Impact of Climate Change on Disruption to Urban Transport Networks from Pluvial Flooding

For a given scenario of rainfall (duration, intensity), terrain, and boundary conditions, a hydrodynamic model is employed to produce outputs for each time step of the simulation to give flood depths and velocities. The flood model used in this study is the City Catchment Analysis Tool (CityCAT), a two-dimensional hydrodynamic model developed to simulate pluvial inundation at high resolution that is already applied to and calibrated for a number of cities. Photos and records from the Toon Monsoon (Fig. 1) were used to validate the model for Newcastle (Glenis et al. 2010, 2013). Rainfall time series of precipitation intensity (considered uniform across the model domain in this study) during an event were propagated over the surface using a set of shallow-water equations. The surface water flows took into account building locations and footprints, permeability of the ground, and topography at a high resolution. Water depth and velocity were calculated dynamically throughout the simulation period and reported at each time step as a raster grid (in this study, at a resolution of 4 m) which can be used for further analysis. To reduce computational burden, the subsurface drainage network was not simulated as a dynamic network. The outputs from the flood model were used to calculate impacts on the transport network (Fig. 3).

Synthetic design storm events were generated following the standard U.K. procedure from the Flood Estimation Handbook (Robson and Duncan 1999) for the current scenarios and extrapolated to the 2080s epoch. To explore the implications due to climate change and potential future rainfall intensity changes, a series of uplifting factors were employed. Regional and global climate models are too coarse to fully resolve clouds, moisture, convection, and topography. Convective processes were therefore parametrized as a subgrid process leading to an inability to represent extreme subhourly precipitation (Ban et al. 2015). The computational expense of convection permitting models requires that runs are at the subregional scale if they are to simulate multiple decades (Prein et al. 2015). The uplift factors used here were derived from high-resolution climate simulations over the United Kingdom that are able to capture convective storm processes that give rise to pluvial flooding (Chan et al. 2014; Kendon et al. 2014). Dale et al. (2015) derived uplift factors from the high-resolution (1.5 km) climate model using the RCP8.5 climate scenario (IPCC 2014). This analysis projects that the intensity and total rainfall of short-duration events such as the Toon Monsoon will increase. These were not available in previous climate scenario reports (e.g., IPCC 2014) and recently have been made available to the water industry, but to the authors’ knowledge this is the first time they have been applied to understanding transport risks. The storms considered are summarized in Table 1.

The second element in the modeling framework is a simulation of the transport network in a geographic information system (GIS)–based accessibility model, as outlined in Ford et al. (2015). This model simulates journeys across a transport network, defined by spatial data of links and nodes, using generalized cost of travel to assess the shortest route between an origin and destination. Free-flow speeds on the links were defined using classes from the U.K. cost benefits analysis (COBA) model (DoT 2004) inferred from attributes in Ordnance Survey MasterMap data, and speed-flow curves were used to simulate congestion effects on those links (DoT 2004). These routes were then used for the assignment of trips from observed U.K. census journey-to-work flows using an iterative assignment routine (see De Ortuzar and Willumsen 2011) in order to assess the number of users along any road in the network.

A number of transport processes were represented at reduced complexity to ensure the model is computationally efficient. There is no stochastic variation in the speeds of the vehicles along each link; all traffic on a road link travels at either the maximum free-flow speed or at a reduced speed accounting for congestion. Minor residential roads were removed from the analysis because it was observed during the flood event of June 28, 2012 that the major roads were impacted to such an extent that minor roads were quickly overwhelmed by the volume of traffic and did not offer alternative route choices. Moreover, this also reflects the lack of perfect knowledge that many road users have, being unaware of alternative minor residential roads away from major or regular routes. Only commuting journeys were simulated because disruption during the morning or evening peak has the potential for the greatest economic disruption (Hallegatte and Przyluski 2010). The timing of the June 28, 2012 event was during the evening peak, and so such a situation has been observed.

The transport model was applied to simulate all commuting journeys across the metropolitan county of Tyne and Wear ($538{\mathrm{km}}^2$). Middle Layer Super Output Area (MSOA) population-weighted centroids for the 2011 U.K. census (freely available from the Office for National Statistics, United Kingdom) were used as origins and destinations for a total of 43,681 of these journeys, with routes computed for baseline and flood conditions. Following the same process as Ford et al. (2015) the model was validated for baseline conditions against census journey flows and observations from automatic traffic counters to ensure that the busiest simulated roads corresponded to the busiest observations. However, only a small proportion of roads have a traffic counter, and commuting flows make up only approximately 20% of flows on the road network (DoT 2016), so the validation can only be partial. The impact of flooding was considered by integrating the depth-disruption vulnerability function described in the following section with information on the flood hazard to recalculate (lower) traffic speeds, and where there was deep flood water, to block the road entirely.

The third and last stage involved translating flood depth, via the transport network model, into journey travel time increase and ultimately into economic cost. Chen et al. (2016) considered driving safety as a type of flood impact and recognized that this is related to the depth of flooding. Pregnolato et al. (2016a) developed a depth-disruption function by synthesizing experimental reports (e.g., Ong and Fwa 2008; Galatioto et al. 2014), safety literature (e.g., Chung and Recker 2012), experimental data (e.g., Boyce 2012), analysis of videos of cars driving through floodwater, and expert judgment (e.g., The Automobile Association 2016). Data were from the EU, Canada, and Australia and for asphalt roads, and therefore were comparable (Fig. 3). This moves beyond the crude assumption that the road is either open or closed according to a single arbitrary depth threshold, which is consistent with observations from real flood events that drivers travel slowly through floodwater, including during the Toon Monsoon event (Newcastle City Council 2013).

A baseline transport scenario was initially generated by running the transport model under normal settings (i.e., speeds defined by the speed/flow curves), and then multiple disruption and adaptation scenarios were evaluated. The hazard maps for each flood event showing water depths across the city (Fig. 6) were integrated with the vulnerability curve, enabling the speed reduction according to the depth of floodwater to be calculated for each road link. The uncertainty bounds in Fig. 3 capture a range of vehicle sizes, but with incomplete information available on vehicles in Newcastle and their individual routes, the central estimate of the depth-disruption function was applied to each road link. When network characteristics were modified by hardening one or more links, traffic flows were recalculated and disruptions assessed in terms of additional journey time and delays. This allows an assessment of the effectiveness of one or more adaptation options in reducing network-level disruption from flooding.

By overlaying the water depth from flood simulations onto the road network, vehicle speeds and subsequently journey travel time can be recalculated. A single metric of person-hours delay (PHD) to measure the city-wide disruption is calculated by aggregating all of the delays to each passenger journey across the networkwhere $A^S$ = aggregate journey time across the $N$ origin and destination zones in the city for scenario $S$ (BA is the baseline scenario); and $T$ = number of trips and $C$ = cost (in time or money) between each origin and destination. Other metrics, such as percentage of roads flooded or severity of damage to infrastructure, could be used to assess the impact, but during the Toon Monsoon the most notable and least-understood impact was the loss of road network performance. The resultant delays, due to rerouting and speed reduction, are used to compare the impacts of scenarios and the benefits of adaptation.

Delays can be converted into monetary terms using the value of time (VoT) (Ford et al. 2015; DoT 2014b; De Ortuzar and Willumsen 2011). The additional time required by journeys when the network systems is disrupted means an overall economic cost (e.g., business interruption) which through the use of the VoT can be converted into monetary terms accounting for the time of delay and vehicle operating costs (Ford et al. 2015). The cost per vehicle delayed $C_{\mathrm{veh}}$ is calculated bywhere $\mathrm{\Delta }T$ = variation in journey time (h); and VoT = value of time ($\pounds /\mathrm{h}$). The value of commuting time is properly defined as nonworking travel time, which differs from working time (4 times higher) for business trips or journeys made in the course of work, because commuting trips usually use the commuter’s own time. Commuting time includes “all non-work journeys purposes, including travel to and from work” (DoT 2014b). The VoT used in the model was the 2010 market price for commuting time per person, which was £6.81 per hour (US$10.56 in June 2012 prices).

Although this VoT measure is defined for use in normal road conditions, it can be considered a low bound to the level of economic cost, because the VoT is likely to be higher during disruptive events (Jenelius et al. 2011; Mattsson and Jenelius 2015). Other impacts could also be quantified, such as the increase in air pollution due to vehicle emissions and a higher total ${\mathrm{CO}}_2$ for the journey (de Palma and Lindsey 2011; Mao et al. 2012), or social impacts in terms of driver health and well-being (Quah and Boon 2003; Abu-Lebdeh 2015). Using the census journey-to-work data, the individual delay for journeys between each pair of locations can be multiplied by the observed number of commuting trips between those points to give a combined person-minute delay for those journeys [Eq. (1)]. This captures the wider effects of the delay to transport links, weighting the delay to journeys by the number of people currently using those portions of the transport network.

The benefit of climate adaptation actions are usually realized over multiple years. The net present value (NPV) of the benefits in terms of risk reduction is one criterion for deciding which action is more cost effective. Net present value computes the long-term costs and benefits, discounted to present-day rates to account for inflation. The NPV of the benefits in terms of risk reduction, ${\mathrm{NPV}}_r$, is calculated by summing the disruption cost, $D(x)$, and likelihood, $\rho (x)$, of a range of flood events:

In line with HM Treasury (2013) guidelines, a life span, $N$, of 50 years for infrastructure and a discount rate, $r$, of 3% were used.

Potential changes to rainfall intensity over Newcastle benefit from high-resolution climate model simulations (Kendon et al. 2014) that have been analyzed using the approach described by Dale et al. (2015) in Table 2. Low, central, and high projections of change are made to a range of rainfall events—showing, for example, that a 30-mm, 1-h storm in the present epoch would be 36–42 mm, with a central estimate of 38.4 mm, in the 2030s. The central estimate is used in this analysis.

In order to ensure resilience of urban areas in the future, infrastructure must be adapted to cope with such change in future extreme events (IPCC 2014). The vulnerability of the road transport network of Newcastle upon Tyne to pluvial flooding was assessed by Pregnolato et al. (2016b), with multiple locations at risk from rainfall-induced flooding (and therefore in need of adaptation) identified in the city. One strategy to make infrastructure in those locations more robust is to intervene with measures of hard engineering, such as improved drainage or raising the level of the link. Such strategies are referred to in this paper as link hardening. When a link is considered hardened in this study, it means that such a link has been made completely invulnerable to flooding. There are many options available for hardening a link (e.g., better drainage or road elevation), and the results for one such option are presented in the “Results” section, along with a simple cost–benefit analysis.

During the analysis process, network metrics are used to select the critical links in the network and target adaptation options. The most at-risk locations in the road network are identified through a matrix (Larsen et al. 2010; Naso et al. 2016; Pregnolato et al. 2016b) combining the hazard, i.e., the depth of water on the road, and the exposure, i.e., the average daily traffic flow along the road (Fig. 4). Network measures, such as betweenness centrality, have been used in other studies to identify critical locations (Pregnolato et al. 2016b) based on their topological importance, although that approach has not been adopted here.

The matrix was applied to identify and rank the criticality of road stretches in Newcastle’s road network. The six most critical stretches, where both the exposure (i.e., traffic flow) and hazard (i.e., water depth) were in the highest categories, were selected for analysis of adaptation options. Road stretches can comprise a number of neighboring links and nodes (e.g., protective measures would not be taken for just one spur of a roundabout). These stretches, shown in Fig. 5, in order of criticality are

Eleven different adaption scenarios were considered and simulated using the framework. Each of the six options was tested independently (i.e., LH_A,…,LH_F). Five additional scenarios considered the cumulative effect of adaptation portfolios that included the next most critical link (i.e., LH_A, LH_AB,…,LH_ABCDEF).

The methodology in this paper was developed with standard tools and practitioner appraisal methods in mind. For example, any flood or transport model could be used. Similarly, the calculation of the generalized cost of travel and value of time is in line with U.K. government guidance, ensuring that the results are of direct value to the policy appraisal process (DoT 2004; EA 2010; HM Treasury 2013; DoT 2014b), but this stage of the calculation is readily adapted to suit other national approaches (e.g., FHWA 2001).

The framework provides a means of assessing the benefits, in terms of reduced disruption, of flood risk management and highway drainage interventions. To date this has not be considered in flooding appraisals in such a comprehensive way. Moreover, the method provides a mechanism for city-wide screening of priority locations for climate change adaptations based upon analysis of road networks and traffic properties. The results for Newcastle show that just two interventions provide a substantial reduction in transport disruption. This may seem unintuitive; however, both are important roads for commuters and are susceptible to surface water flooding, but, crucially, if they are blocked, alternative routes are limited and markedly longer. This rate of return from two adaptation interventions is not expected to be the same in every city because the transport network structure and level of redundancy, travel patterns, and topographies will be different, but application of the criticality analysis is shown to prioritize investment interventions effectively. The analysis also shows the importance of considering a range of events because of the different flood footprint and depths, which alter the viability and possible travel speed of different routes. Furthermore, by considering multiple events it is possible to identify a balance of the costs and benefits from the size of adaptation at each road stretch. Although designing for more extreme events is more costly, the returns are greater and provide greater resilience to projected changes in rainfall.

Because this analysis only considers the benefits in terms of disruption, and not other benefits associated with urban flood management, it is likely that the overall returns would be higher. Previous work by Pregnolato et al. (2016a) showed that green infrastructure provides notable benefits across the city. However, because these are more diffuse interventions than the targeted adaptation options considered here, although they provide a city-wide reduction in flood levels they are less effective at mitigating the impacts on the transport system. There are a range of other options that could be tested in future work—e.g., upgrading the urban drainage network. Extension of the impact analysis to use land-use transport modeling to consider future urban development and socioeconomic change would provide future traffic flows and enable testing of other adaptation options, such as a modal shift away from private cars, increased home working, or adding redundancy to the road network.

Although the use of the high-resolution climate simulations (Dale et al. 2015) provides a marked improvement on other transport impacts studies, these currently use only a single climate scenario. Because the scenario modeled by Dale et al. (2015) is a higher-emissions scenario, the change in flood frequency in the 2080s might be considered an upper estimate, although Sanford et al. (2015) note that global emissions are proceeding along, and even exceeding, this trajectory.

The transport model used here provides a low-complexity representation of driver behavior—e.g., it does not consider vehicle-to-vehicle interactions at road junctions, and assumes that travelers have perfect knowledge of the network and associated journey disruptions. This provides a computational advantage while still capturing the macroscale transport interactions that this work seeks to understand; however, the disruption estimates are therefore expected to be a lower bound.

An agent-based transport model could be used to better capture micro effects, albeit at additional computational cost. Dynamically linking an inundation and transport model would allow a simulation of disruption over the course of the flood event and an understanding of when transport patterns return to normal after the rain, which would vary according to the magnitude of the flood. Similar approaches have already been implemented to understand the risk of drowning (e.g., Dawson et al. 2011). The effect of blockages such as broken-down vehicles that continue to obstruct roads after waters recede could also be captured. However, this is unlikely to be significant in the context of an extreme event because the transport model used here does not route vehicles through very deep water, and although these obstructions may cause local disruption they are unlikely to be significant in the context of a city-wide flood. Nevertheless, the framework applied here enables assessment of the peak disruption impact, thereby providing important information for policy makers to determine the benefits of adaptation options on the transport network.

Validation of transport models during disruptive events is challenging because weather extremes are relatively infrequent, and therefore provide limited observations. The vulnerability curve used here captures within uncertainty bounds the observations from a range of studies based on data from similar asphalt roads for a range of vehicles. The model replicates the types of delays reported by commuters and accurately captures the areas that flooded during the Toon Monsoon event (Newcastle City Council 2013). However, the exact impact of any particular event is sensitive to initial conditions such as the number of vehicles on the road and driver choices about whether to delay their journey. The presence of traffic sensors enables the automatic counting of vehicles, at least for a portion of the road network. Incorporation of this data and other emerging data sets from driver monitoring systems offers the potential for a far richer understanding of vehicle response during disruptive events and should greatly improve the validation and calibration of this type of model (Jenelius and Mattsson 2012; Batty 2013; Kermanshah et al. 2014; Osei-Asamoah and Lownes 2014; Kermanshah and Derrible 2016).

The models and data used in this study are readily available for many locations around the world, enabling the approach to be readily transferred to other cities. Moreover, although this study focuses on the flood risk to the road network, the framework could be applied to other weather-related phenomena (e.g., wind gusts, heat waves) and other transport networks (e.g., the railway network). This would require an appropriate hazard model and a relationship between the hazard and disruption (e.g., a rail-buckling function that is conditional on temperature) to support the system-wide analysis of their direct and indirect impacts.