Comparison of SCIPUFF predictions to SO2 measurements from instruments on the MetOp-A, MetOp-B, Aura and Suomi satellites from the 2016 fire at Al-Mishraq

Abstract

Abstract On October 20th, 2016, Daesh (Islamic State) set fire to the sulfur production and storage site at Al-Mishraq, Iraq. The fire burned continuously for seven days resulting in the release of a buoyant dynamic sulfur dioxide (SO2) toxic plume extending geographically for over 300\xa0km. This Al-Mishraq event provides an opportunity for scientists to address probabilistic atmospheric transport and dispersion (PATD) hazard model performance across multiple downwind scales over a large area. In using any PATD model, a fair evaluation of the model depends on the quality of the source term – especially for a very dynamic source term such as a burning sulfur stockpile. An innovative approach to derive such a source for Al-Mishraq was explored by the Swedish Defence Research Agency (in Swedish Totalforsvarets forskningsinstitut, or FOI). The experts from FOI characterized the SO2 source term based on satellite observations, which reflect vertically-integrated concentrations and are expressed in Dobson units (DU). The work throughout this paper is a novel approach and leverages the FOI analysis to evaluate a PATD, the Second Order Closure Integrated Puff (SCIPUFF) transport and dispersion code as adapted within the Hazard Prediction and Assessment Capability (HPAC). SCIPUFF predictions are compared to the satellite-based vertically-integrated SO2 measurements from the Moderate Resolution Imaging Spectro-radiometer (MODIS) instruments. Satellite measurements were available at 07Z and 11Z daily from 21 to 26 Oct 2016. A 1-DU background was assumed. SCIPUFF over predicts (factor of 2) column density (DU) peaks in the near field (40, 80, and 120\xa0km) and under predicts in the far field (200, 280 and 320\xa0km); best agreement is at the 160\xa0km arc. Plume direction errors of 25–80° are found when comparing the azimuths of peak observed and predicted column densities. Performance metrics traditionally used for model evaluation included fractional bias (FB), normalized mean-square error (NMSE), and figure of merit in space (FMS). The FB at 07Z was −0.2 to +0.09 across the eight arcs, 40–320\xa0km downwind. Upon analyses, SCIPUFF showed approximate equal over and under prediction for the 07Z; and the FB at 11Z under predicted seven of eight arcs, with values that ranged from −0.07 to +0.36. The NMSE was 3.11–5.06 with no identifiable trends on arc distance. The FMS, averaged over each arc, by day, was between 0.31 and 0.56. Higher values were found in the early and later stages of the fire 21–22 and 25–26 October. Lower averaged FMS values were calculated for 23–24 October which corresponds to a period where the fire burned hotter and SO2 mass was injected to higher altitudes of 1–4\xa0km.