Finding a free space to park has become one of the main problems for drivers and urban mobility. Traffic congestion, usually caused by drivers looking for a free parking space, can increase energy consumption and air pollution. In this context, providing information regarding the availability of parking spots is crucial to help drivers find a free parking space faster, thus reducing parking search traffic. This paper proposes an approach based on GCNN and an LSTM model to forecast real-time future parking availability by capturing both the temporal occupancy patterns and the geospatial interactions in traffic flow. Different model configurations were evaluated by varying the information fed to the model (e.g., weather forecast, parking violations), the extent of historical data used, and the forecasting period. Experimental evaluation over historical parking data of the city of Tandil in Argentina showed that the proposed approach was able to outperform other state-of-the-art models significantly.