Extracting knowledge from hybrid instrumental environmental odour monitoring systems: self organizing maps, data fusion and supervised kohonen networks for prediction

High frequency multi-sensor patterns for instrumental odor monitoring in urban areas close to industrial districts requires adequate analytical strategies, hardware and effective data analysis and control procedures. An approach linking data analysis tools is presented based on unsupervised identification of recurrent patterns from data produced by hybrid instrumental odour monitoring systems. A self-organizing map is built and clustered in order to identify air typologies at a e-nose site; they are interpreted thanks to data fusion of sensorial data from citizen complaints, olfactometric measures, air pollutants and meteorological data. A supervised Kohonen network generates quantitative relationships between multisensor patterns and sparse olfatometric measures on air samples timely collected after complaints or instrumental threshold exceedance, allowing high frequency estimates of odour concentrations.