In the context of energy mix, and in order to improve the stability of the power grid, it is essential for a grid operator to know, as precisely as possible, the future quantities of energy produced, particularly from wind and solar sources. To improve wind energy resource forecasts, our study is based on a hybrid neural network, integrating meteorological reanalysis data (ERA5) as well as topographical data geographically close to the prediction site. We are particularly interested in the impact of the size of the considered topographical area on the network's forecasting performance. In this study, we specifically show that the integration of relief data improves predictions compared to the same non-hybrid network, regardless of the forecast horizon considered, with a gain in RMSE of up to 17%.