Building on developments of R-EDA in MACC, EDA's principal objective is to ebnsure increasing and skilful use of new trace gas and aerosol measurements or retrievals, within a scenario of changing earth observation data compositions, retrieval versions and model configurations. These developments will be provided for prototype operational use in MACC-II sub-projects ENS and EVA.
Environement remote sensing data from space-bourne sensors are increasinlgly available, but also subject to changes and modifications. Further to this, the waelth of data of present satellite missions in the troposhere is not as straightforward to handle as in the stratosphere, since in many cases information is only avaialble in terms of column data, such as NO (2) tropospheric columns or AOD. Making best use of these both valuable and expensive data for air-quality forecasting is the key objective of MACC-II sub-project EDA.
The ability to provide analysed chemical fields, along with locally and temporally resolved error estimates, is essential, making use of a multi-model ensemble approach.
Assimilated chemical constitues include all rountinely measured species and particulate matter, irrespective of direct or indirect data provision. The resulting analyses of the chemical state of the atmosphere are required to be chemically consistent, that is to preserve the integrity of natural balance within chemical families.
While in MACC R-EDA activities laid the foundations for prototype operational chemical data assimiliation (DA) procedures, with primary focus on in situ data, MACC-II aims to (i) improve the regional DA system with respect to an extended or partly novel set of remote sensing data and in situ observations, (ii) generalise and refine the core data assimilation capacities of individual air-quality models in terms of operational stability, accuracy, and reliability, (iii) validate the assimilation system with refined methods and algorithms, and (iv) provide for validity of the data assimilation in case of extraordinary geophysical or atmospheric events.
Due to the nature of the developments in EDA, all work packages are intepreted as being in compliance with the characteristics of "joint research and development activities" in section 2.2 of FP7-SPACE-2011-1, Guide for Applicants. The work to be undertaken includes items with:
More specifically, the following objectives have been set.
First is the steablishment of an extended suite of data to be assimilated from space-borne sensors:
In addition, further ingestion of the in situ data is planned. These are collected by EEA EIONET, comprising constitutes under EU directive rule, including particulate matters in the limits of PM (10) and PM (2.5) and IAGOS-ERI aircraft data, which provide profiles over some major European cities.
The prototype operational data assimilation algorithms will be improved in terms of operational stability, robustness, and accuracy. The data assimilation algorrithms employed by partners range from variational methods to complexity-reduced Kalman filter techniques. In either case, some features must be updated or designed to perform in a more robust way. Therefore, this work package comprises some general featutres to be included in the data assimilation algorithm, which are applicable for typical cases, rather than special events like fire and mineral dust. This includes
The quality of the data assimilation products will be validated. A posteriori validation is the means in data assimilation, to evaluate the skill of results. It tests the BLUE (Best Linear Unbiased Estimate) propoerties by means of consistency of results. The basic approach is X2 - validation, which allows for a check of covariance consistancy. (However, it cannot test the practical benefits during routine operation, which is left to related work in sub-projects ENS, EDA and VAL.) The precedure is performed in two steps:
In all cases tests are also performed with data withheld from assimilation and used for validation.
The data assimilationalgorithm will be enabled to account for special or hardly predictable events. These include
The data assimilation algorithms will comply with corresponding model devlopments in ENS to simulate these events. As a critical part, the data assimilation algorithm will be enabled to identify the model performance with respect to successfully modelling the event and its elevated constituet burden. The assimilation procedure is then required to estimate and provide the proper forecast error covarainces to the special event data assimilation process.
Feedback will be provided to data providers, the EEA/EIONET in particular in the case of the in situ data, based on lessons learnt from all these activities. It will be organised through WP16 of the OBS sub-project, in which RIUUK, the lead organisation for EDA, will participate.
In EDA expertise and efforts are combined to provide for the most efficient and synergetic assimilation system developments. In practice, this is especially important for satellite data, which do not only vary from sensor to sensor, but are always made available with changing data set versions and mostly improving quality.
It is a typical feature of data assimilation algorithm prototype operations that properties of the system procedures emerge by statistical evaluation in the course of routine applications. Any lessons learnt, such as bias and covariance estimations, must be fed back to the assimilation system developers.
While EDA primarily serves ENS nad EVA, it also linked with the MACC-II sub-projects on integrated global data assimiliation, production and services (GDA), global reactive gases (GRG), global aerosols (AER), acquisition of observations (OBS), emissions (EMI), and fire data assimilation (FIR). Apart from the core data flow from OBS, information from these sub-projects will be either expertise of coded information of mostly auxiliary data (in terms of assimilation parlance). This collaboration ensures that EDA partners will be provided with the same advanced data bases from related sub-projects.
Building on existing code, partners will further develop their own data assimilation code individually, which will be of variational or complexity-reduced Kalman filter type, both of type BLUE. In analogy to the multi-model ensembles used in R-EVA and R-ENS, the diversity of existing assimilation algorithms at individual centres will be maintained, apart from further developments, where exchange of modules and common improvements will be implemented. This ensures continuity from MACC and previos acheivements.
Milestones and description of work
The organisation of work follows the four identified objectives, and is detailed below. Data assimilation developments need to be assessed in an operational-like environment, with comparisons based on sound statistics. To avoid ambiguous interpretation due to multiple system changes, the milestone allocation is decided to include only two steps, that is a mid term upgrade, and a finalupgrade. For the first milestone, gas-phase related improvements and x 2 - based validation of a standard case study, selected from the prototype operational runs is scheduled. For the second milestone, aerosol-related improvements and special-event focussed work is stipulated.
A special proto-type operationalization work package is omited intentionally in MACC-II EDA, as this work step is now integrated in the individual work tasks.
WP91 EDA: Satellite data assimilation
Satellite data for the troposphere will be prepared for prototype operational and case study use in data assimilation, with emphasis on efficient data processing, applying most recent error estimates from retrievals and balancing with in situ data. The character of most of the work is "Adapation on the service chain to the new input data".
Task EDA.1.1 NO (2) column data assimilation modules
No(2) tropospheric columns from SCIAMACHY, OMI and GOME-2 will be integrated in the data stream, and related observation operator will be developed or updated, making use of averaging kernel information. Care will be taken that errors due to limited model top height will not degrade the assimilation result. Since NO(2) and O(3) are closely related, CERFACS will evaluate the impact of the analyses of those gases using 3D-VAR/Valentina suite.
Task EDA 1.2 MOPITT data assimilation module
CO data from MOPITT will be used to ingest coarse troposheric CO profile values. The impact of MOPITT data assimilation will be evaluated against in situ observations. MOPITT data will also be taken as an opportunity to update model boundary values, as an alternative to global model boundary.
Task EDA 1.3 IASI data assimilation module
IASI data will be integrated in the data stream, and related observation operator will be developed or updated, and used to upgrade free troposheric ozone values.
CNRS-LISA will continue assimilation if IASI ozone observations, to consolidate and optimize assimilation work performed already in MACC. Beyond this, CNRS-LISA will provide an IASI ozone product, for example partial troposheric 0-6km ozone columns, over the GEMS/MACC area for assimilation in regional CTMs. The averaging kernel, a major part of the observation operator, is part of the product. At present, data can be delivered for case studies, i.e. over entire summer seasons, with a delay of several months.
Operational NRT delivery of these data is an option for the future that could be pursued if of interset for MACC-II. CERFACS will test the analysis of partial O(3) column retrievals performed CNRS/LA FROM THE iasi DATA.
Task EDA 1.4 MODIS, SEVIRI AOD data assimilation module
AOD from MODIS and SEVIRI will be integrated in the data stream and used to update particulate matter values. The related AOD observation operator of the aerosol module of the chemistry model (and its adjoint) will be developed and integrated in the DA algorithm.
WP92 EDA: Assimilation algorithm extension
This work package introduces a couple of improvements of the assimilation algorithm, which will substantially extend its applicability, while improving robustness, and stability. The work is made in the frame of "Making minor improvements to existing production systems to maintain performance". The final task "investigates performance problems arising" from preoperational experiences.
Task EDA 2.1 Aerosol data assimilation
The special problem in aerosol data assimilation is that the information is available as integrated particulate matter, PM(10) in most cases, or as AOD. On the othr hand, air quality models typically have a very detailed multi-component aerosol chemistry and dynamics module. The aerosol data assimilation is either introduced or refined to address the following issues:
Task EDA 2.2 Bias correction scheme
The bias correction scheme will account for typical time-dependent model biases as found by EVA and ENS evaluation. The bias correction schemes will follow the suggestions of Dee et al..
Task EDA 2.3 Dynamic covariance evolution
This work task focuses on the optimal exploitation of background/forecast information from ensemble runs and the related formulation as covariances. The forecast/background error covariance matrix will be made flow dependent. This is done either by application of the "NMC"-method or by exploitation of the ENS model ensemble, or by exploitation of the model ensemble used of rthe Ensemble Kalman filter. The choice is based on individual partners' assimilation algorithm configuration. Special set-ups are also admissible; For example for CNRS-LISA, the model ensemble used in the ENKF method is used.
The aim is to have a model ensemble which is physically sound taking into account uncertainty in meteorology, emissions, chemistry, deposition, and boundary condidtions. This ensemble will be refined with respect to MACC. The evolution of the uncertainty fields will be analysed.
Task EDA 2.4 4D-Var emission inversion
The emission data inversion procedure based on the 4D-Var will be applied to identify emission correction factors, but using the MACC-II grid configuration. Emission factors to be improved comprise observed species in the first place, but also not observed, yet influenced by chemical coupling.
Task EDA 2.5 Prototype operationalization update due to EVA and ENS feedback
The work included here is reserved for activities perceived at necessary following the evaluation of experience from sub-projects ENS and EVA, but not covered by other deliverables. It is expected that regional-, seasonal- or weather-dependent deficiences will be identified, which are due to the set-up of the data assimilation algorithm. These activities include software proven too time consuming, boundary layer height dependent and chemical convariances, and similar features.
WP93 EDA: A posteriori validation
This work package uses the state-of-the-art approaches of a posteriori validation of data assimilation algorithms. While testing of developments for individual compents of other work tasks are part of the related work packages, the overall validation of the data assimilation algorithms is performed in this work package. On the basis that all partners use some type of BLUE data assimilation algorithm, it must be noted that only least square type optima can be considered. (Other skill scores will be applied in EVA.) The first validation activity within the time span of the first milestone makes use of x2-valiadation with spatial and temporal adjustments. For the second milestone, caste studies with NRT data withheld for control are performed. All work in this package is to guarantee the quality of deliverables.
Task EDA 3.1 X2-validation regionally and seasonally resolved
Models perform with different skill levels in regions with different emission impact, seasons or weather conditions. Likewise, the observation representativity may vary. The X2 testing activities will ensure a proper balance between the observation and forecast error covariance matrices, based on pertinent related performance statistics of models and observations.
Relaxing a constraint made in MACC, the application of X2-tests focussing on proper definition of forecast errors and observation/representativity errors will now be broken in terms of typical regions (remote to urbanized) and times (season, weather conditions). The data base for assimilation will be restricted to NRT operational data.
Task EDA 3.2 Case study analyses with non-NRT data
Two test episode types, a standard case and special event cases (biomass burning, mineral dust, volcanic emissions) will be selected to be analysed with the prototype operational data assimilation algorithm. However, an expanded data set, the observations of which are not available in NRT, will be taken for a more comprehensive control. These data comprise scientific in situ data of opportunity, as made available by partners from most probably national or other third-party sources, and additional remote sensing data. These include EARLINET lidar, CALIPSO, IAGOS-ERI (as long as not available in NRT) and TES.
WP94 EDA: Mineral dust, volcano and fire data assimilation
This work package comprises the delopment of measures to address data assimilation modifications for special transient atmospheric conditions. These include biomass burning, mineral dust events, and, based on recent experience in Europe due to the Eyjafjallajokull eruption, volcanic emissions. This work package engages in "Developing suitable solutions and upgrades to maintain product delivery", while exceptional events occur.
Task EDA 4.1 Wild fire data assimilation module
The data assimilation algorithm will be extended to identify the chemistry models' change toward local fire emission, estimate the forecast errors, adapt the covariance formulation, and perform the data assimilation procedure.
Task EDA 4.2 Mineral dust data assimilation module
The data assilimation algorithm will be extended to identify the chemistry models' change toward mineral dust load and, if applicable, dust controlled lateral boundary values obtained from global models, estimate the forecast errors, adapt the covariance formulation, and perform the data assimilation procedure.
Task EDA 4.3 Volcanic emission data assimilation
The data assimilation algorithm will be extended to identify the chemistry models' change toward vlocanic emission load and, if applicable, vlocanic emission controlled lateral boundary values obtained from global models, estimate the forecast errors, adapt the covariance formulation, and perform the data assimilation procedure.
Figure 11 Information exchange between EDA and other sub-projects within MACC.