Recent years have seen the rise of global atmospheric composition models for several applications including climate monitoring, provision of boundary conditions for regional air quality forecasting, and energy sector applications, to mention a few. Typically global forecasts are provided up to five days ahead as it is the case for those issued by the ECMWF/Copernicus Atmospheric Monitoring Service (CAMS). Forecasts are initialized with analysis derived mostly from satellite observations. Specifically for aerosols, assimilated observations are aerosol optical depths (AODs) from the NASA MODIS sensor on board of the Aqua and Terra satellites and the EUMETSAT Polar Multi-sensor Aerosol properties (PMAp) product.
In the ARAS project, started in April 2018, the feasibility of direct aerosol reflectance assimilation is investigated. The motivation of this study is to develop a fast radiative transfer code based on the Optimal Retrieval of Aerosol and Cloud (ORAC) algorithm to be used in the ECMWF's Integrated Forecast System (IFS) for aerosol assimilation at the visible wavelengths. The study also aims are understanding the relative benefits of using AOD products with respect to using the multi-wavelength reflectances (or radiances). While it is still a matter of debate whether it is more advantageous to assimilate raw measurements instead of products, this has not been tried in the IFS system for aerosol analysis.
Moreover, the study will use the dual control variable approach to extract more information from the measurements than the current system. The advantage of using a dual control variable (fine and coarse mode mixing ratios) over a single aerosol variable (total aerosol mixing ratio) stems from the fact that the size information contained in the observations can be propagated to the aerosol fields via the minimization. The dual control variable approach has been revamped as part of ARAS’s development work and it is currently being used to run analysis experiments with the multi-wavelength Aerosol Optical Depth product from MODIS. The system with dual control variable will be used as baseline. The following experiments of 3 months of length each will be run:
1. control experiment with no data assimilation (free-running experiment)
2. baseline experiment with assimilation of MODIS AOD at two wavelengths (670 and 865nm) over ocean only
3. experiment with assimilation of MODIS reflectance at two wavelengths (670 and 865nm) over ocean only
4. baseline experiment with assimilation of MODIS AOD at two wavelengths (670 and 865nm) over land (dark targets only) and ocean
5. experiment with assimilation of MODIS reflectance at two wavelengths (670 and 865nm) over land (dark targets only) and ocean.
Two seasons of interest for aerosol science will be covered: June-July-August (peak of the Saharan dust season) and September-October-November (biomass burning season). The year was chosen to be 2017 due to an intense episode of biomass burning from the Portuguese/Spanish fires and desert dust from the Sahara advected to Northern Europe by hurricane Ophelia (October 2017). The model domain will be global with spatial resolution of 40km (T511) and 60 vertical levels. The two periods will be run in parallel to optimize use of computational resources. Experiments in points 4. and 5. are subject to the development of an adequate surface reflectance model over land and initial tests will focus on the use of data over ocean only. The control experiment will be run with no aerosol data, to assess the relative merits of assimilating AOD and reflectance. The output of the experiments will be verified with independent multi-wavelength AOD data from AERONET and with ground-based PM data. Preliminary results of the assimilation runs and verification will be presented.

An accurate quantification of dust’s spatial and temporal distribution is crucial to correctly characterize the effect that dust has on the earth’s energy balance, and to improve the skill in forecasting not only its concentrations in the atmosphere and on the surface, but also the weather (Pérez et al., 2006; Chaboureau et al., 2011). High accuracy in the characterization of dust can be best achieved by optimally combining observations and model simulations, through data assimilation. Data assimilation has in fact impacted significantly the forecast and monitoring of aerosols (and dust; Benedetti et al., 2009; Niu et al., 2008; Wang and Niu, 2013), and is operational in some of the main aerosol forecasting centres (Sessions et al., 2015). The majority of the aerosol forecasts rely on the assimilation of aerosol optical depth (AOD) retrieved from observations in the visible part of the spectrum (e.g., MODIS Dark Target observations), and pertaining to all aerosol species.

We report here on simulations specifically targeting dust aerosols, and assess the benefit of assimilating dedicated satellite dust observations. Data assimilation simulations are performed with the Multiscale Online Nonhydrostatic AtmospheRe CHemistry model (NMMB-MONARCH; Pérez et al., 2011). This model provides operational dust forecast for the Barcelona Dust Forecast Center under an initiative of the World Meteorological Organization (WMO). It also participates in the multi-model dust ensemble of the International Cooperative for Aerosol Prediction (ICAP) over the global domain, as well as to the WMO Sand and Dust Storm Warning Advisory and Assessment System (SDS-WAS) for Northern Africa, Middle East and Europe.

NMMB-MONARCH is coupled to an ensemble-based technique known as Local Ensemble Transform Kalman Filter (LETKF; Hunt et al., 2007; Miyoshi and Yamane, 2007; Di Tomaso et al., 2017). Ensemble-based techniques use flow-dependent model error amplitudes and structures which evolve during forecast. They should therefore be able to capture better instabilities in the background flow compared to techniques requiring pre-calculated, and constant in time, model error structures. More specifically, our ensemble forecast is designed considering model uncertainties in the physics-based emission scheme and its parameters (with respect to surface winds, soil humidity, and vertical flux distribution at sources), as well as in the meteorological initial and boundary conditions.

We will show here results from the assimilation with such a scheme of satellite observations particularly relevant for dust applications, like MODIS Deep Blue and IASI dust AOD, where the evaluation of the simulations, and comparisons among the different analyses, is performed through data assimilation internal diagnostics, and through the agreement with independent ground-based observations. We will discuss in particular the settings for the representation of model and observation uncertainty, which are key factors in the estimation of the dust analysis.

References

Benedetti, A., Morcrette, J.-J., Boucher, O., Dethof, A., Engelen, R. J., Fisher, M., Flentje, H., Huneeus, N., Jones, L., Kaiser, J. W., Kinne, S., Mangold, A., Razinger, M., Simmons, A. J., and Suttie, M.: Aerosol analysis and forecast in the European Centre for Medium-Range Weather Forecasts Integrated Forecast System: 2. Data assimilation, J. Geophys. Res., 114, D13205, doi:10.1029/2008JD011115, 2009.

Chaboureau, J.-P., Richard, E., Pinty, J.-P., Flamant, C., Di Girolamo, P., Kiemle, C., Behrendt, A., Chepfer, H., Chiriaco, M., and Wulfmeyer, V.: Long-range transport of Saharan dust and its radiative impact on precipitation forecast: a case study during the Convective and Orographically-induced Precipitaion Study (COPS), Q. J. Roy. Meteor. Soc., 137, 236–251, doi:10.1002/qj.719, 2011.

Di Tomaso, E., Schutgens, N. A. J., Jorba, O., and Pérez García-Pando: Assimilation of MODIS Dark Target and Deep Blue observations in the dust aerosol component of NMMB-MONARCH version 1.0, Geosci. Model Dev., 10, 1107-1129, https://doi.org/10.5194/gmd-10-1107-2017, 2017.

Hunt, B. R., Kostelich, E. J., and Szunyogh, I.: Efficient data assimilation for spatiotemporal chaos: A local ensemble transform Kalman filter, Physica D, 230, 112–126, 2007.

Miyoshi, T. and Yamane, S.: Local ensemble transform Kalman filtering with an AGCM at a T159/L48 resolution, Mon. Weather Rev., 135, 3841–3861, doi:10.1175/2007MWR1873.1, 2007.

Niu, T., Gong, S. L., Zhu, G. F., Liu, H. L., Hu, X. Q., Zhou, C. H., and Wang, Y. Q.: Data assimilation of dust aerosol observations for the CUACE/dust forecasting system, Atmos. Chem. Phys., 8, 3473–3482, doi:10.5194/acp-8-3473-2008, 2008.

Pérez, C., Haustein, K., Janjic, Z., Jorba, O., Huneeus, N., Baldasano, J. M., Black, T., Basart, S., Nickovic, S., Miller, R. L., Perlwitz, J. P., Schulz, M., and Thomson, M.: Atmospheric dust modeling from meso to global scales with the online NMMB/BSC-Dust model – Part 1: Model description, annual simulations and evaluation, Atmos. Chem. Phys., 11, 13001–13027, doi:10.5194/acp-11-13001-2011, 2011.

Pérez, C., Nickovic, S., Pejanovic, G., Baldasano, J.M., and Özsoy, E.: Interactive dust-radiation modeling: a step to improve weather forecasts, J. Geophys. Res., 111, D16206, doi:10.1029/2005JD006717, 2006.

Wang, H. and Niu, T.: Sensitivity studies of aerosol data assimilation and direct radiative feedbacks in modeling dust aerosols, Atmos. Environ., 64, 208–218, doi:10.1016/j.atmosenv.2012.09.066, 2013.

Data assimilation as the key component of meteorological forecasts has been established several decades ago and showed strong positive effect on the quality and durability of the weather forecasts. The DA record in the atmospheric composition problems is much shorter and less impressive. Apart from common problems of the assimilation procedures - heavy computations and difficulties of handling strongly non-linear systems - the short memory of the atmospheric composition models is arguably the main roadblock. Assimilation of the model state (concentrations of pollutants) typically loses its impact within less than 24 hours.
The reason for the difficulties is that atmospheric composition problem is heavily driven by boundary conditions (if take them in a generalized sense including emission fluxes) rather than by the model initial state.
The presentation summarizes the experience of the SILAM modelling team in formulating and solving the data assimilation problems, which adjust the emission fluxes instead of (sometimes, in addition to) the model state.
It is shown that in many cases, the task has strong similarities with inverse problems and top-down emission assessment. There are however also significant differences: for the purposes of the forecasts, pre-emptiveness and smoothness of the emission fluxes has to be guaranteed at a time scale of a few days instead of at least the monthly period required for top-down emission assessments.
Several cases will be considered:
- assimilation of MODIS AOD over the African continent, finally aiming at the representation of all aerosol emission sources. It is shown that the quality of the SILAM predictions with such top-down emission stays high even if the simulations are done for another year.
- combined assimilation of in-situ and satellite data for determination of multi-pollutant emission in China
- first approaches to variational short-term assimilation of satellites and in-situ data in Europe, aiming exclusively at the best AQ forecast - and thus softening the regularization requirements to the adjusted emission fluxes
A dedicated discussion will be dedicated to observation operators, whose role in the efficiency of the assimilation procedures can be very high, especially for the satellite instruments.

Korea Meteorological Administration (KMA) has established a plan to automate synoptic observation of Cloud Amount (CA), instead of the naked-eye observation. Thus, National Meteorological Satellite Center (NMSC) of KMA developed CA retrieval model using Himawari-8 observation. To retrieve CA that is determined from 0 to 10 (0~100%), we used Random Forest (RF) which is one of the machine learning. For training the RF model, Target value is calibrated CA of ground at the 22 points in the South Korea during daytime for 2016. As a predictors and input for this model, 48 variables exploiting the brightness temperature difference (BTD) of IR channels from Himawari-8 were used and considered for the 9×9 pixels (18×18 km2) area to reflect ground observation characteristics. In addition, the accuracy of estimating the CA range from 1 to 9 was improved through sequential system of individual RF models generated by classification learning because of the high frequency of 0% or 100% CA. Details will be given at the conference. As a result of validation from September to December 2015, averaged bias, RMSE and correlation is –0.05, 1.58 and 91%, respectively. Proportion correct ±0 (PC±0) that means the percentile of no difference between observer and model CA is 43.9% and PC±2 is 89.3%. Also, model`s PC±2 in the range of 1 to 9 has increased 20.1% (29.9% → 50.0%) compared to COMS CA for January 2017 in the test. In future plans, Optimal Estimation (OE) method would be applied to retrieved CA to get more accurate CA. This is one of the Optimal Cloud Ananlysis (OCA) only for CA. The retrieved CA is considered as observation data and a priori or background data would be calculated using a regression equation of first guess. Finally, this product will be used to operationally enter CA code for meteorological message from August 2020. And also, we will try to apply the GeoKompSat-2A data that will release from 2020.

The Climate Modelling User Group brings a modelling perspective to ESA’s CCI programme. It does this through the application of CCI data sets in climate models and reanalyses, by developing data analysis tools, by maintaining a user dialogue on the ‘added value’ of CCI datasets and by engaging with key international actors. All this helps to ensure that the CCI data needs of the climate research community are being met.
Using CCI data the CMUG has started publishing the results of using these data in climate models. An assessment by CMUG of the CCI CDRs of the sea surface temperature, ocean colour, sea ice, sea level, soil moisture, land cover, fire, aerosol and ozone datasets will be presented. This will comprise not only looking at the primary variables of each ECV but also the associated uncertainty estimates necessary for climate modelling and reanalysis applications. Results showing consistency between ECVs from the integrated CMUG assessments and through assimilation experiments will also be shown. The tools developed by CMUG for assessing climate models and data will also be demonstrated.
The use of CCI data for climate services will also be discussed.

The Copernicus Atmosphere Monitoring Service (CAMS), implemented by ECMWF on behalf of the European Commission, provides daily analyses and forecasts of global atmospheric composition and European air quality. The global fields are produced using atmospheric composition equations embedded in ECMWF’s Integrated Forecasting System (IFS), which is at the core of ECMWF’s services for Numerical Weather Prediction (NWP). However, there are differences in the set-up of the NWP and CAMS configurations. One key difference is that the NWP analyses use an Ensemble of Data Assimilations (EDA) to ensure flow dependent ‘errors of the day’ form part of the background error covariance matrix in the data assimilation process. CAMS currently does not run such a system and so has to use a static or climatological background matrix.

In data assimilation, the background matrix is responsible for spreading the information from an observation to nearby locations, both horizontally and vertically. The vast majority of satellite observations in atmospheric composition are of total column quantities. Hence, how the observational information is distributed in the vertical through the background matrix is vital to the performance of the CAMS system. This talk looks at the different choices that can be made for the static background matrix and how this can have a large impact on the quality of the analyses. Although using forecast statistics can be considered as the main method for generating the background errors, the importance of understanding both the model and the observational information is demonstrated. It is shown that sometimes model knowledge provides a better method than pure statistics. The advantage of using EDA information and the possibilities of incorporating this in to the CAMS system without running a separate NRT EDA is also discussed.