Operational Forecasting

One of the objectives of the ESA-funded project VAST is to improve volcanic ash forecasts based on atmospheric models by a seamless integration of inverse modeling and ensemble prediction approaches. While some of the methods applied are already well-established and considered as state-of-the-art in the scientific domain, they have not yet been fully implemented and tested in an operational environment with the characteristic time and resources constraints. VAST develops an operational demonstration system which comprises a modeling and ensemble modeling service, an ash source term estimation service using state-of-the art EO products and various post-processing and graphical tools. The service is tested and largely developed at the National Meteorological Center (NMC) Vienna (ZAMG) in tight collaboration with the partners of the project. After successful demonstration, validation and documentation, the service will be made available to the interested VAACs and related institutions.

Operational modeling

The operational demonstration tool, figure 1, includes all the automatized and user-selected steps with the exception of the multi-model multi-input ensembles and assumes independent processing of the satellite and meteorological data to be ready in case of an event. The system will produce a two level output product. The first, faster, level is obtained after an initial forecast based on past information of eruptions and simplified parameters. The output is then within the first hours after event notification. The output shall include, among others, concentration forecasts at selected flight levels using the usual low, elevated and high risk levels for ash ( 0.2, 2 and 4 mg/m3) and time series of vertical profiles above European airports and measurement stations. The second, advanced, level includes both inversion and ensemble modeling. The inversion is iterated in case of an extended event and the obtained a posteriori source term(s) is/are used for FLEXPART forecasts with the operational ECMWF forecast and to drive the FLEXPART runs for the multi-input ensemble modeling. The results here include dynamical source term estimations and combined ensemble results of concentrations and concentration time series.

Operational Demonstration Tool

Figure 1: diagram of the operations procedure. The start of the event triggers the process via automatic notification or by hand. Once triggered, the system reads in an internal database to produce a first fast forecast and processed output. At the same time, all the steps needed by the [dynamic]inversion are produced iteratively during the whole length of the event. The trigger also starts the cluster analysis of the 50 EMCWF ensemble members to produce a sub-set of them for a multi input ensemble with the a posteriori(s) obtained in the inversion.

Operational ash source term service

An accurate description of the eruption source term is crucial for a good forecast of the ash/SO2 atmospheric transport and dispersion and its potential influence on the aviation sector. Inversion modeling, combining satellite observations with model-derived source-receptor sensitivities and an a priori estimate, has proven to be a good system to produce accurate estimation of the volcanic eruption source term strength and vertical distribution (Stohl et al. 2011, Kristiansen et al. 2010, 2012). Whereas a research-oriented ad-hoc inversion is a one time inversion run that may involve time- and resource-demanding FLEXPART calculations, the timely treatment of an ongoing event, of a certain duration, requires a different approach with updated warm-started FLEXPART runs, updated meteorology and satellite data. For this, the concept of a dynamic inversion (figure 2) has been developed at NILU and matured within VAST to be implemented in the demonstration operational service. This system, when fully completed, shall allow the update of the source term for fine ash and SO2 and thus the related disperison forecasts several times within the duration of the event.

Dynamic inversion

Figure 2: coarse diagram of a dynamic inversion system with input data every six hours coming in with a 5 hour delay. The first orange simulation corresponds to the initial output level, where fast output is needed and no inversion is made. Each discontinuous green line means satellite data available. Each box represents a one hour interval and shaded grey correspond to releases. Boxes surrounded by a dim grey line means runs (or part of the run) already realized. PD stands for particle dump, needed for warm starts of FLEXPART runs. When we reach a thick black line, and inversion is made, a source term derived and a forecast with it performed (blue arrows)

Ensemble modeling

Multi-input and multi-model (MI and MM hereafter) ensemble techniques have become popular during the last two decades as a pragmatic means to account, under certain assumptions (e.g. Potempski and Galmarini, 2009, Solazzo et al. 2013), for the uncertainty in the meteorological data driving the atmospheric transport models (MI) and the uncertainty introduced by the models themselves (MM), or a combination of both (MI-MM) (e.g. Galmarini et al. 2004 a,b, Galmarini et al. 2013). Ensemble modeling is largely used in fields ranging from, amongst others, weather prediction (with a longer history) to air pollution research applications (Galmarini et al. 2004 a, b), decision support systems (http://ensemble2.jrc.ec.europa.eu) and, more recently, also volcanic ash atmospheric transport and dispersion studies (e.g. Kristensen et al. 2012). After the recommendations from the European Volcanic Ash Cloud Experts Group (EVACEG) the ensemble modeling technique is expected to be as well included in the European operational forecasting systems of ash and SO2 for aviation applications. VAST moves towards this direction.
Multi-input ensembles in VAST
Within the VAST demonstration operational service, MI ensemble modeling is performed with instances of FLEXPART driven by members of the European Center for Medium-range Weather Forecast (ECMWF) Integrated Forecasting System (IFS) [citation needed]. Due to the operational approach, where timings and resources are limited, the currently available 50 ECMWF lower-resolution forecast ensemble members need to be cut down to a lower amount in a meaningful way. This is achieved by following the approach described in Klonner (2013). Briefly, the system performs a first guess forecast to make an estimate of the area affected by the ash cloud. In this process this area is automatically expanded by 1 degree every 12 h after a forecast time of 12 h as exemplified in figure 3.

Safety area 2 degrees Safety area 6 degrees

Figure 3: examples of the addition of an uncertainty area (light blue) to the initial area given by the first guess run. The plume itself combined with the uncertainty area defines the area of interest for the cluster analysis.

Figure 4: sample of concentration forecast of SO4 (used as a typical tracer) for an idealized Ejya event and six of the 50 ECMWF ensemble members. All the concentrations are showed at forecast time t+72 h and height between 9.5 and 10 km agl.

Once the area (geographical extension) per timestep is defined, then the horizontal wind velocity components for a pre-defined vertical level (roughly 10 km agl) and for all the 50 ensemble members are extracted and clustered using a partitioning around medoids method with a weighted Euclidean distance measure (for details the reader is referred to the original work). For each medoid of the final set of 5 to 7 clusters, the full meteorological dataset needed by FLEXPART is extracted and used to drive an instance of FLEXPART that is afterwards evaluated together with the other FLEXPART runs using medoids as input. The variability of the dispersion results for each of the datasets has proven to be significant (figure 3), depending, of course, on the meteorological conditions and event characteristics.
Multi-model / multi-input ensembles in VAST

[This section will be updated with results as results become available]

The partners of VAST have the possibility to work with various atmospheric transport and dispersion models, namely FLEXPART (Stohl et al. 2005), driven by GFS and ECMWF, SILAM (Sofiev et al., 2009), driven by ECMWF, and REMOTE (Langmann 2000), driven by the German Weather forecasts. This allows the possibility of a relatively small MI-MM ensemble, which includes a combination of Lagrangian and Eulerian models. In order to make preliminary evaluation of the data transfer process and results of the ensembles for the cases in the VAST database, a set of exercises was defined. For the operational demonstration system, each of the contributions to the ensemble runs by the different institutions is centralized at the NILU facilities. The demonstration too retrieves then the results from the NILU ftp server and proceeds with the analysis and production of a comprehensive set of summarized output that will ultimately be uploaded onto the website.

Testing and system validation

[This section will be updated with results as available]

The demonstration operational system needs to be tested and validated from two perspectives: firstly, the results should be scientifically accurate and realistic. Secondly, the operational needs must be fulfilled. Both aspects need to be considered together. The database generated in VAST includes several - and different in characteristics - test-cases and data that may be used to this aim. On one hand, accurate ad-hoc inversions as accurate as possible and with no computational constrains (allowing for large number of particles, high vertical resolution at the source and long flexpart runs) are made [link to where this will be] for all the test cases starting for the Grimsvoetn eruption of 2011. For the same cases, constrained inversions are performed at ZAMG to estimate the optimum set-up for an operational system in terms of grid sizes, number of particles, release heights and regularization parameters while keeping track of the computer wall-time, memory and storage required. Comparisons will be shown when available.


1.      Galmarini S., et al. (2004 a) Ensemble dispersion forecasting, Part I: concept, approach and indicators. Atmospheric Environment , 38, 28, 4607-4617
2.      Galmarini S., et al. (2004 b) Ensemble dispersion forecasting, Part II: application and evaluation. Atmospheric Environment, 38, 28, 4619-463
3.      Galmarini, S., Kioutsioukis, Solazzo, E., 2013. E Pluribus unum: Ensemble air quality predictions. Atmospheric Chemistry and Physics. ACP-2012-912, in press
4.      Klonner, R. (2013): Clustering ECMWF ENS ensemble predictions to optimise FLEXPART plume dispersion ensembles. Masther thesis. 
5.      Langmann, B., 2000. Numerical modelling of regional scale transport and photo- chemistry directly together with meteorological processes. Atmospheric Envi- ronment 34, 3585e3598.
6.      Potempski, S., Galmarini, S., 2009. Est modus in rebus: analytical properties of multi-model ensembles. Atmospheric Chemistry and Physics, pp. 9471-9489.
7.      Kristiansen, N. I., A. Stohl, A. J. Prata, A. Richter, S. Eckhardt, P. Seibert, A. Hoffmann, C. Ritter, L. Bitar, T. J. Duck, and K. Stebel (2010): Remote sensing and inverse transport model-ing of the Kasatochi eruption sulfur dioxide cloud. J. Geophys. Res. 115, D00L16, doi:10.1029/2009JD013286.180.pdf
8.      Kristiansen, N. I., A. Stohl, A. J. Prata, N. Bukowiecki, H. Dacre, S. Eckhardt, S. Henne, M. C. Hort, B. T. Johnson, F. Marenco, B. Neininger, O. Reitebuch, P. Seibert, D. J. Thomson, H. N. Webster, and B. Weinzierl (2012): Performance assessment of a volcanic ash transport model mini-ensemble used for inverse modeling of the 2010 Eyjafjallajökull eruption. J. Geophys. Res. 117, D00U11, doi:10.1029/2011JD016844.216.pdf
9.      Sofiev, M., Vankevich, R., Lotjonen, M., Prank, M., Petukhov, V., Ermakova, T., Koskinen, J., Kukkonen, J., 2009. An operational system for the assimilation of satellite information on wild-land fires for the needs of air quality modelling and forecasting. Atmospheric Chemistry and Physics 9, 6833e6847.
10.      Solazzo, E., Riccio, A., Kioutsioukis, I., Galmarini, S., 2013. Pauci ex tanto numero: reducing redundancy in multi- model ensembles. Atmospheric Chemistry and Physics Discussion 13, 4989-5038
11.      Stohl, A., Forster, C., Frank, A., Seibert, P., and Wotawa, G.: Technical note: The Lagrangian particle dispersion model FLEXPART version 6.2, Atmos. Chem. Phys., 5, 2461-2474, doi:10.5194/acp-5-2461-2005, 2005.
12.      Stohl, A., A. J. Prata, S. Eckhardt, L. Clarisse, A. Durant, S. Henne, N. I. Kristiansen, A. Minikin, U. Schumann, P. Seibert, K. Stebel, H. E. Thomas, T. Thorsteinsson, K. Toerseth, and B. Weinzierl (2011): Determination of time- and height-resolved volcanic ash emissions and their use for quantitative ash dispersion modeling: the 2010 Eyjafjallajökull eruption. Atmos. Chem. Phys. 11, 4333-4351.188.pdf