Flooded Roads Can Trap Villages Within Hours. Can Dynamic Routing Change That?

The research focused on the Huayang River Flood Retention Area, a 1,721-square-kilometre region spanning parts of Hubei and Anhui provinces that contains lakes, low-lying farmland, settlements and an extensive road network.

Flood evacuation plans can quickly become unsafe after a levee breaks because rising water can cover roads that were expected to stay open. A new paper titled "Research on a Dynamic Flood Risk Evacuation Model Based on Flood Information and Road Network Reconstruction", published in the journal Sustainability, explores a system that repeatedly recalculates evacuation routes as flooding spreads across communities and transport networks. Shouwei Shang, Yong Liu and colleagues developed the research and tested the approach in the Huayang River Flood Retention Area along China's Yangtze River.

The researchers linked changing flood conditions to road accessibility so the evacuation system could recognize flooded road sections, adjust expected driving speeds and identify another route or shelter when the original choice became unsafe. Their results showed that fixed evacuation planning can leave far more communities without usable roads than a system that responds to changing water conditions.

The research focused on the Huayang River Flood Retention Area, a 1,721-square-kilometre region spanning parts of Hubei and Anhui provinces that contains lakes, low-lying farmland, settlements and an extensive road network. Its geography complicated evacuation because floodwater can move through lake systems, channels and polders at different speeds, creating very different evacuation windows for communities only a few kilometres apart. Researchers examined an extreme 300-year flood scenario and used levee breaches at Huikou, Wangzhou and Yangwan to understand how deep water could become, where it would travel and when individual roads might fail.

Roads Are Rebuilt Digitally as Floodwater Moves

The model combines one-dimensional and two-dimensional hydrodynamic simulations with a digital representation of the road network, allowing flood depth and movement to influence route calculations at different stages of the emergency. Road intersections are represented as network nodes and road sections as connecting edges, with travel time as the weight that determines the quickest escape. The Dijkstra shortest-path algorithm then searches the updated network for the fastest available route instead of assuming that the road conditions present at the start of the flood will remain unchanged.

Water depth determines whether vehicles can continue using a road because the researchers assumed that evacuation would rely mainly on large buses and trucks. A road becomes impassable when water exceeds 0.50 metres, and vehicle speeds are progressively reduced below that threshold to account for the difficulty of driving through accumulated water. Villages are assessed separately, with communities facing water depths above 1.0 metre or flow velocities above 0.5 metres per second considered candidates for external relocation rather than local sheltering.

Static Routes Begin Failing Within Hours

The difference between fixed and changing routes became clear during the first day following the simulated Huikou breach. Between three and six hours after the breach, 15 of the 66 static evacuation routes had already become unusable, representing 22.7% of villages requiring external relocation, whereas dynamic recalculation left only three villages without an accessible route during the same period. By the 21-to-24-hour stage, 34 villages, or 51.5%, had become inaccessible under the static plan compared with 21 villages, or 31.8%, under the dynamic system.

The results also show why the shortest distance does not always produce the fastest or safest journey during a flood. Sanxing Village initially had a 43.5-kilometre route requiring 38.2 minutes, then floodwater forced traffic onto a shorter 38.4-kilometre alternative that required 45.5 minutes because it relied more heavily on slower county roads and included water-affected sections. A later detour stretched the journey to 50.2 kilometres and 65.2 minutes, and every road from the village became unusable nine hours after the breach.

Changing Shelters Can Extend the Time Available to Escape

Tongxing Village demonstrates one of the model's most useful features because the recommended destination changes as different parts of the road network disappear under floodwater. Residents could travel to Qianling during the first 21 hours, switch to Chengling between 21 and 57 hours, then use Liangquan from 57 to 63 hours, extending the possible evacuation period from 21 hours under static planning to 63 hours with changing destinations. The routes became progressively longer, rising from 41.5 kilometres to 48.2 kilometres and then 61.4 kilometres, but they kept an escape option open after the original road had failed.

The analysis also uncovered a serious weakness in evacuation schedules based mainly on when floodwater is expected to reach a village. Ten western villages, including Qiaobei, were originally placed in the third evacuation batch because flood arrival was expected after 24 hours, but their road connections disappeared within nine hours. The researchers argue that evacuation priority should therefore be based on when the last usable route will be cut off, because a village may become physically isolated far earlier than its homes are directly flooded.

The Model Offers Better Decisions but Still Has Practical Gaps

The modelling work was supported by calibration against the major 1998 Yangtze River flood and validation using flood events from 2010 and 2012, with simulated river levels closely following observations at Hukou and Pengze stations. The one-dimensional river model produced high accuracy indicators and small maximum water-level errors, giving researchers confidence in its use for severe breach scenarios comparable to the 1998 event. The two-dimensional inundation component carries more uncertainty because the retention area has not been operated since 1954, leaving no measured inundation records available for direct validation.

Several real-world complications also remain outside the present framework because travel times represent theoretical free-flow conditions rather than the congestion likely to occur during a mass evacuation. The model does not include road-capacity limits, queues, delays in warning households, mobilisation time, pedestrian movement, non-motorised transport or the needs of people without access to vehicles. Future development proposed in the paper includes traffic-assignment modelling, behavioural simulations, multi-modal evacuation systems and better hydrodynamic calibration using satellite observations from future flood events.

Flood evacuation cannot be treated as a single route selected before an emergency begins because both roads and safe destinations can change repeatedly as water spreads. Linking hydrodynamic information with road-network reconstruction gives emergency managers a clearer picture of how much evacuation time remains, which communities should leave first and where people should be redirected when familiar roads disappear. For places facing fast-moving levee breaches, recognising a closing escape window several hours earlier could turn evacuation planning from a fixed map into a continuously updated life-safety tool.