scale model used for Bristol in this analysis simulates the movement of each individual vehicle separately as shown in Figure 3.

**Figure 3.** Example of SUMO micro-simulation.

For the road network the Bristol model has been built using OpenStreetMap (OSM) data [21]. Using SUMO's "netconvert" tool, OSM data was converted into a network file suitable for use within SUMO that contains road property information including but not limited to, the number of lanes, junctions, and traffic light locations. In the absence of traffic data, the traffic flows within the network were derived via generating Origin-Destination (OD) matrix database using data from the National-Receptor-Database

(NRD) [22]. For this process, we assumed vehicles start from either Residential locations or from the boundary of the road network extent (Network Entry Points) (assuming from outside the city) and that the journeys terminate either at a place of work or they leave the network at a boundary (Network Exit Points). Table 3 shows the composition of the origin-destination points with Table 4 showing the percentage distribution of the Origin and Destination locations accordingly. For the 'School' classification, some vehicles can use the school as a mid-way point in their journey to simulate school drop-offs during the morning.


**Table 3.** Origin Destination points within the road network.


**Table 4.** Origin Destination percentage distribution.

Using the spatial information of land-use points from the NRD, SUMO's 'Duarouter' tool was used to generate the OD catalogue of vehicular journeys within the network. An additional rule applied states that each journey must have a journey length equal to or greater than 1 km.

To simulate morning rush hour flows, a sigmoid style curve was used in determining the number of vehicles that were added to the network over time during the simulation. Figure 4 shows the number of vehicles being added to the network over time for a 5000 and 10,100 vehicle scenario respectively using the same curve function.

**Figure 4.** Number of vehicles added to the network over time for the 5000 and 10,000 traffic volume scenarios.

Figure 5 shows a comparison of the SUMO model outputs for dry weather scenarios, in which the different volumes of traffic present within the network over time whereby vehicles are only being added to the network (starting their journeys) between the hours of 5 am and 11 am. The result demonstrates that a doubling of vehicle journeys from 5000 to 10,000 vehicles within that period results in a seven-fold increase of the number of vehicles present within the network at its peak and a subsequent long tail section as the vehicles leave the network. For both scenarios, between the hours of 7 and 9 am there is a large increase in the number of vehicles being added to the network (approximately 4000 and 8000, respectively). In each scenario, the vehicles added to the network are subsequently removed from the network upon completion of their respective journeys.

**Figure 5.** Comparison of Traffic Volumes in the Network for 5000 journey and 10,000 journey scenarios.

The reason behind this substantial difference in the number of vehicles within the network is a result of traffic jam formation leading to the delay in the completion of vehicle journeys. Figure 6 highlights the variations in the Average Journey Speed (Equation (2)) of vehicles for both the 5000 and 10,000-journey scenarios. The majority of vehicles within the network during the 10,000-journey scenario are travelling at relatively low speeds (less than 10 km/h) whereas the average journey speed for traffic in the 5000-vehicle scenario is around 30 km/h.

$$Average\text{ }fourreg\text{ }Speed\_{Velocity} = \frac{Total\text{ }fourreg\text{ }Distance\_{Velocity}}{Total\text{ }fourreg\text{ }Time\_{Velocity}}\tag{2}$$

**Figure 6.** Average Journey Speed of Vehicles within the Network during the hours of 7 am and 9 am for 5000 and 10,000 journey scenarios.

Figure 7 shows a section within in the network at the peak times (determined in Figure 5) for both journey cases, where the 10,000 journey scenario presents a considerable worse congestion. Because of the congestion both here and in other sections of the network the time for the traffic to clear the network (complete their respective journeys) becomes dependent on the interval timing of the traffic lights and the settings in place within the model "time-to-teleport" to handle these obstruction issues. Traffic signal timing data is often unavailable and has a dominant influence on intersection capacity and network performance [23]. If the timings of these traffics lights are not configured correctly, under high volumes of traffic a stalemate scenario can occur whereby traffic can neither enter nor exit an area thus resulting in severe gridlock. To minimise gridlock scenarios within the traffic model (due to imperfections in the network design) and to deal with instances of vehicles becoming an obstruction, the teleportation rule is applied. In the examples shown in Figure 5, if a vehicle is stationary 40 min (flood duration of 30 min plus and arbitrary 10-min window), it is deemed to be assumed to be erroneously stuck and is teleported to the next edge within its route. Note that it is important that the teleportation rule has a time-limit set to be equal or greater than the duration of a flood event to prevent traffic teleporting past the blocked roads under flooded conditions.

**Figure 7.** Comparison of Traffic Volumes at peak times for (**A**) 5000 journeys and (**B**) 10,000 journey scenarios.

For the purpose of the study within this paper, we selected 5000 vehicle traffic scenario as a means of analysing the effects of flooding on obstructing or causing diversions to vehicles within the network and to minimise the implications of imperfections in the network configuration itself causing disruptions to traffic flows. Ten 6-h duration traffic scenarios were generated, where each scenario contains 5000 randomly selected journeys from the OD catalog whereby the Origin and Destination's match the percentage distribution outlined in Table 4. The synthetic sigmoid curve, as shown in Figure 4, was applied to stagger the start times of the vehicles during the simulation to generate a pseudo morning rush scenario. Figure 8 shows the variation/range of the number of vehicles present in the network over time across the ten scenarios, highlighting the two temporal peaks in traffic volumes within the network during the morning rush hours between 7 am and 9 am. Figure 9 shows the extent of the Bristol traffic model and the percentage route distribution of the ten scenarios (10 × 5000 journeys) whereby the higher percentage values correspond to the road sections where vehicles have traversed the most within the 10 scenarios. Here, we observe that within the modelled scenarios, there is a preference for vehicles to traverse the river section (that bisects the city) across the bridges.

**Figure 8.** Range of vehicle distributions from 10 scenarios under dry weather conditions in Bristol road network.

**Figure 9.** Percentage Route Distributions under dry conditions.
