Conceptual analysis on synergetic retrievals of self-consistent aerosol property climate data records

1 Introduction

The influence of aerosols on climate is determined not only by their global distribution but also by their composition. Therefore, long records of all relevant aerosol properties are needed to allow monitoring of the global and regional climatic impact of atmospheric aerosols. In order to enable climate studies, it is essential that all information obtained from different algorithms is fully consistent, i.e., in agreement with each other matching physical understanding and within the combined uncertainties (Popp et al., 2020). This requirement applies to long time series (“Climate Data Records”, in short CDR) of the same variable obtained from different subsequent sensors as well as to different aerosol parameters obtained from independent retrieval approaches.

So far, a large fraction of the development efforts for satellite-based aerosol retrievals has focused on inverting aerosol information from different single instruments (e.g., summarized in Kokhanovsky and de Leeuw, 2009). The primary output is Aerosol Optical Depth (AOD), as measure of the total abundance of atmospheric aerosols. The information content of single instruments is smaller than necessary to retrieve a comprehensive quantification of all parameters of atmospheric aerosol (as for example, shown by Klüser, et al. (2015) for dust size and composition from IASI). Consequently, there is a need for using auxiliary information to fill this gap (e.g., pre-defined optical aerosol properties, “climatological” vertical profiles, …). Retrievals capable of inverting further aerosol parameters from one instrument (e.g., Fine Mode or Dust AOD) also depend on some auxiliary assumptions and on the sensitivity of the available instrument channels to the various aerosol properties. This leads to inconsistencies even between AOD results from different sensors and certainly for additional parameters inverted from them. Theoretically, synergetic retrievals exploiting several sensors together increase the information content and therefore hold the potential to reduce the dependence on assumed properties and thus improve self-consistency. Consistency is defined as agreement within combined uncertainties and physical understanding of results obtained independently from different sensors, both for the same variable (long records or simultaneous results inverted from different sensors) and for different variables (Popp et al., 2020). Self-consistency means consistency which is achieved “automatically” by inverting several sensors together and thus avoiding many of the reasons for inconsistencies of results obtained from independent retrieval systems.

Inconsistencies between different aerosol products from different single sensor retrievals arise from:

• Different sensors have different sensitivities and sampling
• Different retrievals use different pre-defined aerosol components and cloud masking
• Climate Data Records composed from separate pieces from subsequent similar sensors show steps between them or have possibly larger uncertainties in earlier pieces (weaker sampling, higher pixel-level uncertainties)
• Different aerosol parameters obtained from different sensors exploit different signal sensitivities in different parts of the spectral range or for polarization properties and thus have to use different definitions for the same target aerosol type
• The conversion of retrieval results from sensors in different parts of the Earth radiation spectrum requires complete knowledge of the aerosol properties (spectral dependance of extinction, but also sensitivity to absorption of all chemical components, like hematite only absorbing in the visible); it is therefore difficult to link UV-VIS and TIR instruments (converted to 550 nm)

As an example for inconsistency case (3), Figure 1 shows a CDR of monthly mean AOD at 550 nm from subsequent dual view sensors which exhibits a step between the first part (instruments ATSR-2 and AATSR, 1995–2012) and the second part (SLSTR instruments, 2017–2023) due to opposing oblique viewing directions (back- and forward scattering range of aerosols) as is visible in the larger difference to the ground-based AERONET (Holben, et al., 1998) sun photometer reference time series (black dashed line). Another assessment of strengths and limitations of creating a long CDR from subsequent similar NASA sensors (MODIS, VIIRS) has been done by Sawyer, et al. (2020).

The second example of possible consistency issues regarding case (4) is shown in Figure 2, where global time series of total AOD and Fine Mode AOD from one instrument (SLSTR) and Dust AOD from another instrument (IASI) are plotted together with Sea Salt AOD (from the CAMS model; Antje et al., 2019). It is shown that the sum of the AOD components (DAOD, FMAOD, SSAOD) agrees with the total AOD within its uncertainties (shaded range). However, this good agreement is not achieved on regional basis or with other algorithms.

Overall, it can be stated that there is high confidence in AOD distributions and records as shown by reasonable agreement (offsets, but similar temporal evolution of regional records) despite differences in sensitivities and sampling between satellite instruments. The confidence level is lower for further aerosol properties with decreasing confidence from AOD components (Fine Mode AOD, Dust AOD) over inverted optical (Single Scattering Albedo, Ångstrœm exponent) to microphysical properties (effective radius, refractive indices); qualitative “climatologies”/multi-annual averages can be deduced from most capable sensors (e.g., from MISR multi-angular observations, Lee et al., 2016).

Based on the experiences of the European aerosol retrieval community embedded in the ESA Climate Change Initiative (CCI) in Aerosol_cci (also in dialogue with international colleagues through AEROSAT), best practices have been developed (Popp et al., 2016) to guide a practical implementation of the principles specified by the international Global Climate Observing System (GCOS, 2009) for Climate Data Records. These include full validation (statistics, coverage and capabilities to observe the known regional and seasonal patterns of atmospheric aerosol), assessment of long time series stability, an open development approach, early involvement of users, pixel-level uncertainty propagation.

Synergetic methods exploit a combination of different sensors to overcome some of the limitations of single-sensor retrievals. Synergies can be exploited on different product levels which have unique advantages and disadvantages. On the level of gridded monthly products (level 3), the coverage of AOD can be improved but on relatively coarse resolution (e.g., Mélin et al., 2007: combining AOD from different versions of the same sensor; Tang, et al., 2016 or Yang and Gueymard, 2021: combining AOD from two or more different sensors). Consistent long records can be extracted by quality weighting from monthly gridded data of a larger number of different sensors/algorithms (Sogacheva et al., 2020 for AOD; Sofieva, et al., 2024 for stratospheric aerosols). Gridded daily products can be exploited to combine alike products from polar and geostationary orbiting sensors which allows integration of AOD with higher spatial resolution with diurnal cycle information (Xia et al., 2023; Gupta, et al., 2024).

On the other hand, synergies on the level of the measured TOA reflectances (level 1) can be exploited through retrieval of the combined measured data from several sensors which provide complementary information content. By mathematical principle, exploiting a larger number of linearly independent measurements through combining different sensors has more degrees of freedom which allows retrieval of more aerosol properties. Those properties are then mathematically self-consistent within the retrieval framework, since they are inverted together in one minimization algorithm. However, the combined multi-sensor exploitation faces a number of technical issues (e.g., calibration consistency, temporal alignment) which needs to checked and if filtered can lead to reduced coverage. Also, this does not guarantee automatically, better physical accuracy or independence from common assumptions. On the other hand, the main benefit lies in reducing the need for assumed auxiliary information and thus it has large potential for physical better results (which needs to be proven by their validation). Furthermore, its success depends significantly on a good quantitative understanding of the covariances between the different measurements which is not easy to obtain.

A rather simple sequential two-sensor approach (SYNAER applying a simple least square fit, using AATSR, SCIAMACHY reflectances) has been demonstrated (Holzer-Popp et al., 2002; 2008) and its capability to separate different aerosol types has been studied theoretically (Martynenko, et al., 2010), but also large-scale application showed the limited number of converging pixels by depending on suitable dark fields. Figure 3 illustrates the critical dependance of the information content on AOD (left side) and shows how aerosol types with different dominant components can be separated in a 2-dimensional principal component space (right side). Further examples of synergetic retrievals based on level 1 input are exploitation of NASA’s A-train (e.g., combining the wide-swath radiometer MODIS with the multi-angle radiometer MISR or with the lidar CALIOP), combined use of two radiometers (AATSR, MERIS) onboard ESA’s ENVISAT (Sobrino, et al., 2016) or EUMETSAT’s PmAP combining GOME-2, IASI and AVHRR (Grzegorski, et al., 2021).

The GRASP algorithm (deduced from the AERONET inversion algorithm and extended to a multi-pixel approach) has been applied to several sensor combinations and demonstrated the extended capabilities (Pavel et al., 2025). An information content analysis for a merged optimal estimation retrieval using input datasets from 3 sensors covering UV to TIR spectral range (GOME-2, SLSTR, IASI) has proven the increased degrees of freedom (Stöffelmair, et al., 2025) as shown in Figure 4. Note that absolute numbers of degrees of freedom are too high for all sensors and the combination due to neglecting covariances between the measurements (channels and sensors) and temporal mismatch. On the other hand, calibration errors have been taken into account and common model assumptions avoid technical inconsistencies but still my lead to common biases versus reference measurements. The full value of synergetic multi-sensor retrievals for long CDRs needs to be exploited and proven with future research.

In general, temporal co-location, spatial resolution differences, radiometric calibration uncertainties, angular sampling, cloud screening, quality filtering and cross-sensor error correlations can affect synergetic retrievals. For level 3 synergies, these are (if at all) corrected statistically based on biases versus external reference data such as AERONET (Sogacheva et al., 2020). For level 2 synergies, matching uncertainties can be reduced by using small matching windows in space and time or by multi-pixel algorithms which are able to take into account related temporal differences (Dubovik, et al., 2021). For the purpose of medium resolution multi-annual data records, changes of the aerosol type and to large extent also of AOD can be assumed as small within these differences. The application to quasi-simultaneous sensors entails limiting those differences and also differences in observation geometries. Sampling differences by different cloud screening or quality filtering are also less critical within the quasi-stationary assumption of aerosol parameters as long as the number of remaining valid pixels suffices to observe the overall characteristics of atmospheric aerosol within the joint “super-pixel”. In addition, suitable algorithms need to be able to quantitatively consider all the other reasons by using covariance information (including calibration uncertainties, cross-sensor correlations). Those can be obtained statistically from the sensor measurements (level 1 datasets), from aerosol model datasets and in some cases must be adjusted experimentally (e.g., covariance weighting of different sensors).

2 Review of aerosol retrieval algorithms: principles, limitations, inherent issues

2.1 Aerosol retrieval principles and sensor characteristics

Aerosol retrieval algorithms for passive satellite instruments exploit radiance measurements at the top of the atmosphere (TOA), which are typically converted to reflectances (0–3 µm) or brightness temperatures BTs (3–12 µm). Different passive instrument types offer various combinations of such measurements in a set of different observation channels.

• At different wavelengths from ultra-violet (UV) to thermal infrared (TIR) range,
• In various observation geometries (nadir imagers, dual or multiple views, limb sounders, star occultation),
• At daytime or nighttime (TIR only),
• With spectral resolution varying typically between several 10s or even 100s of nanometers (radiometers) down to sub-nanometer (spectrometers),
• With or without additional measurements of polarization (polarimeters)

Another important characteristic of each satellite sensor is its sampling capability which is determined by its swath width. Best instruments provide ∼2,000 km which enables global coverage at every 1–2 days. Spatial resolution ranges typically from few hundred meters to 10 s of kilometers. Most satellite instruments with channel combinations suitable for aerosol retrieval operate on polar orbiting platforms, which means that on each location on ground at best one observation per day is taken (TIR instruments adding one additional night time observation) usually at a fixed local time (sun-synchronous orbits). All these aspects lead to a significant under-sampling of atmospheric aerosol in time and space. In addition, all algorithms apply different cloud masking to avoid cloud contamination and different algorithms use different quality control–both also contribute to differences in sampling. Altogether, this can lead to significant biases between gridded averaged products from different sensors. Sogacheva et al. (2020) could prove that regional averaged AOD times series from different instruments and algorithms show similar temporal patterns but with constant regional offsets as shown in Figure 5. The highest information content of all passive sensors is provided by multi-spectral, multi-angle polarimeters (MAP-type instruments), but even such powerful sensors miss typically the TIR and UV dimensions and have limited swath width. Other instrument types with complementary strengths and weaknesses are operated on geostationary platforms (providing measurements of the diurnal cycle, but usually with less information content, lower spatial resolution and no global coverage) or using active remote sensing (lidars: with very high accuracy and significant capabilities for vertical profiles, but very limited spatial coverage/no imaging capabilities) or exploiting horizon-viewing geometries (“limb sounding”) to measure aerosol extinction profiles in the stratosphere.

The focus of this conceptual paper is on algorithms to extract long Climate Data Records from passive polar orbiting instruments. All these satellite aerosol inversions deliver information on total column aerosol amount (AOD, column-integrated extinction) at one or more wavelenths. With growing number of independent observations further aerosol properties can be estimated: Components of AOD (multi-spectral/dual view: Fine and Coarse Mode AOD; MAP-type: Absorbing AOD, in short “AAOD”), column effective optical properties (Ångstrœm exponent, Asymmetry Parameter, Single Scattering Albedo). Some sensors are only sensitive to a part of AOD (TIR to Dust AOD only due to its silicate spectral behaviour in the TIR range) and thus can provide complementary pieces of information on the total aerosol loading. Aerosol types (typical prevalent mixtures) and chemical composition need interpretation of clusters of different optically retrieved parameters (e.g., FM-AOD fraction and AAOD to differentiate sulfate and carbonaceous fine mode aerosols) which are however different for each sensor type due to its different sensitivities. Most powerful (MAP-type) instruments can provide column-effective microphysical properties, such as effective radius, non-spherical fraction, refractive indices, size distributions.

2.2 Information content and its limitations

In the inversion algorithms, aerosols are mathematically represented by their spectral optical properties: absorption, scattering (together these add up to extinction), the scattering phase function (directionality of the scattering) and in the TIR their emissions. These optical properties which are directly linked to the measurements of the satellite instruments are related (by Mie or T-matrix theory) to microphysical particle properties (spectral complex refractive index, size distribution, non-sphericity fraction and particle shapes).

As aerosols show no unique spectral lines (as trace gases do), the inversion can only provide indirect results by fitting simulated signals at TOA with the measured reflectances/BTs. Typically, this inversion problem is ill-posed, i.e., there are fewer independent channels as needed to invert all aerosol properties. The more linearly independent channels an instrument can offer, the higher its information content can get. As the measured TOA signal depends also on other parameters in the Earth system (e.g., surface brightness, trace gas concentrations), additional dependencies come into play. While some external dependencies can be avoided (using “window channels” where only few absorbing lines of trace gases occur) others mean a major task in the retrieval (foremost separating the surface signal which often dominates). Here, additional channels (with different viewing angles or with polarization) help and different strategies are used to benefit from them. Finally, the complexity of microphysical and chemical varieties of atmospheric aerosol particles requires a simplifying lumped approach by using pre-defined aerosol components or binned (size) contributions.

All these dependencies of the measured signal on the observing conditions are usually non-linear: wavelength, geometry (observing zenith angle, solar zenith angle, relative azimuth between the two), surface brightness, aerosol loading. Consequently, the available information content of a sensor has a complex dependance on all these parameters (including AOD, itself). This is the basis why combining different channels increases the information content. As further challenge, there are observation conditions when the measured signal has only weak sensitivity to the aerosol content (e.g., by compensating effects for increasing AOD of additional signal from aerosol scattering versus reduced signal by extinction of the underlying surface signal). These conditions are usually quantified by the partial derivatives of the signal as function of AOD (Jacobians) which should not be close to zero (meaning no signal sensitivity to AOD).

Different mathematical concepts are applied to minimize the differences between simulated and measured TOA signals of these channel combinations. Most aerosol inversion algorithms apply the independent pixel approximation neglecting 3D radiative transfer/spatial context and memory effects. Few examples exist like the multi-pixel inversion GRASP (Dubovik, et al., 2021). For low aerosol loading (AOD at 550 nm smaller than 0.1 to 0.15 depending on sensor) the signal shows only very small dependance on other aerosol optical properties so that the information content of the measurements for these properties is highly limited and thus constraining the well-fitting optical properties provides a large range of plausible output. On the other hand, even the retrieved AOD inherits uncertainties from highly uncertain optical properties which become larger with growing AOD. Increasing the information content by combining several instruments does not automatically increase retrieval sensitivity, However, if a sensor with higher sensitivity for low AOD (e.g., dual view) is combined with another sensor with other complementary information on aerosol composition (e.g., IASI for Dust AOD), the overall sensitivity of the combination can be increased.

When the number of independent observations is smaller than the number of output variables (extinction, absorption, phase function at different wavelengths), additional auxiliary information is needed. Depending on the capabilities of an inversion algorithm, this auxiliary information is used as pre-scribed “climatological” average conditions (to fill in missing information) or as a priori information (only as starting point to speed up minimization iterations) and/or as covariance information (defining the range of variability and correlations between different parameters). The way of using these different types of auxiliary information can have significant impact on the quality of the results but cannot be avoided for weaker algorithms/sensors. On the one hand, where needed to use pre-scribed “climatological” information, i.e., fixed properties over the years, these may potentially introduce retrieval biases (in case of true changes of this parameter). Using a priori knowledge should be less critical, but may come with the risk of consequential biases in case of low sensitivity (when a secondary minimum is chosen by a retrieval). On the other hand, if possible, the use of covariances provides only mathematical constraints in a true Bayesian sense and does not create such biases. An example for prescribed monthly 1-degree information is the aerosol component mixing climatology based on AEROCOM multi-model median and AERONET (Kinne, et al., 2006) which is used by Aerosol_cci algorithms with low information content on aerosol properties (Popp et al., 2016).

Finally, each algorithm needs to solve a number of practical issues. For example, applying the algorithms to global, multi-annual datasets requires also technical computing speed optimization, e.g., by replacing time-consuming radiative transfer calculations with interpolation in pre-calculated discretized look-up tables (or recently by machine learning trained on full radiative transfer results). Inversions do sometimes not converge and therefore thresholds for quality filtering need to be determined and applied. Also cloud masking (which means applying thresholds in the continuous transition of any observable from aerosols to clouds) is optimized differently by different algorithms.

2.3 Use of auxiliary information

The most relevant auxiliary information in aerosol retrievals concerns assumptions on optical properties and on vertical distribution of aerosol. Algorithms which rely on pre-computed look-up tables are confined to a limited number of different aerosol mixtures and vertical profiles representative of main typical average situations which encompass a large range of realistically occurring variability.

For this paper we use the term (pure) “aerosol component” to describe a collection of optically homogeneous particles with a unique log-normal size distribution, refractive index and non-spherical shape which may be a mixture of different chemical components which are optically similar. The optical properties of a component are then described by a unique spectral extinction, absorption and phase function (or matrix, if polarization is considered). An “aerosol mixture” is then composed from two or more pure aerosol components by external mixing; as a simplification to reality, internal (core-shell) mixing is not considered. Finally, an “aerosol type” is an aerosol mixture which occurs widely in space and/or time, e.g., “continental aerosol”. Aerosol types show some variability of their mixing fractions, but similar aerosol types usually cluster in a parameter space (e.g., single-scattering albedo vs. Ångstrœm exponent).

One approach specified by Aerosol_cci for aerosol properties after intensive community dialogue defines a small set of 4 aerosol components grouping optically homogeneous aerosol particles which are then externally mixed in their optical properties (de Leeuw, et al., 2015). With these, look-up tables can be created for a limited number of aerosol mixtures which encompass a realistic variability range of atmospheric aerosol properties. The 4 components represent major aerosol species in the global atmosphere: sea salt, dust, fine mode particles. Their average microphysical properties have been defined using AERONET statistics (to extract median values of the effective radius), in-situ measurements (to define a standard log-normal size distribution) and for dust following a published non-spherical transported mineral dust (Dubovik, et al., 2006). In the case of sea salt and dust, two mean conditions (size, absorption, non-sphericity) have been defined. For the fine mode the approach is different in that it avoids defining basic chemical components (black carbon, organic carbon sulfate/nitrate) but rather defines two optical components with one median size (from AERONET statistics) but two extreme absorption states (representing sulfate as weakly absorbing and black carbon as strongly absorbing) to allow describing any intermediate stage by combining these two. While this approach is practically suitable to create a limited number of look-up tables, further variance (e.g., the large size variety of mineral dust particles as a function of lifetime due to faster deposition of larger particles, size changes of fine mode particles due to hygroscopic swelling, dust absorption differences between different source regions on the globe) has to be neglected with such a simple 4-component model. On the other hand, the 4-component set-up matches the typical single instrument information content of 2–5 (Martynenko, et al., 2010). As explained in Section 2.2., any use of pre-scribed “climatological” information may potentially introduce retrieval biases in as far as fixed parameters cannot represent regional aerosol regimes and their hygroscopic growth, internal mixing, dust aging and source-dependent optical properties or any long-term trends in them. This means that algorithms/sensors which fully rely on the 4 components together with their mixing “climatologies” can only take into account long-term average aerosol regimes. On the other hand, cases where a retrieval allows to invert parts of this mixing (e.g., between fine and coarse mode), the dependance is reduced (e.g., on the regional extent of dust outbreaks).

If an inversion has sufficient information content to differentiate the fitting quality between different mixtures, the retrieval output can include not only the aerosol loading (AOD), but also the set of well-fitting aerosol mixtures. Any further step of deducing aerosol types (equal to typical mixtures of the basic components) entails then interpretation of this information and leads to issues of nomenclature and concrete transparent definitions. Since different instruments have different information content and sensitivities, this interpretation between different single-sensor results is subject to intrinsic inconsistencies. For example, an AOD component named as “dust” derived from a UV-VIS dual view instrument and from a TIR spectrometer may not be perfectly matching.

Regarding vertical distribution of aerosols, two common approaches in aerosol retrievals are the use of a small set of standard atmospheres for latitude bands (tropical, mid latitudes, Arctic) and seasons (summer, winter) or for separate aerosol components (Aerosol_cci: de Leeuw et al., 2015), since the TOA signal in visible bands (but not as clearly in the UV and the TIR) depends less on the vertical aerosol distribution as compared to the influence of surface brightness and AOD.

3 A conceptual framework for synergetic aerosol retrievals

3.1 Underlying user requirements for climate applications

Underlying user requirements for aerosol products for climate applications (monitoring, process studies, trend and anomaly analysis) are documented by the Global Climate Observing System (GCOS-245, 2022) and furthermore in the User Requirements Document of the ESA Aerosol_cci (Aerosol_cci, 2026) based on the AEROCOM modelling community needs. As satellite retrievals are sensitive to optical properties of aerosols, AOD is the best-established concept for its products. Also, for the utilization of satellite-based aerosol products by modelers, a well-established meeting point is AOD, and if applied to components (as AOD contributions) or to other properties (e.g., absorbing AOD) a definition conflict can be avoided. These requirements are not focusing specifically on satellite products (and are thus not limited to their expected capabilities), but are meant to allow answering the relevant scientific questions. GCOS specifies requirements for horizontal, vertical and temporal resolution, uncertainty and stability on three levels with increasing strictness (threshold, breakthrough, goal). AEROCOM additionally discusses accuracy needs as function of the horizontal and temporal resolution (which relates well with different satellite product levels and model grid resolution and translates accuracy requirements to regional quantities). The parameters required by GCOS start from total AOD and extend to Single Scattering Albedo (as measure of absorption), extinction profiles (as measure of vertical distribution), chemical composition, numbers of cloud condensation nuclei (CCN) and size distribution. Not all of those quantities are directly feasible for satellite retrievals (chemical composition, CCN numbers), but proxies are in scope for them (optical for chemical components, Fine Mode AOD or Ångstrœm exponent * AOD for CCN numbers). AEROCOM adds specifically AOD optical components (Fine Mode AOD, Dust AOD, Absorbing AOD) which avoid using misleading intensive parameters (Ångstrœm exponent, Single Scattering Albedo) with very large uncertainties for low AOD. Other requirements for climate applications are not specific for aerosol. The required length of Climate Data Records (CDR) to be able to estimate a climatological state is considered as 30 years, while the capability to identify a trend in them depends on the strength of this trend as compared to the uncertainties in the retrieval results and their stability–this is usually tested by estimating the significance of any trend estimate in a CDR. Finally, self-consistency of CDRs (if put together from separate instruments, even if they are similar) and between parameters (e.g., Fine Mode, Dust and total AOD) are required to understand processes in the Earth System (Popp et al., 2020).

The Aerosol_cci User Requirements Document (Aerosol_cci, 2026) has also explicitly discussed requirements for vertical profile information so that it can be used as “climatological” or a priori information for passive satellite retrievals. Overall, a monthly or seasonal 3D vertical extinction “climatology” (multi-annual median) broken down to major aerosol components is needed as baseline dataset. Reflecting the practical limit for numbers of discretized conditions in look-up tables commonly used in most aerosol retrieval algorithms, this 3D “climatology” needs to be clustered into a small number of different profile types (e.g., exponential, single tropospheric layer, near-surface-layer), possibly with a main parameter such as average layer height or thickness. The “climatology” should then be converted into a mapping of these profile types (and their associated parameters) per month and latitude/longitude grid cell. Additionally, long-term average profiles per season and region or latitude band should separate a background mean state (low to average AOD) from an average (or range) of extreme case conditions (high AOD tail of probability distributions). Overall, each “climatological” parameter should be associated with its variance as input for uncertainty propagation. The impact of any true trends outside this variance is neglected by these “climatologies” and can only be assessed through validation of the results versus any available external reference. Otherwise, such trends have to be added to those variances or strategies have to found using any (however sparse) knowledge on such trends to correct for them. As far as possible, the 3D extinction “climatology” should be provided for separate (optical) aerosol components which are consistent (or “translatable”) to optical components assumed in the retrievals. This last point requires cross-community dialogue (between communities for historic passive retrievals, recent MAP-type retrievals and lidar inversions). Finally, the typing/component definitions need to be absolutely transparent (and best included in the product files) so that passive retrieval experts can easily exploit quantitatively the profiles of aerosol extinction and their underlying optical properties.

In order to be able to serve international climate treaties (e.g., the Paris agreement, 2015), satellite-based CDRs of aerosol properties should provide history as far as possible together with consistent operational continuation into the future to support analysis on 5-yearly intervals for the Paris agreement Global Stocktake. In most cases, regional monthly long records are sufficient for climate applications while the need for high resolution products (on scales similar to air quality applications) may be limited to impact and adaptation purposes.

3.2 Available sensor lines

Table 1 summarizes the polar orbiting passive satellite instruments operated since the early 1980s until the end of the 2020s (specified end of life of currently operational satellite platforms) which offer (more or less suitable) measurements for aerosol retrieval. The sensors are grouped with colors by similar measurement principles and/or common operating agencies. From top to bottom, the table shows:

• Multi-spectral, multi-angle polarimeters (MAP-type instruments): SPEXone (Spectro-polarimeter for Planetary Exploration) and HARP (Hyper-Angular Rainbow Polarimeter) onboard the NASA PACE platform; 3MI (Multi-acquisition Multi-spectral Band and Multi-polarization) onboard the EUMETSAT Metop-SG satellite; POLDER (Polarization and Directionality of the Earth’s Reflectances) onboard the French PARASOL satellite.
• Multi-spectral (and multi-angle) radiometers of NASA’s EOS program onboard the TERRA (and AQUA) platforms: MISR (Multi-angle Imaging SpectroRadiometer) and MODIS (Moderate Resolution Imaging Spectroradiometer.), and the MODIS successor VIIRS (Visible Infrared Imaging Radiometer Suite) operated by NOAA on its Suomi NPP platform
• Dual-view radiometers of ESA: ATSR-2 (Along-Track Scanning Radiometer), AATSR (Advanced ATSR) and SLSTR (Sea and Land Surface Temperature Radiometer) onboard European Earth Observation/Copernicus platforms ERS-2, ENVISAT and SENTINEL-3A and -3B
• Thermal infrared sounders: AIRS (Atmospheric Infrared Sounder) onboard the NASA AQUA satellite and IASI (Infrared Atmospheric Sounding Interferometer) onboard EUMETSAT’s METOP platforms
• UV-VIS spectrometers: GOME (Global Ozone Monitoring Experiment), SCIAMACHY (Scanning Imaging Absorption Spectrometer for Atmospheric Cartography), and GOME-2 (GOME version 2) operated by ESA onboard ERS-2, ENVISAT and by EUEMTSAT onboard METOP platforms
• UV spectrometers: TOMS (ESA’s) on several NASA platforms (Nimbus-7, Meteor-3 to −5, Earth Probe), OMI (Ozone Monitoring Instrument) onboard NASA’s AURA platform and TROPOMI (Tropospheric Monitoring Instrument) onboard the Copernicus SENTINEL-5-Precursor satellite
• Nadir viewing radiometers: AVHRR (Advanced Very High-Resolution Radiometer) onboard the series of NOAA and the METOP platforms, SeaWIFS (Sea-viewing Wide Field-of-view Sensor) onboard Orbview-2, MERIS (Medium Resolution Imaging Spectrometer) onboard ENVISAT and OLCI (Ocean and Land Color Instrument) onboard Copernicus platforms SENTINEL-3A and -3B

TABLE 1. Overview of passive polar orbiting satellite sensors which can exploited for aerosol retrievals with periods covered by them. The sensors are grouped by similar instrument types/operating agencies with the colours. The vertical red dahsed lines separate 3 phases with improving capabilities of the sensors.

In the first one and half decades shown in Table 1, there are only two sensor lines available, but from ∼1995 and in particular after 2000 the number of different instrument types in orbit increases, so that there is overall no major gap in the combined record where at least two different instruments can be integrated. Regarding retrieval capabilities, three separate phases of sensor capabilities can be identified in Table 1 (as indicated with the two vertical dashed red lines):

• Phase 1 (1980–1995):
• Few sensors with few channels/low aerosol information content
• Phase 2 (1995/2000–2030):
• Many sensors with complementary instrument types and good information content for aerosols, but limited capabilities to exploit polarization (POLDER: 2005–2013), multi-angle views (MISR: 2000–2026)
• Phase 3 (2024–2030):
• MAP-type sensors with maximum information content for aerosols by multi-spectral, multi-angle polarization measurements

In addition, active sensors offer high accuracy and vertically resolved aerosol observations but with limited spatial coverage (only narrow curtain below the platform/no imaging capability across track): CALIOP (Cloud-Aerosol Lidar with Orthogonal Polarization), a backscatter lidar (2007–2021) onboard NASA’s CALPSO platform, and ATLID (Atmospheric Lidar), an extinction lidar (2024–2027) onboard ESA’s EarthCARE mission. Finally, recently a new initiative “Geo-RING” started to integrate several geostationary platforms over the different continents into a consistent reflectance/brightness temperature fundamental data record which can be used as basis for additional cloud and aerosol retrievals with diurnal cycle information–given its early stage this is not discussed further in this paper.

A first analysis of possible quasi-simultaneous sensor combinations (morning/afternoon orbits with temporal co-location e.g., ±1 h, spatially overlapping) delivers for example, the following initial list:

• Phase 1 (1980–1995):
• AVHRR + TOMS: combining UV, VIS (no blue channel), TIR, only few channels at all; morning and afternoon orbits
• Phase 2 (1995/2000–2030):
• MODIS + MISR: very long life-time of two single sensors on the same platform; combining VIS (including blue channels), TIR, multi-angle views
• ATSR-2, AATSR, SLSTR + GOME, SCIAMACHY, GOME-2 + (AIRS), IASI: series of similar sensors; combining UV, VIS (including blue channels), TIR, dual-view
• Adding to the above: SeaWIFS, MERIS, OLCI and/or TOMS, OMI, TROPOMI: adding UV and/or VIS (including blue channels); nadir only/wider swath; UV spectrometers: afternoon local time
• Alterations of the above, aiming at better coverage or resolution and overcoming existing gaps (e.g., AATSR–SLSTR: 2012–2016)
• 2005–2013: all of the above, combined with POLDER (with additional short periods in 1992 and 2002/3)
• Phase 3 (2024–2030):
• All of the above, combined with MAP-type instruments (SPEXone, HARP, 3MI)

3.3 Conceptual framework for the exploitation of multi-sensor synergies

While the most powerful recent sensors allow to extract a number of aerosol properties with lower uncertainty or to derive additional aerosol parameters than older sensors, they can for now not yet deliver Climate Data Records due to their recent launches. On the other hand, older sensors which were often not specifically optimized for aerosol retrievals provide long records covering the last 4 to 5 decades. On the basis of priorities (summarized in Section 3.1) for required parameters and length and gaps in historic satellite sensor records and the sensor capabilities available (summarized in Section 3.2), the task ahead and at the same time the challenge lies in making best use of this large suite of satellite instruments with highly divers capabilities but covering a time period of 50 years. A large asset can be expected by “educating” the weaker earlier sensors (Phase 2 and 1) with knowledge from the most powerful sensors available only in the last few years (Phase 3). This paper presents a conceptual framework for a community effort to advancing synergetic aerosol retrieval methodologies to make best use of this wealth of satellite observations. Altogether, this conceptual analysis aims to establish a roadmap for synergetic aerosol retrievals, paving the way for more robust and comprehensive, i.e., self-consistent climate data records. The basic principles of this roadmap are twofold:

• To exploit complementary multi-sensor configurations in Phase 2 (and 1) to increase the information content of retrievals
• To use the most powerful sensors of the recent years as anchor for algorithms (multi-sensor and single-sensor) applied to the earlier two phases, in particular to better define and thus reduce uncertainties of any assumed pre-defined auxiliary information (aerosol components, vertical profile shapes)

Figure 6 illustrates major elements included into the conceptual framework. It has to acknowledge the three phases of highly different polar orbiting passive sensor capabilities. This is treated in a backwards temporal processing logic, by which multi-annual datasets from the most powerful MAP-type instruments in the third phase and the additional recent lidar capabilities are used to cross-calibrate in overlapping periods and to inform multi-sensor retrieval algorithms for the entire Phase 2. Subsequently the weakest sensors of Phase 1 are cross-calibrated in overlaps with improved algorithms and datasets of Phase 2 and benefit also from “climatological” information of Phase 3.

The key methodological developments needed to achieve this define the following research and development roadmap:

• Develop and extend versatile algorithms which can exploit different multi-sensor configurations in Phase 2 (and Phase 1)
• Extract multi-annual “climatologies” from Phase 3 MAP-type (and concurrent lidar) datasets as auxiliary parameters which are not retrievable with Phase 2 and Phase 1 instruments. These need to contain median, variance, covariances of intensive aerosol properties, separated for average and extreme cases (to better specify the pre-defined aerosol components) and a suitable set of vertical profile shapes
• Conduct information content analysis for different quasi-simultaneous multi-sensor configurations from Table 1 to identify best sensor combinations and optimal combination of channels
• Develop and/or apply methods for cross-calibration in overlapping periods (2024–2030 between Phase 2 and Phase 3 sensors, 1995–2020 between Phase 1 and Phase 2 sensors; possibly using POLDER (the earliest available MAP-type instrument, 2005–2013) as a transfer standard
• Develop and apply methods to use those “climatologies” as pre-defined information or as a priori information and covariances for the versatile multi-sensor algorithms; also finding (iterative) approaches to identify mean and extreme cases so that different “climatological” information can be used.
• Develop methodology to quantify uncertainties and identify instabilities as consequence of inabilities of the use of “climatologies” to account for any true trends in these parameters and develop strategies to correct for any known trends by combining the “climatologies” with (however sparse) ground-based indicators of any relevant trend.

Furthermore, implementation of best practices as defined by Popp et al. (2016) defines additional research and development tasks:

• Evaluate the improved algorithms for Phase 2 and Phase 1 instrument combinations with AERONET (and for Phase 1 with BSRN) measurements (binned statistics of the errors vs. AOD, stability analysis)
• Assess and improve the consistency of assumptions (pre-defined components and profile shapes) between the three phases and instrument types, e.g., specifications developed by EarthCARE (Wandinger, et al., 2023) and Aerosol_cci (Popp et al., 2016)
• Assess and improve the consistency between different aerosol products in the 3 phases. This includes avoiding jumps in long records composed from subsequent pieces of different sensor combinations, assuring consistency of the temporal evolution of all aerosol properties obtained from different sensor combinations and algorithms, and physical consistency between different retrieved aerosol variables (e.g., sum of AOD components must equal total AOD within combined uncertainties). Reflecting the different sampling and retrieval capabilities of the 3 Phases, different uncertainty envelopes are expected in the different periods
• Develop systematic uncertainty propagation beyond AOD using variances of the Phase 3 “climatologies” as input for pre-defined components and profile types. Also, further work on quantitative estimates of error correlations between measurements of the different channels and sensors of Phase 1 and Phase 2 instruments is needed
• Test further approaches (machine learning, post-processing corrections, data assimilation) to augment these development steps. For example, speeding up radiative transfer calculations by machine learning emulators of the time-consuming but solved radiative transfer problem can reduce or even overcome the dependence on look-up tables/limited numbers of discretized aerosol components or profile types (by enabling online radiative transfer)

Figure 6 summarizes the concept overview of the suggested community approach. It is centered around versatile algorithms which produce synergetic long multi-sensor (or single-sensor) Climate Data records (AOD, aerosol properties, all with uncertainties) using “climatologies” of aerosol properties and vertical profiles from the Phase 3 sensors as auxiliary information. Methodological steps needed include information content analysis of the Phase 1/2 algorithms with different sensor combinations, methods to use the “climatologies” as pre-defined or a priori knowledge, cross-calibration between algorithms of different phases (including use of the early MAP-type instrument POLDER as transfer standard.

Examples of versatile algorithms are GRASP, a multi-pixel algorithm based on the AERONET inversion algorithm heritage (Dubovik, et al., 2021) or generally optimal estimation algorithms which allow mathematically consistent exploitation of multiple sensors (Phase 2 and 1) and auxiliary “climatological” (or at least multi-annual) knowledge (a priori, covariances between channels and state vector variables). Also, versatile algorithms need some tuning (e.g., for weights and correlations between different sensors). This is the case as the covariance information is usually incomplete since datasets used to derive it from are not fully able to cover all relevant aerosol and other environmental conditions. Accordingly, even a versatile algorithm needs some experimental optimization for such missing or weakly covered aspects which requires iterative testing and research on efficient strategies.

The “climatological” knowledge can be derived from the more powerful sensors in Phase 3 (MAP-type instruments, EarthCARE) or also from reanalysis (combining model and satellite information, e g. CAMS: Antje et al., 2019; MERRA-2: Randles, et al., 2017). Information content analysis is used to determine most beneficial/relevant sensor and channel combinations. One promising multi-sensor combination for Phase 2 exploits different spectral ranges (UV, VIS, TIR) and dual view capabilities (sensors: GOME-2, SLSTR, IASI) to derive AOD and its contributions of major aerosol components (see Stöffelmair, et al., 2025 for the underlying information content analysis).

Practical methods need to be determined to be able to use this auxiliary information. For example, look-up table approaches are limited in the number of discretized aerosol mixtures or standard vertical profiles for computing cost reasons, so that any “climatology” (regular gridded monthly multi-annual averages) or the large atmospheric variance of optical properties of aerosols needs to be clustered into similar “types” or “components”. Special attention shall be given to treat separately extreme cases (e.g., dust outbreaks where the sea salt fraction within the coarse mode is much lower than in “climatological” background, so that a pre-scribed dust/sea salt mixing fraction needs to be altered) since these are highly important for climate change impact and adaptation studies.

Cross-calibration of older “weaker” sensors in overlapping periods with more powerful younger sensors between the three phases of historic sensors with different levels of information content/capabilities links the non-aerosol specified sensors (Phase 1) or earliest aerosol sensors (Phase 2) with the next-generation powerful aerosol sensors (Phase 3). Such cross-calibration comes on top of the classical validation against ground-based sun photometer station networks, as it allows much higher and equal numbers of matching pixels between algorithms and assures coverage in remote areas without AERONET stations. Temporal and spatial colocation can request overlapping pixels/averaged to similar size and should apply the same approaches as typically used for AERONET validation (e.g., time window of ±30 min, temporal averaging coincident by mean wind speed with pixel size). With this, the diurnal cycle needs not to be considered for quasi-simultaneous observations.

The earliest MAP-type sensor POLDER (2005 - 2013) can be used as transfer standard from Phase 3 to Phase 2 with its much longer overlapping period. However, this sensor flew in an afternoon orbit so that suitable methods for comparison to multi-sensor retrievals in the morning orbit need to be chosen (e.g., on level 3 of gridded products or using additional information on diurnal aerosol stability from sensors on geostationary platforms).

In summary, the roadmap has to start from producing and evaluating multi-annual datasets from Phase 3 instruments as basis for extracting “climatologies” of aerosol properties which cannot be retrieved from Phase 2 sensor combinations. These “climatologies” need then to be clustered into similar conditions usable as pre-defined or a prior knowledge in Phase 2 algorithms. Finally, one needs to adapt the versatile algorithms and their use of pre-defined knowledge to the Phase 1 sensor combinations with lowest information content (i.e., more use of pre-defined information is required). Altogether, this has the prospective goal of producing 50-year long aerosol Climate Data Records from ∼1980 to 2030. Uncertainties will be larger in Phase 1 than in Phase 2 and even lower for Phase 3 and the number of variables which can be retrieved will increase from Phase 1 to Phase 3. This has to be evaluated quantitatively during (by uncertainty propagation) and after (by independent validation) the creation of the new CDRs.

Optimal estimation algorithms support fully consistent uncertainty propagation in line with metrological principles as specified for satellite Earth Observation CDRs by FIDUECO (Mittaz, et al., 2019). An example of consistent error propagation for a simple Phase 1 sensor (AVHRR) from measurement uncertainties to a gridded monthly AOD record over Europe is shown in Figure 7 (from Popp and Mittaz, 2022). The underlying simple algorithm uses climatological mixing (Kinne, et al., 2006; from AERONET and AEROCOM multi-model median) of four basic aerosol components defined by Aerosol_cci (Popp et al., 2016) and applies the FIDUCEO principles to implement stringent systematic metrological recipes including use of error correlation structures for averaged products. In Figure 7, the 10-year AVHRR CDR over Europe agrees well within propagated uncertainties of the AVHRR AOD CDR shown by the green shaded area with a reference CDR from a merged dataset (Sogacheva et al., 2020) which represents an example of a cross-calibration with a Phase 2 result (combining different instruments and algorithms by merging). The effective radii of the fine mode components in this approach were set to 0.14 µm based on AERONET statistics. This value can be re-confirmed from a 1-year average of the RemoTAP dataset derived from the SPEXone instrument (Fu et al., 2025; Fu et al., 2026), where the RMSE of the effective radius has been evaluated as 0.021 (over land) and 0.026 (over ocean). This illustrates how the use of Phase 3 instrument “climatologies” as auxiliary information for a Phase 1 algorithm can be made.

4 Discussion and conclusions

The underlying goal behind this conceptual analysis is to produce long (50 years) satellite-based aerosol Climate Data Records with the best available suite of instruments. Single-sensor retrievals provide significant value and allow monitoring of AOD consistently and with good confidence between different algorithms despite of their differences in sensitivities and sampling. However, these are often not self-consistent due to different necessary assumptions and sensitivities. The strongest instruments are able to provide also quantitative estimates of further aerosol properties (e.g., effective radius, refractive indices) but they are only available on the last few years. Therefore, one promising next step for the aerosol retrieval community is to develop intelligent solutions in order to integrate the satellite measurements (level 1) into synthetic super-sensors to conduct versatile multi-sensor retrievals on these combinations for the long historic records and anchor them into the highest information content of the recent new sensors. The large benefit to be expected lies in improved quality of additional aerosol properties and in their self-consistency. Fortunately, the computing capabilities have grown significantly over the last decades, so that such multi-sensor processing is realistic.

Synergy of different satellite instruments can be exploited on different product levels. These combine either independently retrieved single-sensor aerosol products (level 2 and 3) or directly invert their radiance measurements (level 1) as a super-sensor with higher information content. Level 2 synergy has been demonstrated combining aerosol products from similar sensors onboard different platforms (polar orbiting with geostationary) while level 3 combinations allow integrating different algorithms and/or sensor types on the level of the same output quantity. Level 1 synergy may lead to reduced coverage due to limitations in spatial and temporal overlap windows, while level 3 synergy inherits inconsistencies from the independent retrieval algorithms with differing information content and sensitivities. This paper focuses on level 1 synergy.

It can be argued that each new generation of satellite instruments offers new additional capabilities so that the most recent era of multi-angle, multi-spectral polarimeters (called “MAP-type” instruments) provide significantly larger information content and are thus able to invert more aerosol parameters. However, the wealth of satellite-based aerosol climate-relevant time series dating back to early 1980s comes from much simpler instruments. Here lies the largest potential of synergetic retrievals, ranging from combinations of AVHRR with TOMS over combinations of MODIS and MISR or AATSR/MERIS/SCIAMACHY and IASI and their successor instruments.

The analysis of sensors which are suitable for aerosol retrievals in the last 50 years show three phases with increasing capabilities. From 1980 to 1995 (Phase 1) only few instruments (not specifically aimed at aerosol retrieval) with small number of independent observables are available. Since 1995 (Phase 2), a growing number of different sensors with more channels or observing geometries have been available. The different sensor types also provide complementary information (e.g., different spectral ranges). Since the mid-2020s (Phase 3; with one limited duration predecessor instrument in 2005–2013 in Phase 2), the most powerful multi-angle, multi-spectral polarimeters specifically designed for aerosol retrievals are in orbit which allow inverting the largest number of aerosol parameters, but cannot (for long time) provide Climate Data Records (CDRs). In addition, the first extinction lidar in space (ESA ATLID) has also started to provide measurements on vertical profiles over the last 2 years. Making the best towards best possible consistent aerosol CDRs from this highly divers dataset is the target of this paper.

Accordingly, a conceptual framework is presented for producing (up to 50 years) long Climate Data Records of aerosol parameters (AOD and further aerosol properties) with synergetic multi-sensor retrievals applied to satellite instruments with limited information content but long historic coverage. The basic principles to achieve this are a) exploiting complementary multi-sensor configurations by versatile algorithms to increase the information content of retrievals and b) using the most powerful sensors of the recent years as anchor for those algorithms to better define and thus reduce uncertainties of any assumed pre-defined information (aerosol components, vertical profile shapes).

Identifying optimal sensor/channel combinations can be done with such algorithms by information content analysis (considering also measurement errors). With this, quasi-simultaneous sensors (morning/afternoon orbits, co-location/within ±1 h, coarsest resolution; radiometric cross-calibration check, similar angular constellation, additive cloud masking) are assessed for their complementarity in information content. The target is to increase the number of independent observation channels by using all spectral domains (UV, VIS, NIR, TIR) and multi-angle capabilities, but also combining instruments with different swath width for better sampling. Depending on the capabilities of retrieval algorithms for older sensors, they have to rely on more or less auxiliary information; in the weakest cases, only AOD can be inverted and all other aerosol or other Earth system variables need to be provided externally. Implementing this concept requires research and development on methods to use the “climatologies” of most difficult properties from newest sensors for early sensors and to quantify the sensitivities of the results to those prescribed properties.

Regarding the versatile multi-sensor algorithms, optimal estimation is most powerful. The expected disadvantages of synergetic products with reduced coverage and resolution due to matching needs of the different instruments can be overcome by multi-pixel approaches (GRASP, Dubovik, et al., 2021) which reduce the loss in coverage by adding smoothness constraints in space and time for aerosol and surface parameters respectively.

The added value from most recent sensors focuses on aerosol properties (which can be clustered into representative components) and on vertical profiles from lidar instruments. This holds the potential to overcome after some years of data collection the current sampling and accuracy limitations due to limited coverage of ground-based AERONET stations and due to the fact that AERONET inversion products are also retrieval results with intrinsic assumptions and uncertainties. Also airborne campaigns and in-situ measurements have such representativity limitations and therefore need to be largely extended to develop a systematic understanding with statistically significant sampling of relevant air masses on the globe as discussed in the “Three-way street” for air-borne in-situ measurements (Kahn et al., 2023). On the other hand, the need to systematically collect comprehensive airborne aerosol air mass characterizations across the globe persists as validation reference for any satellite retrieval including MAP-type sensors) and also for climate models.

Further supporting research and development is encouraged which should cover machine learning (e.g., emulators to speed up radiative transfer calculations to allow avoiding use of look-up tables), post-processing approaches for bias corrections, etc. Another underlying task is the full development of level 1 (radiance measurements) uncertainties according to FIDUCEO principles to allow propagation to geophysical and gridded averaged products taking into account uncertainty correlation structures.

Practically, one has to accept less information/larger uncertainties for early phases due to the smaller number of sensors, their lower information content and also their weaker radiometric calibration. Care for best possible level 1 calibration (including uncertainty characterization) remains therefore important. Fundamentally, the basic difference between cloud-free biased satellite products and gap-less model grid averages cannot be altered by synergetic retrievals unless geostationary satellites are included. Also, any consequences of the assumption of external mixing of aerosol components versus the reality of frequent internal mixing in nature will not be altered through multi-sensor products, but may still be somewhat reduced by more realistic aerosol components (multi-annual averages and understanding their variances) from Phase 3 retrievals.

Over time with growing length of the Phase 3 datasets, “climatologies” can be improved and variability/trends of the parameters can be quantified which translate into the Phase 2 retrievals as additional uncertainties. Any long-term trends of any auxiliary parameter used as auxiliary “climatology” will introduce an instability in the Phase 2 retrievals as far as present-day aerosol properties are not representative of historical conditions (due to changes in emissions, aerosol composition, source regions, climate and atmospheric circulation). This could affect the transferability of Phase-3 climatologies to Phase-1 and Phase-2 retrievals and needs to be assessed by existing long-term datasets (AERONET, reanalysis, in-situ) within their given (sampling) limitations. Furthermore, the sensitivities of the synergetic algorithms to those trends should be quantified and strategies need to be developed to combine such “climatologies” with (however sparse) ground-based indicators of any relevant trends.

In conclusion, the promises from this synergetic roadmap are for long consistent Climate Data Records of AOD and also further aerosol properties (at least AOD components or column effective optical properties) covering the entire 50 years of satellite aerosol retrieval monitoring, derived from quasi-simultaneous sensor combinations with complementary information content, following best practices, pragmatic approaches and anchored into the most powerful recent new sensors by using their “climatologies” as best auxiliary information. This conceptual analysis discusses the needs and defines a roadmap of research tasks to achieve this goal including the full demonstration of the CDR capabilities with multi-sensor algorithms for AOD and further aerosol properties. The results are then evaluated by validation with independent ground-based measurements and cross-validation with the stronger sensors in overlapping periods. The roadmap requires a community effort to take this next step in evolving satellite retrieval capabilities over the last 50 years by smart integration of newest most powerful instruments with long records from weaker older sensors. This conceptual analysis aims at stimulating this community effort and providing some guidance to the needed research tasks as basis for further discussions.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Alexander Kokhanovsky, University of Marburg, Germany

Reviewed by: Daniel Perez Ramirez, University of Granada, Spain

Pasindu Liyanage, Nanjing University of Information Science and Technology, China

Yuheng Zhang, Generalized Retrieval of Atmosphere and Surface Properties SAS, France