Satellite remote sensing
The satellite is an advanced tool to monitor aerosols and to improve our understanding of aerosol properties. An advantage of the use of satellites is that they can provide routine mea-surements on a global scale.
Presently, the « A-train » constellation of satellites has been established, which includes GCOM-W1 (SHIZUKU), Aqua, CloudSat, CALIPSO and Aura satellites, and travels across the equator at around 1:30 p.m. local time each afternoon. The « A » stands for « afternoon » (see Figure 1.6). These five satellites contain more than 15 separate scientific instruments to observe atmospheric components. For example, the Cloud-Aerosol Lidar and Infrared Pathfinder Satel-lite Observations (CALIPSO) spaceborne lidar provides backscattering coefficient profiles of aerosols and clouds at 532 and 1064 nm from 2 to 40 km above ground level. The radar on-board CloudSat, the polarisation sensitive radiometer POLDER onboard Parasol and MODIS spectra onboard Aqua satellite provide data, which produce the most complete description of clouds and aerosols on a global scale.
In terms of information content, satellite data can be classified into three categories. The first is devoted to scanning the spatial and temporal distributions of aerosols. The second aims at columnar aerosol properties retrievals, through the use of spectral, polarisation, and angular characteristics of backscattered solar light. The third aims at providing information on the vertical profile of aerosols from the surface into the stratosphere. In certain satellites, a combination of sensors may be used.
In terms of the observation geometry, there are two basic types of satellite instruments: vertical (nadir) and horizontal (limb) measurements. In vertical observations, the instrument looks straight down to sense the radiation coming from the Earth, and measures columnar observation. Most instruments employ this concept to provide column integrated products (see Figure 1.6). Horizontal observations can probe the Earth’s atmosphere at various altitudes. They provide a longer path through the atmosphere than vertical observations (see Figure 1.6).
Aerosol properties retrievals from satellites depend on the interaction of the radiation scat-tered and/or absorbed by the atmospheric components and the Earth’s surface. There are two basic types of sensors, passive and active, to receive the radiation. Passive sensors receive the radiation emitted by the Sun and reflected by the atmosphere and the Earth’s surface. Active sensors receive the energy emitted by the sensor itself, such as CALIPSO. Usually, the observa-tions of satellite remote sensing mix all information of atmospheric gases, aerosols, the surface, and clouds [Lee et al., 2009].
Although satellite remote sensing can provide vertical observations, it is very expensive and data are often limited to low horizontal and temporal resolutions. Passive instruments can only retrieve column-integrated aerosol concentrations [Kaufman et al., 2002]. Spaceborne lidars (such as CALIPSO) improve the vertical resolution of aerosol measurements at the global scale [Winker et al., 2003; Berthier et al., 2006; Chazette et al., 2010]. However, spaceborne lidar measurements are only performed along the satellite ground track.
Ground-based lidar networks
The light detection and ranging (lidar) is, along with radiowave detection and ranging (radar) or sound detection and ranging (sodar), a widely used tool for atmospheric remote sensing. It can provide vertical information on molecules and aerosols, as well as the altitude of clouds.
The lidar system consists of a laser transmitter and an optical receiver in parallel, an analog-to-digital converter and data processing by a computer (see Figure 1.7). The intensity of the backscattered light is measured versus time by the optical receiver. The signal profile will be stored by a fast analog-to-digital converter or by a photon counting device. Since the laser is normally vertically directed, it is possible to obtain information on vertical profiles.
The basic lidar measures aerosol backscatter signals at only one wavelength. For obtaining aerosol optical properties from these lidar signals, critical assumptions have to be make in the inversion of the lidar signal. Many techniques have been used in the inversion of the lidar signal in the past years, such as column closure by the use of ancillary optical depth information [Chazette, 2003]. However, basic lidar measurements only estimate either the backscatter or the extinction if only basic lidar data are available. It is because the backscatter to extinction ratio (BER) or the lidar ratio (LR, inverse of BER) that actually depends on the microphysical, chemical, and morphological properties of aerosols, must be assumed constant. On the other hand, measurements of two independent signals can provide accurate retrieval of extinction and backscatter coefficients without assuming that BER or LR is a constant. For example, the Raman-N2 lidar and the High Spectral Resolution Lidar (HSRL) allow the independent determination of aerosol backscattering and extinction coefficients. In the near-range, lidar data are not available due to the receiver field of view and incomplete overlap of the laser beam [Royer, 2011]. A good knowledge of this overlap helps for an estimation of the backscatter or the extinction coefficient in the usually most polluted part of the atmosphere.
Lidar measurements were used in several campaigns to study the impact of anthropogenic and/or natural particles, such as the TARFOX (Tropospheric Aerosol Radiative Forcing Obser-vational Experiment) [Hobbs, 1999; Ferrare et al., 2000], ACE-2 (the Aerosol Characterisation Experiment 2) [Raes et al., 2000], the Indian Ocean Experiment [Ramanathan et al., 2001; Hudson and Yum, 2002], ESQUIF (Étude et Simulation de la Qualité de l’air en Île-de-France) [Chazette et al., 2005], MEGAPOLI (Megacities: Emissions, urban, regional and Global At-mospheric POLlution and climate effects, and Integrated tools for assessment and mitigation) summer experiment in July 2009 [Royer et al., 2011] and during the eruption of the Icelandic volcano Eyjafjallajökull on 14 April 2010 [Chazette et al., 2012].
Following the example of the AERONET (AErosol RObotic NETwork) program, a global net-work of systematic column-integrated aerosol optical depth observations, the Global Atmo-sphere Watch (GAW) aerosol program attempts to coordinate and homogenise different exist-ing networks, and to provide the spatio-temporal distribution of aerosol properties on a global scale [Bosenberg and Hoff, 2007]. This global lidar network is referred to as the GAW Aerosol Lidar Observation Network (GALION). Presently, several lidar networks have been established to perform lidar measurements on continental scales. Those lidar networks include MPLNet (the Micro-pulse lidar network), EARLINET (the European Aerosol research lidar network), AD-Net (the Asian Dust Network), CISLiNet (the Commonwealth of independent states lidar network), REALM (Regional East Aerosol Lidar Mesonet) and ALINE (the American LIdar Network) (see Figure 1.8).
The NASA MPLNET (http://mplnet.gsfc.nasa.gov/) is the only tropospheric pro-filing network that claims global coverage. It was designed to continuously (day and night) measure aerosol and cloud vertical distributions and to provide satellite validations. Nowa-days, Micro-Pulse Lidars (MPL) are operated at 22 stations around the world. Most MPLNET stations are co-located with AERONET sites for producing aerosol and cloud vertical distri-butions by synergy with sunphotometer measurements. The combination of MPLNET with AERONET is also a successful example of the application of the lidar-photometer technique.
The European Aerosol Research Lidar Network, referred to as EARLINET (http://www. earlinet.org) is a voluntary association of scientists concerned with studies of aerosol re-mote sensing by lidars. At present, EARLINET comprises 28 stations distributed over Europe. Since most EARLINET lidars existed before the network was established in 2000, its station sites are rather inhomogeneous.
The Asian Dust Network, referred as AD-Net (http://www-lidar.nies.go.jp/AD-Net/) is an international virtual community designed to continuously observe vertical distributions of Asian dust and other aerosols (including industrial, forest fire, volcanic), and to study aerosol effects on the environment in East Asia. Presently, AD-Net provides continuous observations with automatic lidars. Most AD-Net stations are co-located with skyradiometer from SKYNET (http://skyrad.sci.u-toyama.ac.jp/). Measured data are transferred to the National Institute for Environmental Studies, Japan (NIES) in realtime and processed automatically. Since December 2012, measured data are updated at 00:00 local time on the site of AD-Net, such as attenuated backscattering coefficients at 1064 nm and 532 nm, volume depolarisation ratios, aerosol extinction coefficients, dust extinction coefficients, spherical particle extinction coefficients, aerosol depolarisation ratios at 532 nm and mixing layer height.
The Commonwealth of Independent States lidar network (CIS-LiNet, http://www.cis-linet. basnet.by) was established by lidar teams from Belarus, Russia and Kyrgyz Republic. The goal of this lidar network is to carry out lidar measurements from Minsk to Vladivostok in cooperation with EARLINET and AD-Net. All CIS-LiNet lidar stations are equipped with a three-wavelength lidar (355, 532 and 1064 nm) with Raman channel (387 or 607 nm) and can perform aerosol measurements in the troposphere and stratosphere. Certain lidars can also provide depolarisation measurements.
The Regional East Aerosol Lidar Mesonet (REALM), is operative since 2002. The NOAA Cooperative Remote Sensing Science and Technology (CREST) Lidar Network (CLN, http: //crest.ccny.cuny.edu/) is based on the REALM Lidar Network. However, there are only two groups and three lidars voluntarily contributing lidar data to the network. Campaign style activities are contributed by two other groups.
Finally, the American LIdar Network (ALINE, http://lalinet.no-ip.org/) consists of the existing lidar groups in Latin America for measuring aerosol backscatter and extinction coefficient profiles over Latin America.
There are also other lidar networks, such as Network for the Detection of Atmospheric Composition Change (NDACC, previously NDSC), but with a different research aim. NDACC monitors the upper troposphere and the stratosphere for more than 20 years. NDACC consists of more than 70 high-quality remote-sensing research sites. But only a subset of the stations are equipped with lidars. In this subset, lidars are designed mainly for stratospheric O3 and aerosols.
Air quality modelling of aerosols
As described in previous sections, modellers have developed various chemical transport models (CTM) in the past several years, in order to simulate the formation and evolution of aerosols, and predict levels of air pollution. However, a model can only represent a limited number of species, whereas several millions of organic reactants and products are involved in aerosol and ozone formation. Condensed chemical mechanisms were therefore developed. For example, the chemical mechanism CB05 (Carbon Bond version 5) [Yarwood et al., 2005] can simulate more than fifty gaseous species, such as O3, NO2, NH3 and SO2. The module Super-SORGAM (Secondary ORGanic Aerosol Model) [Schell et al., 2001; Kim et al., 2011a] of the air-quality platform POLYPHEMUS [Mallet et al., 2007] includes 20 aerosol species: 3 primary species (mineral dust, black carbon and primary organic species), 5 inorganic species (ammonium, sulphate, nitrate, chloride and sodium) and 12 organic species. In this section, the development, important processes, numerical approach and performance evaluation of CTM are introduced.
Historical model development
Air quality models have first been developed in the 1970s for air quality impact and scenario studies [Seinfeld, 2004; Seigneur, 2005; Vautard et al., 2012]. In the first-generation models, only a few chemical species and reactions were considered. In the second-generation mod-els, the number of chemical species, reactions and physical processes such as deposition and scavenging was expanded. In the third-generation models, chemistry, meteorology, and other physical processes are now coupled. Current air quality models can simulate both gaseous species and particles, e.g. POLYPHEMUS [Mallet et al., 2007].
Nowadays, in air quality modelling, the evolution equations of chemical species used in CTMs can describe atmospheric transport and chemistry (see Seinfeld and Pandis ) as well as dry and wet deposition. Figure 1.9 describes the main processes that drive the evo-lution of species. The essence of a CTM is the simulation of the time evolution of chemical concentrations over a given domain. It takes into account transport, chemical reactions and additional processes such as emissions and deposition. For « off-line » CTMs, the forcing con-ditions (meteorological conditions, emissions) are obtained from meteorological and emission models.
Let ci stand for the concentration of a chemical species i. The evolution of species con-centrations is governed by a reaction-diffusion-advection equation (equation of reactive disper-sion):
where Kz is the vertical eddy coefficient, Ei is the flux of surface emission and videp is the dry deposition velocity.
In this section, the important processes in aerosol modelling, including emissions, transport, diffusion, chemistry as well as dry and wet deposition, are described. Discussions are based on those processes simulated in the air-quality platform POLYPHEMUS.
The most significant limiting factor on the quality of regional models is emissions [Russell and Dennis, 2000]. The accuracy of emissions have a significant impact on the quality of the model output. The large emission uncertainties are unavoidable, since it is difficult to measure the emissions with high accuracy. Emission uncertainties are also partly due to the high temporal variations of emissions. Over Europe, a commonly used anthropogenic emission inventory is provided by EMEP (European Monitoring and Evaluation Program) [Simpson et al., 2003] with a horizontal resolution of about 50 km × 50 km. However, over smaller-scale air quality domains, higher-resolution emission data are needed, for instance, national level emissions. For example, over Île-de-France, Airparif (Association agréée de surveillance de la qualité de l’air en Île-de-France) provides emission data with a spatial resolution as high as 1 km. Biogenic emissions come from natural sources. They are typically computed using a model, such as MEGAN (Model of Emissions of Gases and Aerosols from Nature) [Guenther et al., 2006].
Air quality models rely strongly on the accuracy of the outputs from meteorological models, e.g. the accuracy of the boundary layer height, wind direction and wind speed. The boundary layer height determines the volume in which pollutants are mixed. Errors in either wind direc-tion or wind speed can lead to errors in the dispersion of pollutants. Pielke and Uliasz  showed that limits on the accuracy of meteorological models led to an upper limit to the ability of air quality models. Over Europe, in the current air quality platform POLYPHEMUS, mete-orological inputs for regional modelling are usually obtained from reanalysis provided by the European Centre for Medium-Range Weather Forecasts (ECMWF). When working on smaller scales, e.g. Île-de-France, the meteorological model WRF (Weather Research and Forecasting) [Skamarock et al., 2008] can be used to provide high spatial resolution meteorological data. However, WRF needs global model outputs, like data provided by the National Centers for Environmental Prediction (NCEP), as input data.
Deposition is a sink for atmospheric chemical species. Its intensity depends on the type of pollutants, weather, location (type and density of vegetation), season (state of the vegetation), etc. Deposition is stronger during the daytime due to the radiation which increases the turbulent vertical transport. Above water, deposition increases with the solubility of species. Deposition is split into two types namely dry deposition and wet deposition (see section 1.1).
where F is the flux, e.g. the quantity of the chemical species which is deposited per unit area per unit time. vd is the deposition velocity and c is the mass concentration.
The deposition process is usually interpreted in analogy to electrical resistances. For gases, the deposition to the surface is supposed to be controlled by three resistances: the aerodynamic resistance (Ra), the quasi-laminar layer resistance (R b) and the surface resistance (Rs). The dry deposition velocity is defined as the inverse of the sum of these three resistances [Wesely, 1989; Sportisse, 2007a]:
For particles, settling by gravitational sedimentation is supposed to operate in parallel with the previous processes. Moreover, the surface resistance is neglected because the particles adhere to the surface. The dry deposition velocity is defined [Venkatram and Pleim, 1999] as follows:
Wet deposition can occur in three ways. First, for gases by uptake, e.g. cloud droplets in a cloud or fog. Second, for particles acting as cloud condensation nuclei. Third, when particles or gas molecule collide with a cloud droplet or a rain drop. Wet deposition can occur both inside and outside a cloud. Scavenging coefficients are used to describe wet deposition in air quality models. Wet deposition is the most important removal process for fine particles in the atmosphere [Anthony and Mary-Scott, 1990] (see section 1.1).
Wet deposition does not only affect the lowest layer of the PBL, the precipitation scavenging affects all volume elements aloft inside the precipitation layer. The wet flux of the chemical species to surface [Seinfeld and Pandis, 1998] is for particles, where Λ is the washout coefficient, c the concentration and dp the diameter of the particles and zTOA the altitude at the top of atmosphere.
CTMs model the evolution of gaseous species and particles. This includes the description of the chemical composition and size distribution of aerosols. Both have an influence on the radiative behaviour of aerosols, on microphysical processes (see section 1.1.3) and on the assessment of health impacts (see section 1.1.1). The evolution of the size distribution and chemical compo-sition of aerosols is governed by the General Dynamic Equation (GDE). It describes the impact of processes such as nucleation, coagulation, and condensation/evaporation (see Figure 1.10).
The smallest aerosols are formed by the aggregation of gaseous molecules through thermo-dynamically stable clusters. Since the mechanism and corresponding species are not yet well understood, binary nucleation (H2O-H2SO4) or ternary nucleation (H2O-H2SO4-NH3) [Zhang et al., 2010a, b] are usually parametrised.
In practice, the Brownian coagulation due to thermal agitation is often the only one modelled. Other effects due to gradients in fields (such as temperature, electric fields, van der Waals forces, etc) are often neglected. Coagulation may be neglected for the evolution of particles above a few µm. However, coagulation significantly affects the number concentration of ur-trafine particles while retaining their total mass [Seinfeld and Pandis, 1998].
Table of contents :
1.1 Atmospheric particulate matter
1.1.1 Health effects
1.1.2 Visibility effects
1.1.3 Climate effects
1.2 Aerosol monitoring
1.2.1 Surface measurements
1.2.2 Satellite remote sensing
1.2.3 Ground-based lidar networks
1.3 Air quality modelling of aerosols
1.3.1 Historical model development
1.3.2 Important processes
1.3.3 Numerical approach
1.3.4 Model performance evaluation
1.4 Data assimilation for aerosol forecasting
1.4.4 Ensemble Kalman filter
1.4.5 Choice of DA method
1.5 Objectives and plan of thesis
2 Assimilation of ground versus lidar observations for PM10 forecasting
2.2 Choice of DA method
2.3 Experimental setup
2.3.2 Input data
2.3.3 Observational data
2.4 Observing system simulation experiment
2.4.1 Nature run
2.4.2 Simulated observations and error modelling
2.4.3 Control run
2.4.4 Parameters of the DA runs
2.5 Choice of the horizontal and vertical correlation lengths
2.6 Comparison between AirBase and 12 lidars network DA
2.7 Sensitivity to the number and position of lidars
3 Modelling and assimilation of lidar signals over Greater Paris
3.2 Experiment setup
3.2.1 POLAIR3D model
3.2.2 Modelling setup and observational data
3.3.1 Modelling of lidar signals
3.3.2 Estimation of zref
3.4 Model evaluation
3.4.1 Model evaluation with Airparif data
3.4.2 Model evaluation with AERONET data
3.5 Comparisons with lidar vertical profiles
3.6 Assimilation test of lidar observations
3.6.1 Basic formulation
3.6.2 Construction of error covariances
3.6.3 DA setup
3.6.4 Results and discussions
4 Assimilation of lidar signals: Application to the Mediterranean basin
4.2 Modelling system
4.3.1 Lidar observations
4.3.2 Observations for validation
4.3.3 Case study
4.4 Assimilation parameter tests
4.4.1 Assimilation period length
4.4.2 Assimilation correlation length
4.4.3 Assimilation altitude range
4.5 Results and discussions
4.5.1 Validation with the BDQA network
4.5.2 Validation with the Barcelona network
4.5.3 Validation with the EMEP-Spain/Portugal network
4.5.4 Validation with the AERONET network
5.2.1 Aerosol modelling
5.2.2 Data assimilation
5.2.3 Lidar observations