The transport sector is one of the most greenhouse gas emitting sectors, and users waste a lot of time in transport due to congestion, with relevant socio-economic consequences. Understanding the spatio-temporal traffic patterns could help traffic forecasts and allow to adjust the transportation infrastructure more efficiently in order to cope with these challenges. Deep learning has had a huge success in recent years due to its ability to capture patterns to make predictions that were not able to be properly assimilated by older regression models. More recently, Graph Neural Networks (GNN) have emerged as the state of the art in capturing spatio-temporal traffic patterns thanks to their capability to model the structure of the traffic network and to capture its dynamics. The present study analyzes the impact of weather scenarios in the case of bike-sharing orgin-destination matrix prediction. In this study, a type of Graph Neural Network (ST-ED-RMGC) has been employed. We consider real world bike-sharing data related to urban areas. The results of the study show a strong impact of the weather on the quality of the prediction, whether it is an atypical temperature, a strong wind or the presence of rain. This suggests the need for introducing as input features contextual information about weather conditions as well as data from other sources, such as historical demand information on transport modes alternative to the one being predicted, in order to quickly identify the onset of atypical events