Probabilistic hazard maps for operational use: the case of SO2 air pollution during the Holuhraun eruption (Bárðarbunga, Iceland) in 2014–2015

Abstract

The Holuhraun fissure eruption (Iceland) in 2014–2015, which originated from the Barðarbunga volcanic system, was exceptional in several respects. It lasted 6 months and, throughout its duration, it released up to 9.6 Mt of SO2 in the atmosphere. The main recorded hazard affecting the entire country over the 6 months was the constant presence of a low-level gas cloud that led to recurrent air pollution episodes. The Icelandic Meteorological Office responded to this human health hazard by (1) setting up a forecasting system to anticipate the distribution of SO2 over Iceland and (2) preparing probabilistic hazard maps to support the decisions taken by the Icelandic Civil Protection in demarcating the accessible area around the eruption site. This paper introduces some technical aspects of the application of the CALPUFF numerical model to this eruption like the SO2 dispersal forecasting setup, the volcanic source numerical description, and the Monte Carlo procedure adopted for the creation of the probabilistic hazard maps. CALPUFF-based maps were created in January 2015, when the eruption was still ongoing, with the assumption that the eruption would be continuing with the same intensity. Maps for the entire country and for a smaller domain were produced, the latter showing the likelihood to exceeding an hourly concentration of 2600 μg/m3 (1 ppm) of SO2 for the spring season, a level chosen by the Icelandic Civil Protection for the delineation of the area of restricted access around the eruption site. As during the eruption there was no time for a rigorous evaluation of the model accuracy, we then undertook a retrospective analysis of CALPUFF model performance comparing the forecasted hourly SO2 concentration with real-time measurements at key-sites. The model did reproduce the hourly observations (the maximum within a 24-running window) with a level of agreement of 50.6% in Mývatn (85 km from the eruption site) and 50.4% in Reykjavik (258 km), when instances of null pairs have been removed. In Mývatn, the model overestimated the concentration more than 22% of the time. In Hofn (104 km), the model accuracy is 81.7% and occurrences of underestimation are higher than 11% of the time. In Reyðarfjorður (at 124 km), the model accuracy is assessed to be 82.7% and the model overestimates occurring 15.1% of the time. Possible explanations for the observed mismatch between model results and measurements include the spatial resolution of meteorological data field, the capability into reproducing chemical reactions of SO2 in atmosphere, and the reduced extension of the numerical domain. In addition, the model performance is strongly dependent on the source descriptors (e.g., strength of the SO2 flux and injection height) that, in this contribution, have been kept constant over long periods—neglecting, in this way, the natural variability of a dynamic emission of SO2. This consideration points toward the need of frequent and high-quality observation data for the initialization of dispersal numerical model. In light of this retrospective analysis, the probabilistic hazard maps possibly over-estimated the area exposed to high level of SO2 concentration. All the same, this paper reports on how quantitative probabilistic hazard mapping can be used for mitigating the health risk of volcanic SO2 emissions during a volcanic crisis to the benefit of operational hazard monitoring in support an effective crisis response.