Future prediction of coastal recession using convolutional neural network

Coastal recession resulting from sea level rise and wave action is a significant environmental concern, posing challenges for accurate long-term predictions due to the intricate interplay of multiple factors. The study explores the potential of machine learning methods, specifically Convolutional Neural Network (CNN), for predicting long-term coastal changes caused by sea level rise and wave conditions using beach profile data. The model was trained and validated using historical data, exhibiting a strong similarity between its predictions and the observed beach profile data with mean absolute and root mean square errors ranging between 0.2 and 1 and an R2 value above 0.93. Furthermore, we compared the long-term coastal recession predictions generated by the CNN model, which were extended until the year 2050, with those derived from the widely used Bruun rule, a conventional benchmark in coastal recession prediction. The CNN model's predictions stood out, demonstrating a greater plausibility as compared to the Bruun rule. Modeled profiles closely align with the observed recession of past data, showcasing the CNN model's adaptability, especially for steeper profiles, as seen at NarrabeenCollaroy Beach and Teluk Cempedak Beach. The CNN model's prediction for Narrabeen-Collaroy Beach indicated a coastal recession of 0.33 m/year for the projected 36 -year period (2015-2050), closely mirroring the observed coastal recession of 0.32 m/year over the preceding 36 years (1979-2014). In contrast, the Bruun rule's predicted that recession is notably lower at 0.09 m/year. Similarly, at Teluk Cempedak, the CNN model's forecast of 0.16 m/year falls below the reported 0.8 m/year but surpasses the Bruun-estimated 0.06 m/year. These results emphasize the potential of machine learning methods to significantly enhance our ability to predict future coastal recession, particularly when contrasted with traditional approaches like the Bruun rule.