Quantitative Verification and Calibration of Volcanic Ash Ensemble Forecasts Using Satellite Data

Abstract

In this paper, we address the problem of verifying and calibrating ensemble‐based probabilistic volcanic ash forecasts. The ensemble members are constructed from dispersion model simulations with different meteorological fields obtained from an ensemble meteorological forecast model and different values of ash source parameters such as ash column height and vertical mass distribution. The Brier score is employed to verify the probabilistic forecasts relative to binary‐valued ash detection fields and fully quantitative satellite‐retrieved ash mass load fields. A new ensemble forecast calibration methodology, which treats the meteorological and source term variations on equal footing, is also developed. This enables the creation of an ensemble subset yielding higher Brier skill scores than the uncalibrated ensemble during a calibration time window of about 6 hr after the start of an eruption. It is shown that this can be used to improve probabilistic forecasts up to 24 hr after the start of the eruption using the Indonesian 13 February 2014 Kelut and 6 November 2015 Rinjani eruptions as case studies.

Using Satellite Data to Determine Empirical Relationships between Volcanic Ash Source Parameters

“…Apart from these source-term uncertainties, all forecasts of ash are plagued by uncertainties in the ash transport and removal mechanisms within dispersion models. These include errors in the NWP fields [4][5][6], and parameterization of various physical processes such as sedimentation, deposition, and aggregation.…”

“…The empirical approach, in addition to other problems outlined above, requires specification of the poorly understood fine ash mass fraction, but it has the advantage of being usable regardless of the atmospheric conditions. The new approach uses satellite retrievals obtained from the NOAA VOLcanic Cloud Analysis Toolkit (VOLCAT) software, based on the algorithm of Pavolonis et al [14], with inverse modelling methods based on multidimensional sampling of model source parameter space [3,6,[24][25][26], to estimate the source term (total mass emission rate, spatial mass distribution, and particle size distribution) for 14 eruption case studies in the Darwin VAAC area. These source term quantities are expressed as relatively simple functions of a few parameters such as the fine ash fraction, umbrella cloud diameter, and mean particle radius.…”

Ensemble-Based Data Assimilation of Volcanic Ash Clouds from Satellite Observations: Application to the 24 December 2018 Mt. Etna Explosive Eruption

“…Thus, as previously shown by Dare et al [87], Zidikheri et al [88], ensemble prediction is encouraged with respect to deterministic forecasting.…”

Atmospheric Dispersion Modelling at the London VAAC: A Review of Developments since the 2010 Eyjafjallajökull Volcano Ash Cloud

“…The future state of the atmosphere therefore cannot be completely described with a single deterministic model forecast; instead, an ensemble of model runs is needed to fully predict all the possible outcomes [100]. A good volcanic ash forecast should use an ensemble of met data to communicate a probabilistic assessment of the expected location and concentration of ash in the atmosphere [8,101,102]. The Met Office's MOGREPS-G system produces ensemble forecasts for the whole globe up to a week ahead.…”

“…We can improve distal forecasts and also avoid explicitly modelling near source processes by using observations to modify the source conditions using inversion schemes (e.g., [9,88,89,91,102]), or by creating 'virtual' sources far from the vent using data insertion and Data Assimilation (DA) techniques (e.g., [11,[134][135][136][137][138][139]). DA techniques have the advantage that they go some way to addressing the inaccuracies in the dispersion model forecasts due to the uncertainties associated with source terms, meteorological data and model parametrizations accumulating over the duration of the run [140].…”