Low cost air quality monitoring systems (LCAQMS) are a promising tool to increase the spatial and temporal resolution of Air quality information. Unfortunately, they suffer from limited accuracy due to a variety of factors including lack of specificity, sensitivity and sensor drifts. Fabrication variance and inherent monitored phenomena characteristics requires costly ad-hoc calibration procedures which have to be repeated on a seasonal basis. Here We propose to use an in-network Federated Learning approach to this problem analysing the results obtained in a publicly available dataset. Preliminary results show that the proposed approach could allow to obtain a single, universal and continuously updated city wide calibration law, ultimately reducing the cost burden of LCAQMS field operation.