WP13: Model evaluation, assimilation and trend studies

WP13 Objectives

•    Identify specific ACTRIS-2 Level 1, 2 and 3 data products for: 1) near-real-time data provision and interface with operational prediction models for verification, bias correction anchoring and data assimilation; 2) climate monitoring and evaluation of atmospheric composition re-analyses; 3) climate model error quantification; 4) trend assessments
•    Use ACTRIS-2 quality-checked data for yearly model assessment
•    Establish a routine verification stream of selected ACTRIS-2 variables with daily updates
•    Explore the potential of ACTRIS-2 data for assimilation and satellite bias correction by setting up pilot studies
•    Explore the use of ACTRIS-2 data for calibration of global aerosol-climate model processes
•    Quantify the value of measurements for the reduction of climate model uncertainty
•    Develop combined trend products based on climate models and ACTRIS-2 observations

Description of work

This joint research activity aims at connecting the wealth of data which will be collected, archived and made available through VA1 in ACTRIS-2 with current state-of-the-art aerosol/cloud models with assimilation and prediction capabilities on one hand, and climate models on the other.
Available NRT data (level 1) will be used for the routine evaluation of operational models, while quality-checked (QC) and added-value (level 2&3) products generated in NA2 and NA3 will be used for the retrospective assessments of the model simulations (reanalysis/reforecasts). Outreach and user-oriented data dissemination tools developed in NA5will be applied to the outputs of this WP. Feedback will be provided to users and data providers through online plots of time-series of model and observations, and relevant statistics at the measuring sites.
Another important application of the ACTRIS-2 data will be regional data assimilation. The potential of ground-based measurements of ACTRIS-2 aerosol parameters for improvements in the aerosol regional prediction will be explored through pilot studies for extreme events of public relevance, like volcanic eruptions, mineral dust storms and biomass burning events. Building on the growing interest by the global NWP community in using high accuracy data from ground-based networks to constrain satellite data biases, this joint research activity will also test the use of ground-based lidar data to anchor the bias correction for satellite lidar data, using a variational bias correction scheme.
Another priority for this activity will be to demonstrate the value of continuous high-quality measurements for the reduction in uncertainty of global aerosol-climate models. The ACTRIS-2 data are uniquely useful because of the well characterized uncertainties and representativeness.
Finally, several papers published in 2013 (Collaud-Coen et al. and Asmi et al.) showed the value of high quality GAW data to understand trends in atmospheric composition. The potential of long-term QC ACTRIS-2 data for trend assessment will be further exploited here.

Task 13.1:Model evaluation and online verification
(ECMWF, UGR, MetNorway, NILU, CNR)

The goal of this task is to use ACTRIS-2 observations, with focus on lidar extinction profiles, and surface absorption/extinction coefficients to assess the relevant aerosols fields from the models. Specifically, ACTRIS-2 ground-based and profiling data with NRT delivery will be used to establish a routine verification of the ECMWF’s Composition-Integrated Forecasting System (C-IFS) system for aerosol optical properties (extinction, scattering and absorption). Currently, an online verification system is in place for the MACC-II aerosol system using observations of Aerosol Optical Depth (AOD) from AERONET. This system has been extended to include verification of reactive gases from the surface-based GAW stations.  Developments will be made here to extend this verification system to ACTRIS-2 data. An advanced aerosol microphysics model with modal structure (GLOMAP) is also being implemented in C-IFS. Observations of aerosol number concentrations provided by ACTRIS-2 will be needed to evaluate its performance. Cloud-related fields from the ECMWF model will also continue to be monitored in NRT and improvements will be made based on experience accrued with the Cloudnet component of ACTRIS-2.
Further extension to the ACTRIS-2 data evaluation infrastructure will be established in this task, with the goal to create a prototype system in anticipation of the future evolution of the research infrastructure. Among the benefits of daily updates of model-data comparisons, through ad hoc plots to be displayed on the Data Centre website, there will also be the direct and immediate feedback to the data providers. Also, the monitoring and prediction system is envisaged to be of use to support intensive measurement campaigns according to the needs emerging from the NA2 and NA3 activities.
ACTRIS-2 ground-based in situ and remote sensing QC data will also be used to evaluate the models on annual basis. The reanalysis produced with the C-IFS/ECMWF system will be assessed using daily-aggregated data of aerosol size and composition, and aerosol optical properties. Daily and monthly means of relevant statistics (bias, standard deviation, correlations, mean normalized bias, fractional gross errors, etc.) for selected variables will be produced. Scatter plots and Taylor diagrams will be used to assess the various configurations of the models. Upgraded model versions will also be evaluated. Plots will be displayed on the Data Centre website. A similar analysis will be done for the cloud variables. This assessment will be repeated after the release of the QC data on a yearly basis, depending on the data availability.
BSC-CNS/UGR will contribute to establish a NRT model monitoring/evaluation system as well as a delayed evaluation system using ACTRIS-2 data for the models contributing to the WMO Sand and Dust Sort-Warning and Assessment System (SDS-WAS) Northern Africa-Middle East-Europe (NA-ME-E) Regional Centre. This is managed by BSC-CNS and AEMET and includes 10 dust prediction models (BSC-DREAM8b, MACC-ECMWF, DREAM-NMME-MACC, NMMB/BSC-Dust, MetUM, GEOS-5, NGAC, EMA RegCM4, DREAMABOL). Modelled and observed data will be interpolated to a reference standard vertical profile for comparisons. This definition will also allow generation of multi-model products. The model evaluation will focus on two main features: the description of the aerosol layering (peak altitude and shape of the profile) and the aerosol concentrations for all the models. A set of selected statistics adequate for the model evaluation will be applied.

Task 13.2:Potential of ACTRIS-2 data for assimilation and satellite bias correction

Selected case studies will be chosen to demonstrate the feasibility of assimilating ACTRIS-2 data into state-of-the-art analysis and prediction systems both at the global and the regional scale. The case studies will include high-impact events for European citizens, such as dust volcanic eruptions, biomass burning events, dust outbreaks, and high pollution episodes in Europe.
Surface in-situ observations of aerosol optical properties and profiles of lidar backscatter will be assimilated in the 4D-VAR atmospheric composition analysis and prediction system (C-IFS/ECMWF) for the selected cases. The analysed fields will be fed to the regional models to assess whether the inclusion of ground-based data over Europe has the potential to improve the boundary conditions for the regional models, and hence the subsequent prediction for the European region.  
The value of assimilating ACTRIS-2 data in a dust multi-scale prediction system (NMMB/BSC-CTM) will be investigated with focus on ground-based backscatter profiles, together with an indication of dust aerosol type,  additionally, or complementary, to the assimilation of space-borne backscatter profiles, with a data assimilation scheme based on an ensemble Kalman filter approach (the LETKF assimilation system). The analysis of relevant dust outbreak cases is expected to show whether the assimilation of ground-based backscatter profiles can improve the model three-dimensional structure of dust outflows from sources.
The regional model EURAD-IM with its 4D-var technique with adjoint aerosol modules (ensemble Kalman smoother) will be used to assess aerosol source strengths estimation of volcanic eruptions including error assessments, ash dispersion reanalyses across Europe, including air traffic related threshold values, and mineral dust load and transport, including error estimates. This will be done for specific test cases using the ACTRIS-2lidar backscatter profiles. In addition CALIPSO, SEVIRI and other space-borne sensor data will be used as source of operational background information. The additional information content of the ACTRIS-2profiling observations will be quantitatively assessed.
The ECMWF system will be used to explore a new application of ground-based aerosol observations: the anchoring of satellite data bias correction. Satellite data may present biases, which can be assessed either a priori or a posteriori as a result of the assimilation procedure. Currently, at ECMWF the estimation of the bias parameters for most ingested data is performed as part of the 4D-VAR minimization. A bias model based on regression coefficients is assumed, and the regression variables are selected depending on the datasets. The regression coefficients are obtained from the assimilation. This procedure is most successful if there are observations of the same type, which do not require a bias correction and can serve as anchoring points. In this task, we will use the ACTRIS-2 aerosol lidar backscatter data to anchor the vertical profile of the bias correction needed for space-borne lidar data from the CALIPSO satellite. The results of the experiments with and without bias correction will be assessed with independent ground-based measurements.

Task 13.3: Climate model uncertainty and trend assessment

Data will also be used to assess how ACTRIS-2 can help to constrain uncertain model processes in global aerosol-climate models. The approach will be to use “history matching” to constrain a global perturbed parameter ensemble against the measurements. History matching is used to eliminate parts of model parameter space that are implausible when output variables are constrained by measurements. This approach generates a reduced (plausible) parameter range and hence reduced forward model uncertainty. For example, in the Global Aerosol Synthesis and Science Project (GASSP), global in situ aerosol measurements from surface, ship and aircraft platforms are being used to calibrate the GLOMAP aerosol model. In this task we will assess which model processes in the GLOMAP model can be constrained by the ACTRIS-2 measurements and whether long-term evaluation can be used to identify missing processes or model structural weaknesses. The procedure will be repeated every ~12 months as the model processes are refined (through other ongoing projects). This task will deliver a global model with observationally constrained processes, with an estimate of how data from individual sites contributes to the uncertainty reduction.
The potential of long-term high quality ACTRIS-2 data for understanding of trends in atmospheric composition shall be further developed. A methodology shall be put in place to analyse, to produce and to regularly update, e.g. annually or biennially, site-specific and regional trends. Suitable QC ACTRIS-2 variables could be aerosol size, composition, surface aerosol optical properties, and optical property profiles.
An updated-combined aerosol trend assessment will be developed and distributed to both scientific community, and authorities/stakeholders. The data sets and new products will be made available through the ACTRIS-2 DC. A web interface shall be put in place to visualise for ACTRIS-2 sites both hindcast reanalysis model data (MACC, EMEP, AeroCom) and corresponding observations for long time series. The observations themselves will be complemented with output from a “history matched” global model to see whether the uncertainty in the observations and model mask any trends.  The purpose is manifold: detect and make periods apparent where observations deviate significantly from models, either due to atmospheric anomalies or instrument problems (feedback to data providers). A training workshop is planned to explain trend assessment report and long term data visualisation tools.