Spatial Prediction of Air Pollution Parameter NOx based on GMDH-type Neural Network

Air pollution-induced issues involve public health, environmental, agricultural and socio-economic aspects. Therefore, decision-makers need low-cost, efficient tools with high spatiotemporal representation for monitoring air pollutants around urban areas sensitive regions. pollution forecasting models different time steps forecast lengths are used as an alternative support to traditional quality stations (AQMS). In recent decades, given their eligibility reconcile the relationship between parameters of complex systems, artificial neural networks have acquired utmost importance in field forecasting. this study, machine learning regression methods establish a mathematical meteorological factors from four AQMS (A-D) located Çerkezköy Süleymanpaşa, Tekirdağ. The model input variables included parameters. All developed were intent provide instantaneous prediction pollutant parameter NOx within across stations. GMDH (group method data handling)-type network (namely self-organizing deep approach), five hidden layer structure consisting maximum neurons was preferred and, choice layers made way minimize error. all developed, divided into training (%80) testing set (%20). Based on R2, RMSE, MAE values models, provided superior results regarding (reaching 0.94, 10.95, 6.65, respectively station A) AQMS. yielded B by using A (without input) RMSE 0.80, 10.88, 7.31 respectively. is found suitable being employed fill gaps records across-AQMS.