Volcanic ash forecast using ensemble-based data assimilation: an ensemble transform Kalman filter coupled with the FALL3D-7.2 model (ETKF–FALL3D version 1.0)

Soledad Osores,Juan Ruiz,Estela Collini,Arnau Folch

doi:10.5194/gmd-13-1-2020

Abstract

Abstract. Quantitative volcanic ash cloud forecasts are prone to uncertainties coming from the source term quantification (e.g., the eruption strength or vertical distribution of the emitted particles), with consequent implications for an operational ash impact assessment. We present an ensemble-based data assimilation and forecast system for volcanic ash dispersal and deposition aimed at reducing uncertainties related to eruption source parameters. The FALL3D atmospheric dispersal model is coupled with the ensemble transform Kalman filter (ETKF) data assimilation technique by combining ash mass loading observations with ash dispersal simulations in order to obtain a better joint estimation of the 3-D ash concentration and source parameters. The ETKF–FALL3D data assimilation system is evaluated by performing observing system simulation experiments (OSSEs) in which synthetic observations of fine ash mass loadings are assimilated. The evaluation of the ETKF–FALL3D system, considering reference states of steady and time-varying eruption source parameters, shows that the assimilation process gives both better estimations of ash concentration and time-dependent optimized values of eruption source parameters. The joint estimation of concentrations and source parameters leads to a better analysis and forecast of the 3-D ash concentrations. The results show the potential of the methodology to improve volcanic ash cloud forecasts in operational contexts.

Highlights

Volcanic ash dispersal forecasts are routinely used to prevent aircraft encounters with volcanic ash clouds and to define flight rerouted trajectories, avoiding potentially contaminated airspace areas
This paper presents a first analysis of the ensemble transform Kalman filter (ETKF)–FALL3D system using different observing system simulation experiments (OSSEs) in which synthetic observations of ash column mass loadings are simulated and assimilated
For simplicity and without loss of generality, we will assume here a MER given by the Mastin et al (2009) scheme, which depends on the fourth power of the top height of the eruptive column and does not account for wind effects, and a Suzuki vertical mass distribution (Pfeiffer et al, 2005) that is an empirical vertical ash mass eruption rate distribution that assumes no interactions with the surrounding atmosphere; it is assumed that the shape of the vertical flow rate is the same for all particle sizes and is given by z z λ

Summary

Introduction

Volcanic ash dispersal forecasts are routinely used to prevent aircraft encounters with volcanic ash clouds and to define flight rerouted trajectories, avoiding potentially contaminated airspace areas. In the aftermath of the 2010 Eyjafjallajökull volcanic eruption in Iceland, safety-based quantitative criteria for air traffic disruption were introduced, originally based on ash concentration thresholds and, more recently, on engine-ingested dosage (Clarkson et al, 2016) These scenarios involve the implementation of quantitative ash concentration forecasts, which require better model input constraints, on ash emission rates and/or on model initialization. Ash mass loading error covariance matrix Concatenation of the state vector xt and the estimated model parameters σtf mean and covariance matrix equal to Pat. Ash mass loading error covariance matrix Concatenation of the state vector xt and the estimated model parameters σtf mean and covariance matrix equal to Pat These equations can be difficult to solve explicitly for high-dimensional systems due to the large size of Pt and Rt , but several methods have been proposed to address this issue and to implement the ensemble Kalman filter in high-dimensional systems. One of the main advantages of this approach is that finding the analysis ensemble mean requires inverting a k ×k matrix, which is significantly cheaper than inverting the n×n matrix for the case in which k n (which is usually the case for high-dimensional applications of the filter)

Objectives

Methods

Results

Conclusion