Predictability of nearshore sandbar behavior

Nearshore sandbars are ridges of sand that are commonly observed along sandy coasts in water depths less than 10 m. Sandbars are the last natural line of defense against the attack of storm waves on the coast and, accordingly, human measures to combat coastal erosion often involve changes in sandbar height or position by means of sand nourishments. Over time, the cross-shore location of a sandbar changes in response to the variable wave forcing. The prediction sandbar migration is required to plan and evaluate the effectiveness of nourishments in mitigating coastal erosion. Cross-shore sandbar behavior is usually studied with models based on short- and small-scale (meters, seconds to hours) hydrodynamic and sediment-transport processes. These process-based models struggle to reproduce natural sandbar behavior on time-scales of a few days to weeks and have uncertain skill on the long term (months to years). A potential source of the difficulty to predict long-term sandbar behavior with process-based models is the nonlinear nature of the process equations, combined with the iterative update scheme. This introduces the possibility of exponential error accumulation, causing increasing divergence between modeled and observed states over time. Instead of studying the predictability of cross-shore sandbar behavior from the perspective of the underlying processes, it is also possible to infer sandbar dynamics from observations with data-driven models. Data-driven models extract the structure underlying sandbar behavior directly from the observations, and can therefore be used to study fundamental aspects of the predictability of sandbar behavior. This thesis investigates, develops and applies several methods to study the predictability of cross-shore sandbar behavior from an observation-based perspective. The analyses are based on 12000 daily-observed sandbar-crest locations and wave height observations collected for five sandbars at three different field sites. The importance of nonlinearity in sandbar behavior on timescales of days to years is investigated with nonlinearity-detection methods and neural network models. These analyses show that cross-shore sandbar behavior on timescales of days to weeks can be modeled as a simple relation between sandbar migration, sandbar location and wave height. On longer timescales of months to years, however, the effect of nonlinearity on predictability varies between different field sites. A further analysis of sandbar behavior with an equilibrium-based model shows that the short- and small-scale nonlinear processes underlying sandbar migration drive a sandbar toward a wave-height-dependent equilibrium state. However, sandbars are mostly out-of-equilibrium with the wave conditions, because the wave conditions change faster than the time it takes for a sandbar to reach the equilibrium location. Long-term offshore-directed trends in sandbar location develop when sandbars never reach the equilibrium locations associated with the highest waves. The trends in this type of sandbar behavior are difficult to model from the underlying physical processes because the nonlinear nature of these processes causes exponential error accumulation over time. Sandbars that do reach their equilibrium location during a storm, however, do not exhibit any trends longer than the duration between individual storms, and can be predicted with detailed nonlinear models from known wave forcings over their entire lifetime