The Dutch coast is affected by coastal erosion. In order to limit erosion and prevent inland migration of dunes, sand nourishments have been adopted as a common practice for the maintenance of the coast. The evaluation of the applied coastal erosion policy is achieved by the use of coastal indicators. Increased number of nourishments affects the coastal profiles and coastal indicators are used to describe those changes along the coastal zone. The present work intends to improve the definition of one of the indicators (dune foot position) by proposing a new methodology and to develop a Bayesian network in order to assess the effectiveness of nourishments on the coastal system. Coastal indicators, such as the Momentary Coastline and the dune foot position, are used in the Bayesian network in order to quantify those effects.

Available field measurements of the entire Dutch coastline (JARKUS morphometric database) are used in order to develop the methodology of the dune foot position detection. Taking into account the geometry of the profiles, the dune foot position can be detected by calculating the first and second derivatives of the measured points along the profiles. Comparison with visual, in-situ observations, constitute the validation method of the proposed methodology. Root mean square errors, with respect to visual observations, are used to compare the performance of two methods for the dune foot position detection; the proposed methodology and the current dune foot definition, which detects the dune foot at the most seaward crossing of the profile with the level of +3 m NAP. By looking at the performance of the methodologies at different areas, the results are diverse. However, the proposed methodology shows a general improvement of the detection. Moreover, the methodology is believed to be generic and applicable to other dune systems around the world, since it is purely based on the morphology of the profile.

A Bayesian modelling approach is used to assess the effectiveness of nourishments on coastal indicators for the Holland coast. The selection of the input and output parameters, the design of the network and the optimization of the structure are the main steps for the development of the model. Once the optimum configuration is chosen and the model is trained, the Bayesian network gives the possibility to investigate the relations between variables, by constraining the input nodes on a specific range of values and assessing changes on the indicators. By applying those constraints it can be found that nourishments have positive effects on the coastal indicators, since larger seaward displacement is observed compared to cases in which no nourishment has been implemented. Moreover, initial erosive trends of the dune foot position are constrained by the implementation of nourishments. In addition, the effects of nourishments at different time scales are investigated. To this end, time horizons of 1, 5 and 10 years after a nourishment implementation are chosen. Positive effects of nourishments are present at all the considered time horizons, with beach nourishment to have an immediate effect on the indicators, whereas shoreface nourishments reach a stronger effect 10 years after the implementation.
Finally, the Bayesian inference can be used as a decision support tool for coastal managers, since its user-friendly environment can lead to easy interpretation and use for coastal management purposes. The required sand volume can be estimated in order to achieve a specific magnitude of seaward displacement of the indicators or in order to preserve the coastline at the current state. The estimations concern large spatial scales.