Heavy rain can disrupt an urban road network long after the water itself begins to recede.

Damaged links, reduced capacity and changing traffic patterns can push large numbers of vehicles onto a smaller number of usable routes, creating congestion and potentially concentrating safety risks.

A 2026 study in Reliability Engineering & System Safety proposes a recovery strategy designed to consider both network efficiency and road safety.

<h3>Repair Is More Than Reopening Roads</h3>

Researchers Yu Lin, Shengling Gao and Dongxu Chen examined how urban road networks could be restored after rainstorm-related disruption when connected autonomous vehicles, or CAVs, share roads with human-driven vehicles.

Many existing recovery models concentrate mainly on how quickly traffic performance can return. The researchers argue that this can overlook an important consequence of reconstruction: when damaged roads reopen in stages, drivers redistribute themselves across the network. Some restored links may then attract heavy flows before surrounding infrastructure has recovered, potentially creating concentrated areas of elevated crash risk.

To address this problem, the team developed a resilience-oriented optimisation framework that considers efficiency recovery, repair priorities and safety risk together.

<h3>A Selective Repair Strategy</h3>

At the centre of the model is what the researchers call a selective link-availability strategy, or SLS. Instead of automatically allowing full use of every road as soon as repairs progress, the approach can control when repaired links become available during the recovery period.

Safety risk was estimated using a crash-rate function based on the ratio between traffic volume and road capacity. The researchers also introduced a Value at Risk-based constraint intended to limit exposure to particularly high-risk conditions.

Road capacity after a rainstorm was not treated as perfectly predictable. Instead, the model incorporated multiple disturbance scenarios representing uncertainty in how much capacity would actually be restored during repair.

The optimisation simultaneously considered repair order and the allocation of repair crews. It was solved using simulated annealing combined with the method of successive averages.

<h3>Testing Two Road Networks</h3>

The researchers evaluated the framework using the Nguyen–Dupuis and Sioux Falls benchmark networks. The base Nguyen–Dupuis scenario contained 13 nodes, 19 links and four origin-destination pairs.

Compared with the baseline strategy, selective link availability produced a smoother recovery of network efficiency and a more balanced distribution of safety risk.

Adding an explicit safety constraint also changed which roads received repair priority. According to the researchers, risk-focused optimisation reduced the duration and concentration of high-risk network states, particularly when combined with selective control over link availability.

<h3>Automation Could Change Recovery</h3>

The simulations also examined different shares of CAVs in traffic. Under the study's capacity and volume-to-capacity assumptions, higher shares of connected autonomous vehicles were associated with stronger early-stage recovery and lower estimated safety risk.

Traffic demand had the opposite effect: greater demand increased both performance losses and exposure to risk. Adding repair crews accelerated recovery, although each additional increase in resources produced progressively smaller benefits.

These results come from numerical modelling rather than observations of an actual city recovering from a rainstorm, so they should not be interpreted as direct predictions of real-world crash rates. The safety estimates also depend on the study's chosen capacity and crash-risk formulations.

The research nevertheless highlights an important planning problem. Restoring roads as quickly as possible does not necessarily guarantee the safest recovery. Managing when links reopen, where repair crews are deployed and how traffic redistributes across a damaged network may be just as important as the physical reconstruction itself.