Since their development in the 1970s, mesoscale atmospheric models have been used for a variety of meteorological applications. Over the years, the models have evolved and nowadays, state-of-the-art atmospheric models are capable of operating on spatial resolutions in the order of kilometres. In the hydraulic engineering community, atmospheric models are used for operational flood protection. HIRLAM is used in the Netherlands with typical spatial resolutions of 11 km. These scales are too coarse to correctly predict small scale effects such as squall-lines and convection cells. Furthermore, the resolution is too low to accurately capture the land-water boundary. The state-of-the-art models of today are capable of predicting small scale meteorological events that might be of interest for hydraulic engineers. The goal of this study was to successfully use a state-of-the-art atmospheric model at high-spatial resolutions to investigate the possible added-value of these models for hydraulic engineering purposes. The problem that was addressed concerns wave growth at short fetch. This is specifically important for wave prediction on rivers, harbour basins, and small lakes. Deltares [2013] investigated the predictive capabilities of SWAN for short fetches (< 5000 m) at Lake IJssel. SWAN computations were performed for a selection of 20 cases and they found uncertainties in the model predictions for wave heights (Hm0; rbias = -15%, SI=-11%) and wave periods (Tm-10; rbias= -15%, SI=-11%). It is believed that this is partially due to inaccurate representation of the wind variability near the land-water transition. A model of the atmosphere was set up using the non-hydrostatic state-of-the-art mesoscale model WRF. The model covers a total area of 3240 x 4050 km. Initial and boundary conditions were derived from ERA-Interim and using a series of five nests, a horizontal resolution of 500 m was realized for the area around the northwest of Lake IJssel. Two storms were hindcasted; storm 1 from January 3rd 2012 to January 7th 2012 and storm 2 from December 4th 2013 to December 8th 2013. The model was validated using wind and temperature observations from the KNMI and Rijkswaterstaat at Lake IJssel. Overall, the model is in good agreement with the observations. Statistical analyses of data, showed that the uncalibrated model performed well in terms of wind speed (around Lake IJssel; storm 1: rbias <5%, SI ? 10%; storm 2: rbias 10%–20%, SI ?20%) and wind direction (around Lake IJssel; storm 1: bias ? 3°, RMSE ?10°; storm 2: bias 4°–15°, RMSE ? 20°). Large errors were found for the surface temperature of Lake IJssel. SST updates from ERA-interim do not represent the temperature of Lake IJssel, and were structurally over-predicted (storm 1: 2°C or 70%; storm 2: 2.5–3°C or 70%–75%). A non-stationary SWAN model was set up using calibrated setting (WTI2011) to perform wave simulations at Lake IJssel. To be able to use the friction velocities (instead of the diagnostic 10-meter wind speed) in SWAN, an extra step was required. The friction velocities from WRF were transformed to `pseudo winds' by using the drag relation Wu [1980] from SWAN inversely. The model was validated for the two storms using observations of wave height H_m0, wave periods (Tp, Tm01, Tm02, Tm-10) and 1-D wave spectra. For wind directions along the shore normal (239°N ± 20°), negative biases were found for the wave heights (Hm0; storm 1: 0.13%, -0.21% for location FL48 and FL49 respectively; storm 2: -0.18%, and -0.11% for location FL48 and FL49 respectively) and wave periods (Tm-10; storm 1: 10%, 7% for location FL48 and FL49 respectively; storm 2: –7%, –4% for location FL48 and FL49 respectively). A part of the error was caused by the already existing errors in the wind data from WRF. 2-D wave spectra from SWAN predicted disturbances for the stations FL48 and FL49 during slanting fetch conditions. Alongshore propagating low-frequency wave components were predicted and the bended coastline enhanced the disturbances. No directional observations were available, but the qualitative agreement between the 1-D spectra suggests that the SWAN predictions are realistic and that the measuring locations FL48 and FL49 are disturbed by energy that is not in aligned with the wind direction. To investigate the effects of high spatial resolutions in WRF, both storms were simulated using five different horizontal resolutions of 2700, 1500, 900, 500, and 300 m. Analysis showed that only little differences between the simulations occur during calm periods. The simulations did react differently during periods of rapid variations in the wind field. Investigation of the origin of these rapid variations showed that these mainly occurred during periods of precipitation (often combined with passage of weather fronts). A particular event with very high velocities (>30 m/s) was investigated. The specific event turned out to be a gust front coming from a convection cell. The simulations responded differently to the event; with increasing resolutions, higher velocities, precipitation rates, downward velocities, and other locations were found. Finally, for the highest resolution (?x = 300 m) artificial disturbances in the atmospheric pressure were found, indicating limitation to the validity of the chosen model settings. To assess the effects of the spatial resolution of the wind field in SWAN, both storms were simulated using the 2700, 1500, 900 and 500 m wind fields. The same method, using pseudo wind speeds was applied for the use of the friction velocity in SWAN. However, for the specific purpose of dealing with wave growth at short fetches, problems were found near the land water boundary. Because the resolution of the wind fields were relatively coarse compared the length scales associated with short fetches, wind field data did not accurately model the land-water boundary. Therefore, an extra preprocessing step was used to exclude land points from the WRF wind field, and to extrapolate water-based data to the land-water boundary. Time series of the results showed equal performance for all simulations, no differences due to the wind field resolution were found. Further investigation of the spatial variability of the wave field showed that the effects of the wind streaks were also shown in the SWAN results. This resulted in differences between the simulations up to 6\% near the coast. In conclusion, the results showed that the model was still able to produce results that were in agreement with observations, and simulations with other resolutions, without showing signs (in the variables: pressure, temperature, humidity, and wind) of instabilities. This suggests—in agreement with Hong and Dudhia [2012]— that the model indeed still produces reliable wind results for resolutions up to 500 m. A specific event was investigated that turned out to be a convection cell that clearly showed the different reaction to increasing resolutions, higher velocities (horizontally and vertically), higher precipitation rates, and different locations for the convection cell were found. The coarser resolutions (?x =2700 m and ?x = 1500 m) hardly showed any deviations in wind speed due the cell, while the higher resolutions showed strong surface winds of up to 32 m/s. Simulations with SWAN showed only little variation based on the resolution of the wind field. The added value of these models in relation to wave modelling can be found from the better representation of the surface features such as the land-water boundary, and land-use for the determination of the roughness, which allow for the prediction of wind above harbour basins and river, possibly leading to more accurate predictions of wave characteristics on these locations. Furthermore, the prediction of convective processes such as squall-lines and convection cells could be a valuable addition for operational wave (en wind) predictions.