This contribution presents an assessment at a regional (island) scale of the beach erosion due to storm events under Climate Change. The approach adopted to assess beach erosion at the island scale consisted of three modules. First, the location, dimensions and other attributes of the Cypriot beaches were recorded on the basis of widely-available satellite imagery. Secondly, sea levels and waves were modeled along the coast under different climatic scenarios and dates in the 21st century. Finally, using these projections beach retreat due to the relative mean sea level rise (RSLR) and extreme sea levels (ESLs) was estimated using ensembles of analytical and numerical cross-shore morphodynamic models, respectively. Extreme sea levels (ESLs) were projected to (a) increase by up to 60% in 2100 from their baseline (2000) levels, and (b) vary along the coast, with the highest ESLs (and corresponding waves) projected for the southern and western coasts. The mostly narrow Cypriot beaches (91% recorded maximum widths of < 50 m) showed increased exposure to erosion. In 2100, about 47% and 72% (based on the median model estimates) of the 241 unprotected Cypriot beaches will be permanently eroded, due to mean sea level rise (SLR), to 50% of their present maximum width, depending on the scenario. In addition to the long-term erosion due to SLR, severe storm erosion is projected by 2050 even under the RCP4.5 scenario; the 100-year extreme sea level event (ESL100) may overwhelm (at least temporarily) 49% of the currently unprotected Cypriot beaches without effective adaptation responses, with the most exposed beaches located along the northern coast. As the beach carrying capacity and hedonic value will be severely compromised, effective adaptation policies and technical measures will be urgently required.

In this paper, a deep learning framework combined with isogeometric analysis (IGA for short) called IGA-Reuse-Net is proposed for efficient reuse of numerical simulation on a set of topology-consistent models. Compared with previous data-driven numerical simulation methods only for simple computational domains, our method can predict high-accuracy PDE solutions over topology-consistent geometries with complex boundaries. UNet3+ architecture with interlaced sparse self-attention (ISSA) module is used to enhance the performance of the network. In addition, we propose a new loss function that combines a coefficients loss and a numerical solution loss. Several training datasets with topology-consistent models are constructed for the proposed framework. To verify the effectiveness of our approach, two different types of Poisson equations with different source functions are solved on three datasets with different topologies. Our framework can achieve a good trade-off between accuracy and efficiency. It outperforms the physics-informed neural network (PINN for short) model and yields promising results of prediction.