Aerosol and Clouds II

One of the greater uncertainties in climate observation and modelling is the means by which aerosols interact with clouds. Many mechanisms have been observed and theorised, producing both positive and negative radiative effects. However, the relative real-world importance of these is unclear, which complicates the parametrization of cloud processes within models. This presentation will outline a technique to quantify the variation of cloud micro- and macro-physical properties as a function of aerosol loading.

Satellite observations of localised aerosol sources, such as industrial areas or volcanoes, are used as a natural laboratory where fresh aerosols are injected into an otherwise homogeneous field. Perturbed and pristine conditions can be separated by aligning the retrievals with the local wind vector. Approximately 100 sites around the globe, covering a range of anthropogenic and natural aerosol sources, were analysed. The first indirect aerosol effect is clearly observed, with weaker evidence for cloud invigoration. Liquid water path effects are observed in some circumstances.

The Copernicus missions focused on atmospheric composition and trace gas retrieval operate in the UV/VIS/NIR/SWIR spectral region.
For an accurate trace gas retrieval, also a precise knowledge of the cloud properties in this wavelength region at each given scene is required.
In this work we present the algorithms for retrieving the operational cloud products from the low earth orbit (LEO) Sentinel-5 Precursor and Sentinel-5 missions and the geostationary (GEO) Sentinel-4 mission.
The cloud retrieval algorithms OCRA (Optical Cloud Recognition Algorithm) and ROCINN (Retrieval of Cloud Information using Neural Networks) have their heritage with GOME/ERS-2 and GOME-2 MetOp-A/B, where they have already been successfully implemented in an operational environment.
The ROCINN algorithm retrieves cloud height, cloud optical thickness and cloud albedo from NIR measurements in and around the oxygen A-band, taking as input the cloud fraction computed with the OCRA algorithm that is based on a broad-band UV/VIS/NIR color space approach.
This OCRA color space approach makes use of the assumption that clouds usually have a higher reflectivity than the surrounding surface and that the cloud reflectivity is almost wavelength independent across the UV/VIS/NIR region.
Assigning R, G and B colors to respective broad-band reflectances in the NIR, VIS and UV leads to clouds appearing white in the normalized RGB color space. The scene furthest away from the white point is the one where we expect the least possible amount of cloud contamination.
Joining all these scenes on a global grid allows to generate a cloud-free reflectance background composite map which, together with the measured reflectance, can be used to calculate a radiometric cloud fraction for any given measurement.
The cloud height, optical thickness and albedo retrieved by ROCINN are provided for two different cloud models. One which treats clouds more realistically as layers of scattering water droplets (clouds-as-layers, CAL) and one which treats clouds as simple Lambertian reflectors (clouds-as-reflecting boundaries, CRB).
Initial comparisons and validation results are shown for Sentinel-5P and perspectives are outlined for Sentinel-4.

Windblown desert dust is the most prominent type of aerosol in the low troposphere. It affects the radiation directly through absorption, scattering and emission of light, and indirectly, through interactions with clouds. All radiative effects of dust depend on the horizontal and vertical distribution of that aerosol in the troposphere. The dust vertical distribution is not yet well characterized, at least not enough to allow studying the interactions between dust and clouds or the effect on atmospheric circulation.

The Royal Belgian Institute for Space Aeronomy (BIRA-IASB) has developed a strategy to retrieve dust aerosol vertical profiles from thermal infrared radiances measured by the IASI instrument onboard the Metop satellite series (Vandenbussche et al., 2013). This strategy, based on the optimal estimation formalism, is entitled Mineral Aerosol Profiling from Infrared Radiances (MAPIR) and has been used under ESA’s Climate Change Initiative aerosols project to produce 9 years of dust 3D distributions (Popp et al, 2016). The validation has shown an AOD overestimation but a good mean aerosol altitude (Kylling et al., 2017). Over desert surfaces with low emissivity, there were however frequently convergence issues with MAPIR v3, which lead to a loss of valuable data.

To cope with those weaknesses and to make the processing less costly, current research lead to an updated strategy, MAPIR v4. The algorithm is improved by replacing the radiative transfer code from Lidort (Spurr et al. 2008) to RTTOV (ECMWF NWPSAF) and by making some structural changes in the retrieval method. The better convergence was mainly achieved by implementing the Levenberg-Marquardt modification to the Gauss-Newton iteration technique (Rodgers, 2000). A new dataset of 11 years of dust 3D distributions is currently produced with MAPIR v4 and will be presented. This data set will become available through the Copernicus Climate Change Services.

In this talk, we will describe the updated algorithm, show results of the reprocessing and compare them with Aeronet and CALIOP measurements.

The accuracy of satellite retrievals of tropospheric trace gases (e.g., NO2, HCHO) strongly depends on the accuracy of the corresponding cloud fraction, particularly for small cloud fractions. The main goal of the newly developed MICRU (Mainz Iterative Cloud Retrieval Utilities) algorithm is, therefore, to improve the accuracy of small cloud fractions by improving the retrieval of background reflectance maps. The approach is applicable to various UV/vis satellite cloud retrievals.

The most important feature of our approach is the derivation of the minimum reflectance map for a certain satellite sensor and wavelength range from the measurements themselves. A particular advantage of our algorithm is that it integrates measurements of the entire instrument swath. The algorithm builds on the assumption that the surface is dark compared to clouds. Therefore, it is limited to regions not permanently covered by clouds, ice or snow. The algorithm transforms measured reflectances to Lambertian Equivalent Reflectances (LER), thereby removing the viewing-angle dependent contribution of Rayleigh scattering, yet minimum LER values are found to significantly depend on viewing zenith angle (VZA), time, scattering angle, and reflection angle - mainly due to systematic effects like instrument degradation and surface BRDF effects. Therefore, our approach features a minimum LER parametrised by time, VZA, scattering angle, and reflection angle. The comparison between different MICRU channels suggests that the resulting mean standard error of small effective cloud fractions (CF<20%) retrieved in the wavelength range between 375 and 757nm is smaller than 4%. Moreover, cloud fractions retrieved using different instrument channels are consistent over the entire swath.

In this presentation, reflectance background maps and effective cloud fractions are derived from radiance measurements by GOME-2A between 2007 and 2013. Mean cloud fractions are discussed as well as systematic differences to two operational products of the effective cloud fraction: FRESCO and OCRA. Systematic differences (VZA dependence, coast contrasts, biases) between all three algorithms are investigated. Finally, the potential of applying MICRU effective cloud fractions for the retrieval of nitrogen dioxide (NO2) is illustrated based on the QA4ECV data set.

The KNMI aerosol layer height algorithm is planned to be installed into the Level-2 processors of the Sentinel-4/5 missions and is already operationally processing Sentinel-5P/TROPOMI spectra. Currently, the algorithm uses a forward model that computes top of atmosphere reflectance and its derivatives using line-by-line radiative transfer calculations for the oxygen A-band. In total of 3980 calculations per iteration in the optimal estimation retrieval framework have to be done. Depending on the operational hardware, these calculations can take up to 60 seconds per iteration per pixel, with a minimum of 3 iterations for convergence (over the ocean). Due to this, the near real time and the offline aerosol layer height product is severely hindered in its output capacity.

Another significant issue with retrieving aerosol layer height in the near infrared region is the brightness of vegetated surface, which makes land much brighter and consequently reduces the signal aerosols contribute to the top of atmosphere reflectance. As a result, retrievals over land are less reliable than over land, where the background scene is dark enough to allow aerosol signals to be interpreted by the retrieval algorithm unhindered. The Level-2 processor suffers from wasting valuable computational time retrieving inaccurate aerosol layer height values over land.

There are two problems to solve here: a) improve the algorithm's performance over land, thereby improving the computational efficiency of the algorithm, and b) improve the algorithm's speed per iteration. In this presentation, two methods are introduced in order to overcome these issues:

1. The dynamic scaling method has been designed to force the optimal estimation framework to refocus the retrieval algorithm's attention away from surface photon contribution to the aerosol contribution in the top of atmosphere reflectance. This method is a simple augmentation of the existing KNMI aerosol layer height retrieval framework, requiring very simple changes to the code.
2. The line-by-line calculations of the top of atmosphere reflectance and its derivative with respect to model parameters has been replaced by neural networks. This reduces the number of calculations per iteration from 3980 to less than 5 per iteration, thereby significantly improving the algorithm's speed. This method can calculate forward model outputs in a vectorised fashion, computing 100,000 forward model calculations in 12 seconds.
This oral presentation highlights the recent breakthroughs in the aerosol layer height algorithm, which will cement it to be an operational success in the near future.

Strong convection within the Asian monsoon system quickly transports polluted air masses from the boundary layer into the upper troposphere. The physical and chemical processes within this different environment and the fate of the air is subject of current research. The main sources of altitude-resolved observational data from this region stem from satellite limb and lidar observations, like MIPAS on Envisat, CRISTA on SPAS, MLS on Aura, ACE-FTS on SCISAT, and CALIOP on CALIPSO. StratoClim, the first high-altitude aircraft field campaign in the core of the Asian monsoon, took place from Kathmandu, Nepal, in July and August 2017. During this campaign, we operated the GLORIA (Gimballed Limb Observer for Radiance Imaging of the Atmosphere) infrared limb-imaging spectrometer on board the Geophysica airplane. Two-dimensional cross-sections of various trace gases (e.g. ozone, water vapour, nitric acid, PAN, ethane, ammonia) and aerosol information along the flight paths over Nepal and India have been retrieved from the limb spectra covering an altitude range between about 10 and 20 km with spatial resolutions of 0.5-1 km vertically and 3 km horizontally along track. We analyse this airborne dataset with the help of temporally and locally connected satellite data of nadir-pointing instruments like IASI, to infer the underlying composition in the lower and middle troposphere, and geostationary satellites to deduce the presence of convective influence. Further, the GLORIA observations will be linked to the global context and temporal development by utilizing measurements of limb-sounders like CRISTA and MIPAS.

Vipava valley in southwest Slovenia is a representative hot-spot for complex mixtures of different aerosol types of both anthropogenic and natural origin in mountainous terrain. An investigation of aerosol properties throughout the troposphere in different atmospheric conditions was made possible by a deployment of a two-wavelength polarization Raman lidar system combining with in-situ measurements in the valley (in the town of Ajdovščina) from September 2017. Using its aerosol identification capabilities, which are based on particle depolarization ratio and lidar ratio measurements, it was possible to identify predominant aerosol types in the observed atmospheric structures, for example in different atmospheric layers in the case of the stratified atmosphere. Primary anthropogenic aerosols within the valley were found to be mainly emitted from two sources: individual domestic heating systems, which mostly use biomass fuel, and from traffic. A considerable fraction of natural aerosols (for example mineral dust and sea salt), transported over large distances, were observed both above and entering into the planetary boundary layer. According to the properties of different aerosol types, backscatter contribution of each aerosol type was evaluated and the corresponding extinction contribution was derived from lidar observations. Statistical analysis of the presence of different aerosol types was performed on the entire available dataset from 2017 and 2018.