244
A. Uppstu et al.
observations of both volcanic ash and SO 2 . We demonstrate through a simulated
eruption of the Etna volcano how the EnKF can be utilized to deliver assimilated
forecasts of the dispersion of volcanic ash and how the forecast could be visualized.
38.2 Methods
In order to provide assimilated forecasts, an atmospheric transport model, a volcano eruption model, a data-assimilation method and a set of observations that are
obtainable in near real-time are needed. The ensemble is formed through perturbing
the time stamp of the meteorological forecast, with the standard deviation of the
perturbations being 3.0 h, and through perturbing the emission source.
38.2.1 A Volcanic Eruption Model
The volcanic eruption model is based on an empirical relationship found between
the height of the eruption plume and the eruption rate [2]. However, methods for
determining the vertical distribution of ash within the plume as well as for determining
the particle size distribution of the eruption require more study. In this study, it is
assumed that 75% of the erupting mass is distributed uniformly over the top 25% of
the plume, with the remaining mass being uniformly distributed over the bottom part
of the plume. Within the ensemble, the eruption rate is assumed to be log-normally
distributed. Compared to a normal distribution, this allows for a significantly larger
range of different eruptions to be included within the ensemble, which is essential
for a proper description of the a priori probabilities.
In order to estimate the strength of the eruption, the logarithm of the eruption
rate is included as a part of the control vector of the data assimilation process. To
avoid a collapse of the ensemble and to take into account temporal changes in the a
priori probability distribution, the distribution of the eruption rates is set to constantly
evolve towards a pre-defined distribution within a correlation time of 24 h. Thus, if
no observations of the erupting plume are made, the eruption rate distribution reverts
to the original one within a time scale of one day.
38.2.2 A Simulated Eruption of Etna
In order to test the methodology, we applied it on a simulated eruption of the Etna
volcano. For the simulation, we utilized the emission term estimated for the 2010
Eyjafjallajökull eruption [4]. Simulated retrievals of volcanic ash loadings from the
satellite-borne IASI instrument as well as of ash concentrations from the satelliteborne CALIOP lidar and the EARLINET ground-based lidar network were created
A. Uppstu et al.
observations of both volcanic ash and SO 2 . We demonstrate through a simulated
eruption of the Etna volcano how the EnKF can be utilized to deliver assimilated
forecasts of the dispersion of volcanic ash and how the forecast could be visualized.
38.2 Methods
In order to provide assimilated forecasts, an atmospheric transport model, a volcano eruption model, a data-assimilation method and a set of observations that are
obtainable in near real-time are needed. The ensemble is formed through perturbing
the time stamp of the meteorological forecast, with the standard deviation of the
perturbations being 3.0 h, and through perturbing the emission source.
38.2.1 A Volcanic Eruption Model
The volcanic eruption model is based on an empirical relationship found between
the height of the eruption plume and the eruption rate [2]. However, methods for
determining the vertical distribution of ash within the plume as well as for determining
the particle size distribution of the eruption require more study. In this study, it is
assumed that 75% of the erupting mass is distributed uniformly over the top 25% of
the plume, with the remaining mass being uniformly distributed over the bottom part
of the plume. Within the ensemble, the eruption rate is assumed to be log-normally
distributed. Compared to a normal distribution, this allows for a significantly larger
range of different eruptions to be included within the ensemble, which is essential
for a proper description of the a priori probabilities.
In order to estimate the strength of the eruption, the logarithm of the eruption
rate is included as a part of the control vector of the data assimilation process. To
avoid a collapse of the ensemble and to take into account temporal changes in the a
priori probability distribution, the distribution of the eruption rates is set to constantly
evolve towards a pre-defined distribution within a correlation time of 24 h. Thus, if
no observations of the erupting plume are made, the eruption rate distribution reverts
to the original one within a time scale of one day.
38.2.2 A Simulated Eruption of Etna
In order to test the methodology, we applied it on a simulated eruption of the Etna
volcano. For the simulation, we utilized the emission term estimated for the 2010
Eyjafjallajökull eruption [4]. Simulated retrievals of volcanic ash loadings from the
satellite-borne IASI instrument as well as of ash concentrations from the satelliteborne CALIOP lidar and the EARLINET ground-based lidar network were created