Near-term projection of anthropogenic emission trends using neural networks

This study is the first ever analysis of the global time series 1970–2008 of the Emissions Database for Global Atmospheric Research (EDGAR) for 10 chemical species and more than 3000 subsectors with neural networks, which tries to find non-linear behaviours that several species have in common. plication of the different neural network types, suggests that General Regression Neural Networks (GRNNs) are the most suitable to train a typical Gaussian trend with a very low error level. As such, GRNNs are very suitable for filling the data points missing from the EDGAR time-series, but they are not so good at a making projections outside the time period of the database. Instead Multi Layers Perceptron (MLP) is very suitable for projecting a subsequent year to the database time period of several decades, even though MLP is characterised by a slightly higher absolute mean error than the GRNN. ns of the Principal Component Analysis (PCA), we identified which chemical substances are driven similarly by the activity data over the almost 40 years time period. In all the geographic aggregations, we observed that the emission trends of CO2, SO2 and NOx can be grouped into one cluster, and the emission trends of CH4 and the particulates into another. The best time interval for the prediction proved to be eleven years, and projections seemed to be reliable for three consecutive years following the last year of the database time-series.