Environmental systems status is described via a (usually big) set of parameters. Therefore, relevant models employ a large feature space, thus making feature selection a necessity towards better modelling results. Many methods have been used in order to reduce the number of features, while safeguarding environmental model performance and resulting to low computational time. In this study, a new feature selection methodology is presented, making use of the Self Organizing Maps (SOM) method. SOM visualization values are used as a similarity measure between the parameter that is to be forecasted, and parameters of the feature space. The method leads to the smallest set of parameters that surpass a similarity threshold. Results obtained, for the case of Thessaloniki air quality forecasting, are comparable to what feature selection methods offer.