Determination of Sensitivity of Hysplit Back Trajectories for Input Data Commonly Used in Turkey

Abstract

The Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model (web-version) was used to simulate hourly 96-hr back trajectories arriving at a rural site in the Eastern Mediterranean, at altitude of 1500 m from the surface, for each day between 2010 and 2013 years. Two meteorological data archives (NCEP/NCAR Reanalysis and GDAS1) and vertical velocity methods (isentropic and model vertical velocity) were used as model inputs as these are the most widely used input variables. The sensitivity of trajectories to model inputs was measured by the absolute horizontal transport deviation (AHTD), the absolute vertical transport deviation (AVTD) and the relative horizontal transport deviation (RHTD) statistics. Both the meteorological archive and vertical transport method significantly influenced the trajectories. Trajectories simulated by NCEP/NCAR Reanalysis archive were less sensitive to the vertical transport method than trajectories simulated by the GDAS1 archive. Cluster Analysis by SPSS (k-means technique) was applied for each back trajectory data set to classify them into similar groups (clusters). Based on their speed, back trajectories in each data set were classified into five clusters. To examine the dependence of potential source directions of pollutants (i.e. cluster centroids) to the HYSPLIT model inputs, differences of sulfate concentrations were discussed. The results indicated that there were discrepancies in the interpretation of the source-receptor relationship when different inputs were used to run the HYSPLIT model.

Keywords: Back trajectory, Clustering, HYSPLIT, GDAS1, NCEP/NCAR Reanalysis, Air quality modelling