Eruption forecasting and hazard assessment at INGV during the 2019 crisis exercise at Campi Flegrei.

Abstract

On 16-19 October 2019, the Italian Civil Protection (DPC) organized a crisis exercise at Campi Flegrei (ExeFlegrei 2019) to verify internal procedures and communication flows towards/from institutions and towards population. At INGV ExeFlegrei allowed to verify monitoring workflows and to test and upgrade operational procedures for real-time eruption forecasting and hazard assessment. As regards the latter aspects, after each issued bulletin (time t0), the INGV team provided in real-time:

1) in terms of eruption forecasting, probabilities (with 80% confidence interval) for unrest, magmatic unrest, and eruption in the month following t0, based on BET_EF model (Marzocchi et al, 2008) calibrated in a long series of past elicitation experiments (Selva et al, 2012a)

2) in terms of scenario forecasting, spatial probability maps for vent opening according to two models (Selva et al, 2012b; Bevilacqua et al, 2015; 2019), conditional on the occurrence of an eruption. Both models follow a Bayesian approach, so both the maps were described by a bestevaluation map (aleatory uncertainty), and percentile maps (epistemic). Also, they contained long-term information from Campi Flegrei morphology and geology, and assimilated evidence from the monitoring data given in the bulletin at t0

3) in terms of hazard assessment, based on the vent opening maps produced in real-time in 2):

- hazard and probability maps for tephra load at the ground, conditional to the occurrence of an explosive or of a specific size (so-called Small, Medium and Large explosive scales) eruption. We produced maps for various thresholds in exceedance probability/load, by combining Fall3D model (Folch et al, 2009) simulations of tephra fallout based on Small, Medium and Large explosive scales, run with the most recent wind forecast at t0. We simulated the tephra accumulated in 24 hours given an eruption onset at t0, t0+24h and t0+48h;

- probability maps for the invasion of pyroclastic density currents (PDC), based on the box model (Neri et al., 2015). The PDC size was based on a lognormal statistics of inundated regions by past PDC, and included the main uncertainty on the deposits extent and on a number of not measured but recognized small-sized PDC in the record. PDC size was also correlated to the caldera sector originating the PDC (Bevilacqua et al., 2017).

In ExeFlegrei, DPC expressed high interest on "pre-eruptive" products, i.e., unrest and eruption probability, and vent position probability maps. This may be motivated by ExeFlegrei terminating before the eruption onset, when the Red Alert level (imminent eruption) was enforced.

Due to the short time between subsequent stages in ExeFlegrei and to the high computational cost of mapping, we could not propagate in real-time the epistemic uncertainty on the vent position and on the eruption size to the hazard and probability maps for tephra load: indeed, only the mean maps were produced, quantifying the aleatory uncertainty only. However, after the exercise end, we retrospectively did it, for the last phase of the exercise, along with an ensemble model of the vent position synthesizing the two Bayesian models.

From a practical point of view, the number of maps from hazard assessment can grow very rapidly as consequence of considering multiple:

- threshold values in probability or intensity measure (e.g., tephra load);

- possible scenarios (e.g., maps conditional to the occurrence of an eruption of any size, or of a specific size, or from the most likely vent position);

- hazards (e.g., tephra fallout and PDC in this case);

- maps considering variability in vent opening;

- percentiles to quantify epistemic uncertainty;

- forecasting time windows (i.e., t0 or t0+24h for tephra fallout).

This can make communication with decision makers difficult and results not fully exploitable in the typical short time of an exercise.

In the light of the experience gained in ExeFlegrei, we conclude that:

- given the relevance for decision makers of eruption and scenario forecast, the models used to provide them should be constantly upgraded as new scientific knowledge is gained, and translated in advance into formalized operational procedures;

- the exploitation of the large portfolio of hazard products, which represents a valuable, quantitative information for taking rational decisions, needs a continuous cooperation between scientists and decision makers. This is a necessarily mutual exchange process: on the one hand, in quiet times the decision makers' needs should become clearer to scientists, and together they could could conceive priority levels (e.g., high, medium and low) for the products that may go into reports to decision makers and into internal ones. On the other hand, scientists should struggle to better communicate the amount and quality of information carried in their products;

- periodic exercises represent a fundamental opportunity to improve the response of scientists and civil protection to a volcanic emergency.