Air pollution is one of the most relevant environmental problems, seriously affecting public health. Forecasting air pollution levels and assessing air quality pose significant challenges for both problems. Several studies propose different neural networks to predict the pollution, but they often concentrate solely on spatial distance relationships between monitoring stations, neglecting other important spatial and climate contextual factors. Additionally, they frequently focus on localized or specific case studies, failing to consider their broader global impact. To address these limitations, a comprehensive model based on CNN and LSTM has been proposed. This model integrates pertinent features from air monitoring stations, pollution sources, and climate variables, providing a more holistic understanding of the intricate dynamics and relationships within air pollution data. Experiments are conducted on a nationwide air pollution dataset from Spain and India comparing our model with a LSTM and a GNN models of the state-of-the-art, achieving a high error reduction.