Assessment of error in air quality models using dynamic time warping

Abstract

An estimate of the error between the mean concentration of a released pollutant simulated by an atmospheric dispersion model and the values measured at the ground is obtained using Dynamic Time Warping (DTW). The error measure is relevant to the application with iterative source detection algorithms based on forward numerical transport and dispersion simulations. The new proposed measure is compared with two established error functions commonly used in the literature. A sensitivity study of the error measure to wind direction was performed using real world data from the Prairie Grass field experiment. Whereas both standard measures found smallest error only with a few degrees of wind direction, DTW found the smallest error with a much larger range of wind directions, often as high as 20 degrees.