Evaluation of interpolation methods for the prediction of urban methane concentrations

Abstract

Having data about atmospheric concentrations in an entire urban area is difficult, hence interpolation methods are helpful. Their choice will depend on minimising the error. In this work, twodeterministic (Inverse Distance Weight and Local Polynomial Interpolation) and two stochasticmethods (Simple and Ordinary Kriging) were applied to predict seasonal and annual atmosphericmethane (CH4) concentration means. Two sampling networks were designed in an intermediatecity, covering a wide variety of urban densities, with different sampling site numbers. The mainobjective was to find the interpolation model that best predicts CH4 concentration and to analyseif the network’s expansion improves the metric errors - the mean error (ME) and the root-meansquare error (RMSE). The ME values were close to zero in all cases, and the stochastic methodshad the smallest RMSE for both networks. Besides, adding more sampling sites improved up to50% of the RMSE values. Finally, an integrated map was obtained incorporating all the bestinterpolation models, which gave a difference of <4% between the measured and the estimatedCH4 concentration. This type of study is helpful to evaluate the design of a sampling network, theterritorial planning and future installations of CH4 sources.