AI speeds up climate risk calculations for port infrastructure

Ports across the Mediterranean are facing a mounting engineering dilemma as rising sea levels and intensifying storm patterns threaten the structural reliability of existing coastal defenses. A new study proposes that artificial intelligence (AI) may offer a faster and more flexible way forward.

Published in the Journal of Marine Science and Engineering (JMSE), the study titled An Artificial Intelligence Approach for Coastal Structures Adaptation to Climate Change: Insights from a Case Study in the Mediterranean Sea presents a hybrid modeling framework that integrates probabilistic reliability analysis with artificial neural networks and genetic algorithms to evaluate how coastal breakwaters should be adapted under near-future climate scenarios.

AI meets coastal engineering in climate risk assessment

Coastal structures such as breakwaters are critical to port operations, shielding harbors from wave energy, protecting cargo handling infrastructure, and ensuring navigational safety. However, climate change is altering wave climates in ways that were not fully anticipated during the original design phases of many structures. Rising sea levels, changing storm tracks, and more energetic wave conditions increase the probability of structural failure modes such as sliding, overtopping, and armor damage.

The research focuses on the Port of Valencia in the western Mediterranean, a major logistics hub exposed to evolving wave conditions. The authors selected two representative sections of its north breakwater for analysis: a vertical breakwater section located at the head and a compound breakwater forming part of the trunk. These two structural types respond differently to wave loading and therefore require separate reliability assessments.

Traditionally, engineers rely on deterministic numerical models to evaluate structural performance under projected wave conditions. While highly detailed, such models require significant computational resources, particularly when assessing numerous combinations of wave height, period, direction, sea level rise, and structural parameters. When adaptation strategies such as crest elevation increases or armor reinforcement must be tested across multiple climate scenarios, the time and cost of simulation can become prohibitive.

To address this limitation, the study introduces a hybrid approach. It first uses Monte Carlo simulations to generate large probabilistic datasets of wave and structural parameter combinations. These simulations capture uncertainty in environmental forcing and material properties, enabling reliability analysis of different failure modes. The resulting dataset then trains artificial neural networks to learn the relationship between environmental inputs and structural performance outcomes.

However, instead of relying solely on conventional neural network training, the researchers enhance the process with genetic algorithms. Genetic algorithms iteratively optimize the neural network's weights, improving predictive performance and reducing computational inefficiencies. This combination significantly accelerates scenario testing compared with traditional modeling workflows.

The authors report that the hybrid ANN–genetic algorithm model achieves computational efficiency improvements on the order of 30 to 40 times compared with neural networks trained without optimization. This acceleration allows engineers to explore many adaptation configurations in a fraction of the time, making it feasible to conduct sensitivity analyses and identify optimal adaptation pathways.

Valencia case study signals urgent adaptation needs

Using projected climate conditions for 2035 and 2050, the study evaluates the structural reliability of the selected breakwater sections under intensified wave climates and rising sea levels. The analysis considers multiple failure mechanisms, including sliding instability and wave overtopping for the vertical section, and sliding, overtopping, and armor damage for the compound section.

Results indicate that under the 2050 scenario, adaptation will likely be required to maintain acceptable safety levels. For the vertical breakwater, the framework suggests that crest elevation may need to be increased by up to 1.2 meters to counteract higher water levels and more energetic waves. For the compound section, a smaller crest raise of around 0.4 meters appears sufficient, but reinforcement of the armor layer is recommended to withstand increased wave attack.

These adaptation magnitudes are not trivial. Raising a breakwater crest by more than a meter involves significant construction costs, logistical coordination, and potential disruption to port operations. Yet failing to act could expose infrastructure to overtopping events that compromise cargo areas, equipment, and safety systems.

The study notes that reliability thresholds must account for probabilistic failure rather than relying on single deterministic scenarios. By incorporating variability in wave conditions and structural parameters, the framework provides a more nuanced estimate of risk. It identifies combinations of wave height and sea level that disproportionately contribute to failure probability, helping engineers prioritize interventions.

An important insight from the analysis concerns extreme sea states. The authors note that a substantial share of predictive error in the AI model is concentrated in the upper range of wave conditions. Roughly 60 percent of model error is associated with the highest percentiles of wave energy. This finding is critical because extreme events are precisely those that drive structural design criteria. While the hybrid model performs strongly overall, improving accuracy under extreme conditions remains a priority for future refinement.

The research also highlights the absence of large-scale experimental validation for some breakwater configurations under future climate extremes. While the AI framework can accelerate scenario testing, it does not eliminate the need for detailed engineering verification before implementation. The authors position the tool as a decision-support system that can narrow down adaptation options efficiently before full-scale design studies.

Faster decision-making for climate-exposed ports

Ports operate within complex economic and logistical systems where downtime translates into financial loss. As climate change accelerates, the window for proactive adaptation narrows. Decision-makers need tools that allow rapid comparison of alternative strategies, including crest height increases, slope adjustments, armor reinforcement, and combinations thereof.

By merging probabilistic reliability analysis with machine learning, the framework reduces the time required to assess adaptation scenarios. Instead of running thousands of computationally expensive physical simulations for each design modification, engineers can use trained AI models to estimate structural performance quickly. This capability enables exploration of multiple climate projections, cost trade-offs, and safety targets in parallel.

The computational gains are particularly relevant for public authorities operating under budget constraints. Climate adaptation projects must compete for funding with other infrastructure priorities. A faster assessment process allows planners to identify cost-effective measures and justify investments with quantitative risk estimates.

Energy and resource efficiency also emerge as indirect benefits. By optimizing adaptation magnitude rather than defaulting to conservative overdesign, the framework can prevent unnecessary material use. This aligns with broader sustainability objectives in coastal engineering, where carbon-intensive construction must be balanced against long-term resilience.

The authors also caution that AI models require careful calibration and periodic updating as new climate data becomes available. Wave climate projections evolve as global models improve. If environmental baselines shift, trained neural networks must be retrained to ensure continued accuracy. The system's performance is therefore linked to data quality and ongoing monitoring.

Extreme conditions pose the greatest challenge for predictive modeling. Improving representation of rare but high-impact events may require expanded datasets, integration of field measurements, or hybrid modeling that combines AI outputs with targeted high-resolution simulations. The authors suggest that enhancing performance in the upper tail of wave distributions is essential for reliable climate adaptation planning.