Atmospheric Remote Sensing Based on Satellite Oxygen-Band Observations: A Review

Highlights

What are the main findings?

• Passive oxygen-band observations (O₂ and O₄) establish a physically rigorous reference for constraining atmospheric mass, pressure, and photon path lengths.
• Monomeric O₂ absorption and collision-induced O₄ absorption provide complementary vertical sensitivities, with O₄ strongly weighted toward the lower troposphere.
• Spectral features extend from basic cloud boundaries to advanced retrievals of aerosol layer height, dry-air column mass, and greenhouse gas (XCO₂/XCH₄) light-path corrections.

What are the implications of the main finding?

• Oxygen-band retrieval parameters represent effective radiative scattering metrics rather than direct geometric boundaries, which is crucial for data assimilation and active–passive validation.
• Synergizing oxygen absorption with multi-angle spectro-polarimetry and LEO–GEO constellations breaks surface-atmosphere degeneracies, shaping next-generation climate and environmental satellite monitoring.

Abstract

Oxygen-related absorption features provide fundamental constraints for passive atmospheric remote sensing in the reflected-solar spectrum. Because molecular oxygen is well-mixed in the dry atmosphere, O₂ absorption links measured radiance to atmospheric mass, pressure, and effective photon path length, while O₂-O₂ (O₄) collision-induced absorption provides complementary sensitivity to lower-tropospheric photon paths. This review synthesizes the spectroscopic basis, radiative-transfer mechanisms, satellite implementations, retrieval algorithms, and atmospheric applications of O₂ and O₄ measurements from the ultraviolet to the shortwave infrared. Particular emphasis is placed on the O₂ B-band near 687 nm, the O₂ A-band near 760 nm, O₄ bands in the UV–visible range, and the O₂ band near 1.27 µm. These features support retrievals of cloud fraction, cloud pressure, optical centroid pressure, aerosol layer height, surface pressure, dry-air column abundance, and light-path corrections for greenhouse gas observations. We review major algorithmic approaches, including cloud-as-reflecting-boundary models, cloud-as-layer models, DOAS-based retrievals, optimal-estimation frameworks, photon path-length distribution methods, and machine learning or hybrid techniques. Key applications include cloud climatology, aerosol vertical characterization, air mass factor correction, XCO₂ and XCH₄ retrievals, carbon-cycle studies, and multi-mission data integration. Remaining challenges include spectroscopic uncertainty, aerosol and cloud scattering degeneracy, surface bidirectional reflectance, three-dimensional radiative-transfer effects, wavelength-dependent path mismatch, and inconsistent uncertainty characterization. Future progress will depend on improved spectroscopy, active–passive validation, multi-angle polarimetry, physically constrained machine learning, and harmonized multi-mission retrieval frameworks.

1. Introduction

For satellite atmospheric remote sensing in the reflected-solar spectrum, the measured top-of-atmosphere (TOA) radiance is governed by the coupled contributions of gaseous absorption, surface reflection, and scattering by molecules, aerosols, and clouds. Oxygen (O₂) absorption is uniquely useful in this context because molecular oxygen is well-mixed in the troposphere and stratosphere, with a dry-air mole fraction close to 20.95% [1,2]. For a specified pressure level, the oxygen column above that level is therefore controlled primarily by atmospheric mass rather than by highly variable chemistry. This establishes O₂ absorption features as physically rigorous references for passive remote sensing, where the observed band shape and depth constrain the distribution of O₂ optical depth along the ensemble of photon paths. This relationship is nonlinear in general because of line saturation, pressure-dependent line shapes, surface reflection, and multiple scattering; only under suitably weak-absorption approximations does absorption vary approximately linearly with absorber amount (Figure 1) [3,4,5].

Figure 1. Schematic overview of oxygen-band constraints in passive reflected-solar atmospheric remote sensing.

Within the reflected-solar spectrum, four classes of oxygen absorption features are of particular relevance to atmospheric remote sensing. Collision-induced absorption by O₂-O₂ (denoted as O₄) occurs in several UV and visible (VIS) bands, especially near 360, 477, 532, and 577 nm, is widely used in the Differential Optical Absorption Spectroscopy (DOAS) frameworks to constrain cloud and aerosol effects on photon paths [5,6]. The O₂ B-band near 687 nm exhibits moderate absorption and is less affected than the O₂ A-band by the vegetation red edge. The O₂ A-band near 760 nm is the most extensively exploited and information-rich oxygen absorption band, supporting retrievals of cloud-top pressure, aerosol layer height, surface pressure, dry-air column abundance, and light-path corrections for greenhouse gas measurements [7,8,9]. The O₂ band near 1.27 µm further extends oxygen-based path constraints into the SWIR, where its closer spectral proximity to CO₂ and CH₄ absorption bands is advantageous, although robust retrieval requires explicit consideration of atmospheric airglow and dayglow emission [9,10,11,12].

Over the past decades, the utility of O₂ absorption bands has expanded substantially since their early use for cloud-top height (CTH) retrieval from reflected solar radiation. The first spaceborne demonstrations using Soviet Kosmos measurements established the feasibility of exploiting O₂ A-band absorption as an indicator of the effective scattering height [3,13,14]. This concept was later developed through successive satellite missions and instruments, including GOME, SCIAMACHY, MERIS, OMI, GOME-2, GOSAT, OCO-2/3, TanSat, Sentinel-3/OLCI, Sentinel-5P/TROPOMI, DSCOVR/EPIC, PACE, and recent geostationary air-quality missions. Advanced oxygen-band retrievals now extend well beyond cloud-height estimation, providing physically grounded constraints on aerosol layer height (ALH), dry-air column abundance for greenhouse gas (GHG) retrievals, solar-induced chlorophyll fluorescence (SIF), and spectral and radiometric instrument performance [15,16].

Although numerous studies have addressed specific applications of O₂ absorption bands, an integrated synthesis of their spectroscopic basis, satellite implementation, retrieval methodology, and atmospheric applications across the UV–VIS–SWIR reflected-solar domain remains limited. With the rapid development of wide-swath greenhouse gas imagers, geostationary air-quality spectrometers, and multi-angle spectro-polarimetric missions, a cross-disciplinary review is particularly necessary because O₂ band measurements can provide key constraints for cloud climatology, aerosol vertical distribution, tropospheric trace gas retrieval, and greenhouse gas flux inversion [17]. These applications are based on similar radiative transfer principles and retrieval challenges, yet they have often been studied separately by different research communities.

This review aims to provide a unified and critical assessment of current progress in atmospheric remote sensing using O₂ absorption bands. The core scope is confined to satellite passive remote sensing in the ultraviolet (UV) to shortwave-infrared (SWIR) reflected-solar spectrum; limb-emission physics is discussed only to delineate the physical boundary of this scope, whereas thermal-infrared, microwave, radio-occultation, and active-lidar retrieval techniques are not reviewed. Rather than simply summarizing existing instruments, products, and retrieval methods, it focuses on the common physical principles and algorithmic approaches that link different applications. It also discusses the main scientific and technical challenges that must be addressed to improve the accuracy, consistency, and usability of oxygen-band retrievals in next-generation satellite observation systems. The review is organized as follows. Section 2 introduces the spectroscopic and radiative transfer basis of oxygen-band remote sensing. Section 3 reviews the evolution of relevant satellite payloads in the reflected-solar spectrum. Section 4 summarizes major retrieval algorithms for clouds, aerosols, greenhouse gas light-path correction, and surface pressure. Section 5 discusses representative scientific and operational applications. Section 6 concludes by identifying key uncertainty sources and future perspectives on oxygen-band satellite remote sensing, and Section 7 concludes the review.

To ensure a thorough and structured synthesis, the literature for this review was collected from Web of Science, Scopus, and IEEE Xplore covering studies up to 2026. The search strategy combined keywords such as ‘O₂ A-band’, ‘O₂ B-band’, ‘O₂-O₂ absorption’, ‘cloud-top pressure’, ‘aerosol layer height’, and ‘greenhouse gas light-path correction’. Peer-reviewed journal papers, high-impact review articles, and official satellite ATBDs were prioritized, while purely active-radar or microwave studies without passive optical synergies were excluded.

2. Physical Principles and Radiative Transfer Mechanisms

2.1. Spectroscopic Foundations of O₂ Absorption

O₂ absorption features in the UV–VIS–SWIR spectrum are governed by two main physical mechanisms (Table 1 and Figure 2). Molecular O₂ absorption in the O₂ A-band, B-band, and 1.27 µm band originate from weak magnetic dipole transitions of molecular oxygen from the $X^3{\mathsf{\Sigma }}_g^{-}$ ground electronic state to excited singlet states [2,18]. Despite their weak transition strengths, these bands contain well-resolved spectral structures that are sensitive to pressure, photon path length, and the vertical location of scattering layers. At the same time, their absorption is not so strong as to prevent reflected solar photons from reaching clouds or the surface. Quantitative retrievals therefore require line-by-line spectroscopic treatment that accounts for temperature dependence, pressure broadening, speed-dependent line shapes, line mixing, and collision-induced absorption continua in some spectral region [19].

Table 1. Oxygen absorption bands in the UV–SWIR spectrum for satellite atmospheric remote sensing.

Band	Spectral Range	Spectroscopic Basis	Primary Applications
O4 bands	343, 360, 380, 446, 477, 532, 577, 630 nm	Collision-induced O2-O2 absorption	Cloud optical centroid pressure, effective cloud fraction, aerosol/cloud correction for trace-gas AMFs (air mass factors)
O2 B-band	685~695 nm	Moderate magnetic dipole transition $(X^3{\mathsf{\Sigma }}_g^{-}\to b^1{\mathsf{\Sigma }}_g^{+}, v\prime =1)$; weaker than A-band	Cloud pressure/height, aerosol layer height, reduced red-edge sensitivity relative to A-band, SIF at the red emission peak
O2 A-band	758~775 nm	Strong magnetic dipole transition $(X^3{\mathsf{\Sigma }}_g^{-}\to b^1{\mathsf{\Sigma }}_g^{+}, v\prime =0)$	Cloud pressure/height, aerosol layer height, surface pressure, dry-air column, GHG light-path correction, SIF at the far-red emission peak
O2 1.27 µm band	~1.27 μm	Magnetic dipole transition $(X^3{\mathsf{\Sigma }}_g^{-}\to a^1{\mathsf{\Delta }}_g)$; affected by upper-atmospheric dayglow/airglow	SWIR light-path reference for CO2/CH4 retrievals, dry-air/path constraints close to GHG bands, airglow diagnostics

Figure 2. Spectral locations and oxygen-related absorption features from the UV to SWIR.

O₄ absorption differs fundamentally from monomeric O₂ absorption because it arises from collision-induced absorption by transient O₂–O₂ molecular pairs. O₄ is not a thermodynamically stable tetratomic species; rather, it exists only as a transient collision complex formed upon bimolecular encounter. It is important to emphasize the distinct pressure sensitivities of monomeric O₂ absorption versus O₄ collision-induced absorption (CIA). While monomeric O₂ absorption depends linearly on the O₂ number density (reflecting total air mass and path pressure), O₄ absorption scales quadratically with the square of the O₂ density (O₂). Consequently, O₄ bands exhibit significantly higher sensitivity to the lower troposphere, offering complementary physical constraints when decoupling boundary-layer scattering from upper-level clouds or aerosol plumes [20]. This property gives ground-based instruments strong sensitivity to boundary-layer scattering, whereas satellite nadir observations require a more careful interpretation because the observed O₄ signal depends on backscattered solar radiation along a two-way atmospheric path. Rayleigh scattering, surface reflection, clouds, and aerosols jointly shape the effective photon path and hence the O₄ sensitivity [21]. Although O₄ absorption is intrinsically sensitive to lower-tropospheric path-length changes, this sensitivity can be reduced or redistributed when elevated clouds or aerosol layers shield the lower atmosphere [22].

While nadir passive remote sensing utilizes oxygen absorption features, limb-viewing geometries (e.g., SCIAMACHY limb observations) reveal unique non-local thermodynamic equilibrium (non-LTE) physical phenomena. Above ~35 km in the upper stratosphere and mesosphere, molecular collisions become insufficiently frequent to quench excited states, causing O₂ ($a^1{\mathsf{\Delta }}_g$) excited molecules to emit photons. Consequently, the characteristic O₂ A-band, B-band structures invert from absorption features into distinct emission spectra, which are actively exploited for upper-atmospheric airglow diagnostics, temperature profiling, and wind vector retrievals.

The use of O₂ absorption as a pressure and photon-path reference follows directly from hydrostatic balance. For radiation reflected or scattered at a pressure level $p_r$, the dry-air column above that level is approximately $p_r/(m_{\mathit{dry}}g)$, where $m_{\mathit{dry}}$ is the mean molecular mass of dry air per molecule and $g$ is gravitational acceleration. Because molecular oxygen has an approximately constant dry-air mole fraction (${\mathsf{\chi }}_{O_2}=0.20946$), the vertical O₂ column above $p_r$ level can be written as

$$N_{O_2}^{\uparrow }\left(p_r\right)\simeq {\chi }_{O_2} {\displaystyle \frac{p_r}{g m_{dry}}}.$$

(1)

Here, this expression assumes a dry, plane-parallel atmosphere with constant gravity. Corrections for water vapor, the vertical variation in gravity, surface elevation, and atmospheric sphericity are required in high-precision applications, but they do not change the first-order proportionality between O₂ column amount and pressure. This proportionality is the basis for using O₂ absorption to constrain reflecting-level pressure, surface pressure, and photon path length.

2.2. Atmospheric Radiative Transfer

In a plane-parallel atmosphere where the monochromatic optical depth $\mathsf{\tau }$ increases downward from the TOA, the scalar radiative transfer equation (RTE) for solar radiance propagating in direction $\mathsf{\Omega }=(\mathsf{\mu },$ϕ) is

$$\mu {\displaystyle \frac{\partial I\left(\tau ,\mathsf{\Omega }\right)}{\partial \tau }}=I\left(\tau ,\mathsf{\Omega }\right)-J\left(\tau ,\mathsf{\Omega }\right),$$

(2)

where $I(\mathsf{\tau }, \mathsf{\Omega })$ is the monochromatic radiance (W m⁻² sr⁻¹ nm⁻¹), $\mathsf{\mu }$ is the cosine of the propagation zenith angle (positive upward, negative downward), and ϕ is the azimuth angle. In the reflected-solar UV–SWIR spectral region, thermal emission is negligible and the source function $J$ comprises multiple-scattering and direct-beam contributions only:

$$J(\mathsf{\tau }, \mathsf{\Omega })={\displaystyle \frac{\mathsf{\omega }(\mathsf{\tau })}{4\mathsf{\pi }}}{\int }_{4\mathsf{\pi }}\mathrm{P}(\mathsf{\tau },\mathsf{\Omega },\mathsf{\Omega }\prime ) I(\mathsf{\tau },\mathsf{\Omega }\prime ) d\mathsf{\Omega }\prime +{\displaystyle \frac{\mathsf{\omega }(\mathsf{\tau })}{4\mathsf{\pi }}} \mathrm{P}(\mathsf{\tau }, \mathsf{\Omega }, {\mathsf{\Omega }}_0) {\mathsf{\mu }}_0{\mathrm{F}}_0\mathit{exp}\left(-{\displaystyle \frac{\mathsf{\tau }}{{\mathsf{\mu }}_0}}\right),$$

(3)

where $\mathsf{\omega }\left(\mathsf{\tau }\right)\in [0,1]$ is the single-scattering albedo, $\mathrm{P}(\mathsf{\tau }, \mathsf{\Omega }, \mathsf{\Omega }\prime )$ is the scattering phase matrix (or scalar phase function for unpolarized treatment) mapping incident direction $\mathsf{\Omega }\prime$ to scattered direction $\mathsf{\Omega }$, $F_0$ (W m⁻² nm⁻¹) is the extraterrestrial solar irradiance measured perpendicular to the solar beam, ${\mathsf{\mu }}_0$ is the cosine of the solar zenith angle, and ${\mathsf{\Omega }}_0=(-{\mathsf{\mu }}_0,$ϕ₀) denotes the downward solar beam direction. The first term on the right-hand side accounts for diffuse multiple scattering integrated over all incident directions. The second term represents the attenuation of the direct solar beam at depth $\mathsf{\tau }$ via Beer–Lambert extinction ($e^{-\mathsf{\tau }/{\mathsf{\mu }}_0}$), with the factor ${\mathsf{\mu }}_0{F}_0$ converting the beam irradiance to an equivalent radiance source. For polarimetric retrievals, Equation (3) must be written in vector form using the Stokes vector and a Mueller scattering matrix.

In high-spectral-resolution O₂ observations, Rotational Raman Scattering (RRS) by atmospheric N₂ and O₂ molecules causes inelastic photon frequency shifts, leading to the spectral ‘filling-in’ of strong absorption lines (commonly known as the Ring effect). Because roughly 4% of molecular scattering events are inelastic, photons scattered from neighboring continuum wavelengths fill the cores of O₂ absorption lines. This effect is strongly dependent on spectral resolution, path length, and temperature; for sensors with high spectral resolving power (e.g., Sentinel-4 UVN or GHG interferometers like GOSAT), RRS can contribute >1% to line-core radiance and must be explicitly accounted for in radiative transfer models to prevent systematic biases in cloud and surface pressure retrievals. High-accuracy retrievals should include an inelastic-scattering treatment or a validated Ring correction rather than assume a fixed filling-in fraction [23,24].

Oxygen-band retrieval problems are challenging because the measured TOA radiance is controlled by narrow molecular absorption features together with broadband molecular scattering, aerosol and cloud scattering, and surface reflection. Similar oxygen absorption depths may result from different combinations of cloud fraction, scattering pressure, aerosol optical depth, aerosol single-scattering albedo, surface reflectance, and viewing geometry. Retrieval algorithms therefore need to reduce the number of unknowns through simplified physical state vectors, lookup tables, optimal-estimation constraints, or statistical priors, so that the inverse problem is sufficiently constrained by the available spectral information [10,25].

2.3. Photon Path Length

Oxygen-band measurements can also be interpreted in terms of the photon path-length distribution (PPDL). For a normalized PPDL (denoted as $p\left(l\right)$), the TOA radiance can be expressed schematically as

$$I(\mathsf{\lambda })=I_{\mathrm{cont}}(\mathsf{\lambda }){\int }_0^{\infty }p(l)\mathit{exp}\left[-{\mathsf{\tau }}_{O_2}\left(\mathsf{\lambda }, l\right)\right]\mathit{dl},$$

(4)

where $I\left(\mathsf{\lambda }\right)$ is the continuum radiance in the absence of gaseous absorption, $l$ denotes the effective photon path, and ${\mathsf{\tau }}_{O_2}\left(\mathsf{\lambda }, l\right)$ is the O₂ absorption optical depth accumulated along that path. In a simplified homogeneous atmosphere, ${\mathsf{\tau }}_{O_2}\left(\mathsf{\lambda }, l\right)$ may be approximated as $k_{O_2}\left(\mathsf{\lambda }\right)·l$, where $k_{O_2}\left(\mathsf{\lambda }\right)$ is the absorption coefficient. In realistic atmospheres, it depends on the vertical distributions of pressure, temperature, and O₂ number density.

Equation (4) illustrates why different parts of an oxygen absorption band provide different vertical information. Strongly absorbing line centers are dominated by photons with relatively short effective paths, because photons traveling through deeper or longer paths are strongly attenuated. Weak absorption lines and line wings retain greater sensitivity to photons that penetrate to lower atmospheric levels or the surface. Clouds and aerosols modify this distribution by changing where photons are scattered and how long they remain within the atmosphere. Optically thick clouds generally truncate the photon path at the cloud top, whereas optically thin clouds and aerosol layers may either shorten the path by shielding the lower atmosphere or lengthen it through multiple scattering [26]. Because a continuous photon path-length distribution cannot usually be retrieved uniquely from satellite spectra, operational algorithms often report an effective pressure or height.

3. Evolution of Satellite Payload Architectures

The development of satellite payloads for oxygen-band observations has been closely linked to improvements in spectral resolution, radiometric accuracy, detector sensitivity, and viewing capability. These advances have extended retrieval applications from relatively simple cloud-boundary characterization to more demanding inverse problems involving aerosol height, dry-air column constraints, and photon-path corrections for greenhouse gas measurements. The relevant satellite payloads can be grouped into four main observational categories, as summarized in Table 2 and Figure 3.

Table 2. Representative satellite and payloads using oxygen absorption in the solar spectrum.

Satellite/ Payload	Agency	Launch/ Status	Payload Architecture	Oxygen Band(s)	Companion Spectral Bands	Main Applications
Kosmos 149/320	USSR	1967/1970	Multispectral imaging radiometry	O2 A-band	–	CTH
ERS-2/GOME	ESA	1995	UV–VIS dispersive grating spectrometry	O2 A-band, O2 B-band, O4 band	UV–VIS trace-gas bands	CP, ECF, AMF corr.
ENVISAT/MERIS	ESA	2002	Multispectral imaging radiometry	O2 A-band	VIS–NIR ocean/land-color bands	CTP, ECF, cloud screening
ENVISAT/SCIAMACHY	ESA	2002	UV–VIS–SWIR dispersive spectrometry	O2 A-band, O2 B-band, O4, 1.27 µm O2 region	CO2 and CH4 SWIR bands	CP, aerosol/cloud LP effects, GHG support
Aura/OMI	NASA, KNMI, FMI	2004	UV–VIS dispersive grating spectrometry	O4 mainly near 477 nm	UV–VIS trace-gas bands	ECF, OCP, AMF corr.
MetOp-A/B/C/GOME-2	EUMETSAT, ESA	2006/2012/2018	UV–VIS dispersive grating spectrometry	O2 A-band, O2 B-band, O4 band	UV–VIS trace-gas bands	CP, ECF, AMF corr.
GOSAT/TANSO-FTS	JAXA, NIES, MOE	2009	Fourier Transform Spectrometry	O2 A-band	CO2 and CH4 bands near 1.6 and 2.0 µm	XCO2, XCH4, DAC, scat. corr.
OCO-2	NASA	2014	High-resolution dispersive grating spectrometry	O2 A-band	Weak and strong CO2 bands near 1.6 and 2.0 µm	XCO2, Psurf, LP corr., SIF
DSCOVR/EPIC	NOAA, NASA	2015	Multispectral imaging radiometry	O2 A-band, O2 B-band	UV–VIS–NIR narrow bands	CH, ECF, cloud mapping
Sentinel-3/OLCI	ESA/Copernicus	2016	Multispectral imaging radiometry	O2 A-band	VIS–NIR ocean/land-color bands	CTP, ECF, cloud screening
TanSat	CAS, China	2016	High-resolution dispersive grating spectrometry	O2 A-band	CO2 bands near 1.6 and 2.0 µm	XCO2, DAC, aerosol/cloud LP corr.
Sentinel-5P/TROPOMI	ESA/Copernicus	2017	Wide-swath UV–VIS–NIR–SWIR dispersive spectrometry	O2 A-band, O2 B-band, O4 near 477 nm	UV–VIS–SWIR trace-gas bands	ALH, CP, ECF, AMF corr.
FY-3D/GAS	CMA, CNSA, China	2017	High-resolution NIR–SWIR grating spectrometry	O2 A-band	CO2 and CH4 SWIR bands	XCO2, XCH4, DAC, LP corr.
GF-5/GMI	CHEOS, China	2018	High-resolution greenhouse gas spectrometry/spatial heterodyne spectrometry	O2 A-band	CO2 and CH4 SWIR bands	XCO2, XCH4, DAC, LP constr.
GF-5/DPC	CHEOS, China	2018	Multi-angle polarimetric imaging	O2 A-band narrowband channels	Multi-angle polarized VIS–NIR bands	AOD, AMP, ALH
GOSAT-2/TANSO-FTS-2	JAXA, NIES, MOE	2018	Fourier Transform Spectrometry	O2 A-band	CO2, CH4, and CO bands in the SWIR/TIR	XCO2, XCH4, scat. corr.
OCO-3	NASA/ISS	2019	High-resolution dispersive grating spectrometry	O2 A-band	Weak and strong CO2 bands near 1.6 and 2.0 µm	XCO2, urban/PS mapping, SIF
GK-2B/GEMS	KARI, KMA	2020	Geostationary UV–VIS dispersive spectrometry	O4 mainly near 477 nm	UV–VIS trace-gas bands	ECF, OCP, diurnal AMF corr.
TEMPO	NASA, SAO	2023	Geostationary UV–VIS dispersive spectrometry	O4 near 477 nm; O2 B-band spectral region	UV–VIS trace-gas bands	CP, OCP, diurnal AMF corr.
FY-3F/OMS	CMA, China	2023	UV–VIS dispersive grating spectrometry	O4 in the UV–VIS	UV–VIS trace-gas bands	Cloud corr., AMF support
PACE/SPEXone	NASA	2024	Multi-angle spectro-polarimetry	O2 A-band spectral region covered; not a dedicated high-resolution O2 A-band spectrometer	Continuous polarized VIS–NIR measurements	AMP, aerosol type, ALH
MethaneSAT	EDF and partners	2024; contact lost in 2025	High-resolution dispersive spectrometry	1.27 µm O2 band	CH4 and CO2 SWIR bands	XCH4, LP corr.
MicroCarb	CNES	2025	High-resolution dispersive spectrometry	O2 A-band and 1.27 µm O2 band	CO2 bands near 1.6 and 2.0 µm	XCO2, Psurf, LP corr.
Sentinel-4A/UVN	ESA/Copernicus, EUMETSAT	2025	Geostationary UV–VIS–NIR imaging spectrometry	O4 and O2 A-band	UV–VIS trace-gas bands	Cloud/aerosol corr., AMF support
GOSAT-GW/TANSO-3	JAXA, NIES, MOE	2025	Wide-swath grating imaging spectrometry	O2 A-band	CO2 and CH4 SWIR bands	XCO2/XCH4 mapping, emission obs.
FY-3H/GAS-2	CMA, CNSA, China	2025; testing	High-resolution NIR–SWIR imaging grating spectrometry	O2 A-band	CO2, CH4, CO, and N2O bands near 1.6, 2.1 and 2.3 µm	XCO2/XCH4 mapping, DAC, LP corr.
MetOp-SG-A1/3MI	EUMETSAT, ESA	2025; 3MI in commissioning	Multi-angle multi-polarization imaging	O2 A-band narrowband channels	Multi-angle polarized VIS–NIR–SWIR bands	AOD, cloud optics, aerosol type, ALH
TanSat-2	CAS, China	Planned, around 2026	Wide-swath greenhouse gas spectrometry/cloud–aerosol polarimetric support	O2 A-band and oxygen-related cloud/aerosol channels	CO2 and CH4 SWIR bands	XCO2/XCH4 mapping, hotspot obs., aerosol/cloud LP corr.
CO2M/CO2I and MAP	ESA/Copernicus	Planned, around 2027	Wide-swath imaging spectrometry with multi-angle polarimetry	O2 A-band	CO2 and CH4 SWIR bands; NO2 band for plume context	XCO2/XCH4, aerosol/cloud corr., EV

Abbreviations: CTH = cloud-top height; CTP = cloud-top pressure; CP = cloud pressure; ECF = effective cloud fraction; OCP = optical centroid pressure; ALH = aerosol layer height; AOD = aerosol optical depth; AMP = aerosol microphysical properties; AH = aerosol height; AMF = air mass factor; DAC = dry-air column constraint; LP = light path; Psurf = surface pressure; SIF = solar-induced fluorescence; EV = emission verification; GHG = greenhouse gas; corr. = correction; constr. = constraint; obs. = observations; PS = point source.

Figure 3. Development of satellite instrument architectures for oxygen-band atmospheric remote sensing. Development of satellite instrument architectures for oxygen-band atmospheric remote sensing. Colors correspond to different instrument architectures: blue for multispectral imaging radiometers, green for wide-swath UV–VIS/NIR grating spectrometers, purple for high-resolution greenhouse gas spectrometers, and orange for multi-angle polarimeters/spectro-polarimeters. Solid lines with bars represent historical (ended) missions, solid lines with arrows represent ongoing missions, and dashed lines with arrows represent planned/future missions.

3.1. Multispectral Imaging Radiometry

Multispectral imaging radiometers equipped with O₂ absorption channels represent the earliest and most widely used satellite architecture for O₂-band cloud remote sensing. These instruments measure one or more discrete bands within the O₂ A- or B-band together with nearby continuum channels. Because they do not resolve individual absorption lines, their retrieval information is mainly contained in channel-integrated radiance or reflectance ratios. Early spaceborne demonstrations were provided by the Soviet Kosmos 149 and Kosmos 320 missions, launched in 1967 and 1970, respectively. Operational use was later established by MERIS on ENVISAT, launched in 2002, which used O₂ A-band measurements to support cloud-top pressure (CTP) and effective cloud fraction (ECF) retrievals [27]. This measurement concept was continued by the FY-3/MERSI series, including FY-3A, FY-3B, FY-3C, FY-3D/MERSI-II, and FY-3F/MERSI-III, launched in 2008, 2010, 2013, 2017, and 2023, respectively [28]. Sentinel-3/OLCI, with Sentinel-3A launched in 2016 and Sentinel-3B in 2018, further extended the MERIS heritage for ocean, land, and cloud applications. DSCOVR/EPIC, launched in 2015, provides a distinct L1 viewing geometry and uses O₂ A- and B-band filters for repeated full-disk observations of CTH/CTP [7,29].

The main strengths of this architecture are broad spatial coverage, relatively high spatial resolution, stable imaging capability, and efficient operational processing. Its main limitation is the limited information content of a small number of scalar channels. Cloud optical depth (COD), surface reflectance, aerosol loading, and viewing geometry can produce similar radiance-ratio changes. Therefore, O₂-band multispectral radiometry is most reliable for optically thick clouds or for scenarios in which surface and aerosol properties are constrained by auxiliary channels or prior information. Compared with high-spectral-resolution O₂-band spectrometers, it provides weaker vertical discrimination but much broader spatial sampling.

3.2. Wide-Coverage Dispersive Grating Spectrometry

Wide-swath grating spectrometers were developed primarily for ozone and trace gas retrievals, but their cloud and aerosol information is essential because trace gas vertical columns depend on AMFs [30]. Instruments such as GOME, SCIAMACHY, OMI, and GOME-2 established the use of O₂ A-band and/or O₄ absorption for effective cloud pressure and cloud fraction [31]. Sentinel-5P/TROPOMI launched in 2017 and PACE/OCI launched in 2024 extended this heritage with high spatial resolution and dedicated O₂ A- and B-band information, enabling operational ALH retrievals in addition to cloud products [32]. China has also developed dedicated polar-orbiting UV–VIS atmospheric composition spectrometers, including EMI on the GF-5 series first launched in 2018 and OMS on FY-3F launched in 2023, both of which provide O₄ bands in the range from 300 to 500 nm [33].

Geostationary air-quality missions further extend oxygen-band applications to high-frequency regional monitoring. GK-2B/GEMS uses UV-VIS O₄ information for the Asia-Pacific region [34]. TEMPO includes spectral coverage that supports O₄ and O₂ B-band constraints over North America [35]. MTG-S1/Sentinel-4 provides UV–VIS–NIR observations over Europe with O₂ A-band capability, enabling hourly trace gas and cloud/aerosol corrections.

3.3. Ultra-High-Resolution Greenhouse Gas Spectrometers

GHG satellite missions require high spectral resolution, stable radiometric calibration, and well-characterized instrument line shapes to retrieve XCO₂ and XCH₄ with sub-percent precision [36]. For GHG retrievals, O₂-band measurements are not ancillary cloud diagnostics but core constraints on dry-air column abundance, surface pressure, and scattering-induced photon-path changes [37]. GOSAT/TANSO-FTS and GOSAT-2/TANSO-FTS-2 employ Fourier Transform Spectroscopy to observe the O₂ A-band together with CO₂ and CH₄ absorption bands in the SWIR [38]. China’s GF-5/GMI adopted a spatial heterodyne spectroscopic approach to achieve high spectral resolution in the O₂ A-band for GHG retrieval support [39]. The dispersive grating-spectrometer architecture has been adopted by OCO-2, OCO-3, and China’s TanSat, FY-3D/GAS, and FY-3H/GAS-2 instruments [40,41].

Recent and next-generation GHG missions further shift this architecture toward wider coverage and stronger sensitivity to anthropogenic emissions. GOSAT-GW/TANSO-3, launched in 2025, replaces the earlier GOSAT Fourier Transform approach with a grating imaging spectrometer to improve spatial sampling and emission-source observations [42]. TanSat-2 scheduled in 2026 and the Copernicus CO2M mission planned for 2027 are designed to extend GHG monitoring toward wider-swath CO₂ and CH₄ mapping, with O₂-band measurements remaining essential for dry-air normalization, aerosol and cloud correction, and emission verification [39,42,43].

Although the O₂ A-band remains the most mature reference for reflected-solar greenhouse gas retrievals, its spectral separation from the main CO₂ and CH₄ bands near 1.6~2.3 µm limits the cancelation of wavelength-dependent aerosol, cloud, and surface effects. The 1.27 µm O₂ band is therefore increasingly used as a SWIR light-path reference because it is closer to the GHG absorption bands. Its application requires correction for upper-atmospheric dayglow or airglow and other additive radiance signals, as illustrated by missions such as MethaneSAT launched in 2024 and MicroCarb launched in 2025 [44].

3.4. Multi-Angle Polarimetry and Spectro-Polarimetry

Multi-angle polarimetry provides information that is difficult to obtain from scalar radiance or oxygen absorption alone. Polarized measurements are sensitive to aerosol size distribution, refractive index, particle shape, cloud droplet properties, and surface bidirectional reflectance, whereas O₂ absorption provides complementary constraints on scattering altitude and photon path length [17,45]. The POLDER series provided an early demonstration of this multi-angle polarization concept, with instruments launched on ADEOS in 1996, ADEOS-II in 2002, and PARASOL in 2004, with observations continuing until 2013. Later instruments further developed this approach, including the DPC on the GF-5 series first launched in 2018 and the MetOp-SG-A1/3MI launched in 2025 [13,46,47].

Recent developments are moving from discrete multi-angle polarimetric imaging toward multi-angle spectro-polarimetric observations [48]. PACE/SPEXone launched in 2024 provides multi-angle spectro-polarimetric measurements across the VIS–NIR range, including the O₂ A-band and B-band spectral regions [48,49]. It is not a dedicated high-resolution O₂-band spectrometer and does not provide independent polarization measurements in every spectral channel. By contrast, the cloud and aerosol monitoring instrument planned for TanSat-2 is designed to achieve full-spectrum spectro-polarimetry across the VIS–NIR range with polarization measurements in all spectral channels, providing stronger constraints on aerosol and cloud scattering properties [50]. In synergy with TanSat-2’s greenhouse gas spectrometer, this all-channel polarization capability can improve aerosol and cloud characterization, reduce photon-path correction uncertainties, and support CO₂ and CH₄ monitoring [51]. Despite the absence of oxygen absorption channels, the latest PACE/HARP2 and FY-3G/PMAI missions can still provide valuable information on cloud and aerosol properties from polarized multi-angle observations.

4. Retrieval Algorithms for Atmospheric Parameters

4.1. Cloud Parameter Retrievals

The cloud retrieval algorithms based on O₂-band absorption can be classified into cloud-as-reflecting-boundary (CRB) models, cloud-as-layer (CAL) models, DOAS-based techniques, Bayesian Optimal Estimation (OE) frameworks, and machine learning architectures (Table 3). The retrieved parameters include ECF, CTP, CTH, cloud optical centroid pressure (OCP), and cloud optical depth (COD) [52].

Table 3. Cloud retrieval algorithms using O₂ and O₄ absorption measurements.

Algorithms	Input Measurements	Key Physical Assumptions	Retrieved Parameters	Limitations
Cloud-as-reflecting-boundary models	• O2 A-band or O2 B-band • Continuum reflectance • Surface albedo • Viewing geometry • Auxiliary meteorology	• Cloud represented as an opaque Lambertian reflecting boundary • Effective cloud fraction mixing with clear-sky radiance • Cloud pressure controls the O2 photon path length • No explicit vertical cloud structure	ECF, CTP, CTH	• Bias for broken or optically thin cloud • Sensitive to surface albedo and 3D cloud effects • No explicit in-cloud photon penetration
Cloud-as-layer models	• O2 A-band and/or O2 B-band • Continuum reflectance • Surface albedo • Viewing geometry • Auxiliary meteorology	• Cloud represented as a finite scattering/absorbing layer • Vertical cloud extent and multiple scattering are included • Cloud microphysics and phase are often parameterized	ECF, CTP, CTH, COD	• Degeneracy among cloud height, optical depth, and phase • Limited sensitivity to multilayer clouds • Dependent on radiative-transfer and microphysical assumptions
O4 DOAS-based cloud retrieval	• O4 absorption, especially near 477 nm • Continuum reflectance • Surface albedo • Viewing geometry • Auxiliary meteorology	• O4 slant column constrains photon path length • DOAS isolates broadband scattering effects • Sensitive to cloud/aerosol shielding and enhancement	ECF, CTP, AMF-correction	• Strong aerosol interference • Dependence on O4 spectroscopy and surface reflectance • Less direct sensitivity to geometric cloud top
Optimal-estimation retrievals	• O2 A-band • O2 B-band • O4 absorption • Continuum reflectance • Surface albedo • Viewing geometry • Auxiliary meteorology	• Forward radiative-transfer model with Bayesian inversion • State vector, prior constraints, and error covariance are specified • Cloud and nuisance parameters are retrieved jointly • Posterior uncertainty is propagated	ECF, CTP, CTH, COD, cloud radius	• Computationally expensive • May be ill-posed or sensitive to priors • Requires accurate spectroscopy, calibration, surface BRDF, and cloud/aerosol assumptions
Machine learning or hybrid algorithms	• Any combination of O2 A/B-band and O4 absorption • Continuum reflectance • Surface albedo • Viewing geometry • Auxiliary meteorology • Simulation or collocation training data	• Statistical mapping learned from simulations or collocations • Hybrid schemes may embed radiative-transfer constraints • Physical post-processing or quality control may be applied	ECF, CTP, CTH, COD, cloud radius	• Training-domain dependence • Reduced interpretability • Risk of bias under unseen conditions • Uncertainty characterization can be difficult

4.1.1. Cloud-as-Reflecting-Boundary Models

CRB models approximate a cloudy pixel as a linear mixture of clear-sky and fully cloudy sub-pixels. The Fast Retrieval Scheme for Clouds from the Oxygen A-band (FRESCO) is a representative CRB algorithm developed for GOME, SCIAMACHY, GOME-2, and TROPOMI [53,54]. The cloud is treated as an opaque Lambertian reflector with cloud albedo A_c, usually fixed near 0.8 in traditional FRESCO-type algorithms.

The baseline forward model constructs the simulated TOA reflectance $R_{\mathit{sim}}$ as a linear combination of the clear-sky reflectance (R_s) and the fully cloudy reflectance ($R_c$):

$$R_{\mathit{sim}}\left(\mathsf{\lambda }\right) = f_{c }{R}_c(\mathsf{\lambda },{ A}_c,{ P}_c) + (1-f_c) R_s(\mathsf{\lambda },{ A}_s,{ P}_s),$$

(5)

where f_c is the effective cloud fraction, ${ A}_s$ is surface albedo, ${ P}_s$ is surface pressure, and ${ P}_c$ is the pressure of the equivalent cloud boundary. At continuum wavelengths for which oxygen absorption is weak, the effective cloud fraction can be estimated as

$$f_c={\displaystyle \frac{R_{\mathit{meas}}-{ R}_s}{R_c-R_s}}, 0\le f_c\le 1,$$

(6)

where $R_{\mathit{meas}}$ is the measured reflectance. This formulation is physically clearer than expressions that subtract surface albedo directly from measured reflectance. The clear and cloudy reflectance need to be calculated consistently by the radiative transfer model.

FRESCO and FRESCO+ are representative CRB algorithms. FRESCO retrieves ECF and CTP from selected O₂ A-band windows. FRESCO+ improves the physical treatment by including single Rayleigh scattering and better represents partly cloudy scenes. FRESCO-B applies the same retrieval concept to the O₂ B-band. Because the O₂ B-band is weaker than the O₂ A-band, photons can penetrate deeper into the cloud layer, and the retrieved cloud pressure may therefore be higher than that derived from A-band algorithms, particularly for high, optically thin clouds and vegetated surfaces [53,55]. Optimized by KNMI for Sentinel-5P, FRESCO-S is the default auxiliary cloud algorithm for many standard TROPOMI trace gas products. To compare TROPOMI (CTP) retrievals with MODIS observations, representative single-day cases for four seasons in 2023 (15 January, 15 April, 15 July, and 15 October 2023) were selected. TROPOMI L2 FRESCO-S (v2.4) swath pixels were spatially binned onto a 1.0° × 1.0° grid matching the MODIS L3 COSP product (MCD06COSP, C6.1) resolution, with fill values and unphysical entries 1.0° × 1.0° removed. Qualitative visual comparison demonstrates consistent spatial distribution patterns between the two products across all seasons, successfully capturing low CTP values in tropical deep convective zones and higher CTP values in marine stratocumulus regions. Localized regional discrepancies are primarily attributed to retrieval algorithm differences between O₂ A-band photon penetration depth and thermal infrared emission calculations. Figure 4 shows the cloud-top pressure distributions from TROPOMI and MODIS.

Figure 4. Seasonal spatial distributions of cloud top pressure (CTP) at 1° × 1° grid resolution derived from TROPOMI L2 FRESCO-S (v2.4) and the MODIS L3 COSP product (MCD06COSP, C6.1) for 15 January, 15 April, 15 July, and 15 October 2023. White cells indicate no valid retrieval after filtering.

CRB products are robust and computationally efficient, but their cloud pressure should not be interpreted as the true cloud top in all scenes. It is an equivalent pressure optimized for trace-gas cloud correction and depends on COD, vertical structure, and horizontal heterogeneity of clouds.

4.1.2. Cloud-as-Layer Models

CAL models describe clouds as vertically extended scattering layers rather than as opaque Lambertian reflecting boundaries. They are therefore more physically consistent for clouds with finite optical depth and appreciable in-cloud photon penetration.

In operational TROPOMI and GOME-2 processing, a synergistic two-step framework consisting of the Optical Cloud Recognition Algorithm (OCRA) and the Retrieval of Cloud Information using Neural Networks (ROCINN) is the best-known example [56,57]. OCRA first estimates radiometric cloud fraction from broadband UV–VIS color information relative to a cloud-free background. A schematic form is

$$f_c=\min \left\{1,\sqrt{{\sum }_{i\text{{}R,G,B\text{}}}\mathsf{\alpha }({\mathsf{\lambda }}_i)\max {\left[0, \rho \left({\mathsf{\lambda }}_i\right)-{\rho }_{\mathit{CF}}\left({\mathsf{\lambda }}_i\right)-\mathsf{\beta }\left({\mathsf{\lambda }}_i\right)\right]}^2}\right\},$$

(7)

where $\mathsf{\rho }\left({\mathsf{\lambda }}_i\right)$ is the measured TOA reflectance, and ${\mathsf{\rho }}_{\mathit{CF}}\left({\mathsf{\lambda }}_i\right)$ represents the cloud-free composite reflectance for the specific geolocation. The offset $\mathsf{\beta }\left({\mathsf{\lambda }}_i\right)$ accounts for residual clear-sky offsets and the outer maximum operator truncates negative values at zero, while the scaling factor $\mathsf{\alpha }\left({\mathsf{\lambda }}_i\right)$ maps the color-space distance to radiometric cloud fraction. The exact operational implementation is sensor- and version-dependent, so Equation (7) should be regarded as a schematic representation of the color-space principle. ROCINN then retrieves CTP or CTH and optical properties from O₂-band radiances using pre-calculated or neural-network-emulated radiative-transfer calculations.

The Semi-Analytical Cloud Retrieval Algorithm (SACURA) is another CAL-type approach. It uses asymptotic radiative-transfer theory to derive semi-analytical relationships for optically thick clouds, retrieving COD from visible reflectance and CTH from O₂ A-band absorption [58,59]. Furthermore, multiple scattering within optically thick clouds enhances in-cloud photon paths, so spectrally resolved O₂ absorption contains information on the effective path through and, in some configurations, below the cloud. SACURA uses this sensitivity primarily for optically thick CTH, whereas dedicated high-resolution implementations for OCO-2, GCOM-C/SGLI, and Sentinel-3/OLCI combine continuum and differential absorption information to estimate cloud-base height and geometrical thickness for single-layer liquid clouds. These quantities remain model-dependent and require assumptions about cloud phase, optical thickness, microphysics, and horizontal homogeneity; sensitivity is substantially weaker for optically thin, multilayer, broken, or ice-cloud scenes [60,61,62,63,64]. Its strength is physical transparency and computational efficiency for thick cloud layers, and its limitations arise when clouds are optically thin, vertically complex, or horizontally heterogeneous.

When comparing error characteristics for semi-transparent thin clouds and high-level cirrus, CRB models tend to substantially overestimate cloud-top pressure (underestimating cloud height) because photons penetrate through the cloud down to lower atmospheric layers or the surface. In contrast, CAL models better represent in-cloud multiple scattering and finite optical thickness, providing more realistic scattering heights for thin cirrus, though they remain subject to degeneracies between cloud optical thickness and layer altitude under low signal-to-noise or faint scattering conditions.

4.1.3. DOAS-Based Algorithms

UV–VIS spectrometers without O₂ A-band coverage often retrieve cloud information from O₄ absorption near 477 nm [65]. In a DOAS fit, the measured optical depth is described by differential absorption structures and a low-order polynomial for broadband extinction, surface reflection, and calibration effects [66]:

$$\mathrm{l}\mathrm{n}\left[{\displaystyle \frac{I_0\left(\lambda \right)}{I\left(\lambda \right)}}\right]={\sum }_j{\sigma }_j\left(\lambda \right)S_j+P_m\left(\lambda \right),$$

(8)

where $I_0\left(\lambda \right)$ is the reference solar spectrum, $I\left(\lambda \right)$ is the measured radiance, ${\sigma }_j\left(\lambda \right)$ is the differential absorption cross-section, $S_j$ is the slant column density, and $P_m\left(\lambda \right)$ is a low-order polynomial. Additional terms may be included for Ring-effect correction, wavelength shift, and spectral undersampling.

The fitted O₄ slant column ($S_{O_4}$) is compared against radiative-transfer lookup tables (LUTs) to retrieve ECF ($f_c$) and CTP (${ P}_c$) by minimizing the cost function:

$${\chi }^2={\sum }_j{\displaystyle \frac{{\left[S_{O_4,j}^{meas}-S_{O_4,j}^{sim}(f_c,P_c,A_s,\mathsf{\Theta })\right]}^2}{{\sigma }_{O_4,j}^2}},$$

(9)

where the summation index j denotes different observing geometries or spectral points, and ${\sigma }_{O_4,j}^2$ is the corresponding measurement error variance. This approach (e.g., OMCLDO2) is highly compatible with UV–VIS spectrometers lacking NIR channels, though it yields an optical centroid pressure rather than a physical CTH.

4.1.4. Optimal Estimation Algorithms

Bayesian OE provides a general framework for retrieving cloud parameters from O₂ absorption measurements. These methods combine spectral measurements, forward-model simulations, measurement-error statistics, and prior constraints within a unified probabilistic inversion [67]. The retrieved state vector may include ECF, CTP/CTH, COD, CGT, particle effective radius, surface albedo, and other auxiliary parameters. The standard optimal estimation cost function is:

$$J\left(\mathbf{x}\right)={\left[\mathbf{y}-\mathbf{F}\left(\mathbf{x}\right)\right]}^T{\mathbf{S}}_{ϵ}^{-1}\left[\mathbf{y}-\mathbf{F}\left(\mathbf{x}\right)\right]+{\left(\mathbf{x}-{\mathbf{x}}_a\right)}^T{\mathbf{S}}_a^{-1}\left(\mathbf{x}-{\mathbf{x}}_a\right),$$

(10)

where $\mathbf{y}$ is the measured spectrum, $\mathbf{F}\left(\mathbf{x}\right)$ is the forward model, $\mathbf{x}$ is the retrieval state vector, ${\mathbf{x}}_a$ is the a priori state, and $\mathbf{S}\mathsf{ϵ}$ and ${\mathrm{S}}_a$ are the measurement and a priori covariance matrices, respectively. This formulation enables quantitative characterization of retrieval uncertainty, averaging kernels, and degrees of freedom for signal, particularly valuable for advanced cloud retrievals in which the inverse problem is ill-posed [68].

4.1.5. Machine Learning and Hybrid Algorithms

Machine learning algorithms can be used either as forward-model emulators, where they rapidly predict radiances and Jacobians for a given atmospheric state, or as direct inverse models that retrieve cloud parameters from measured spectra [69]. A representative example is FRESCO-N, which uses artificial neural networks trained with radiative transfer simulations in the O₂ A- and B-bands. Similarly, neural-network acceleration has been adopted in ROCINN-type retrievals to avoid online line-by-line radiative-transfer calculations while retaining the physical information contained in O₂ absorption structures [70].

4.2. Aerosol Vertical Parameter Retrievals

Compared with cloud retrieval, ALH retrieval is more ill-conditioned because aerosol optical depths (AODs) are usually smaller, aerosol optical properties vary strongly with particle type, and surface reflectance can be comparable to or larger than the aerosol signal [71]. As illustrated in Figure 5, O₂ absorption provides height sensitivity through the shielding effect, in which elevated aerosol layers scatter part of the incoming radiation back to space before it samples the full lower-atmospheric O₂ column. The weakened apparent O₂ absorption therefore provides a constraint on the effective altitude of aerosol scattering.

Figure 5. Schematic diagram of the physical principle of the remote sensing of aerosol layer height in the O₂ absorption bands. Due to the scattering by aerosol particles, light scattered by a higher aerosol layer (b) would travel a shorter path length to reach the satellite sensor than those scattered by the lower-altitude aerosol (a), resulting in less absorption by O₂ and thus a higher radiance value detected by the satellite. (Source: [72]).

4.2.1. Radiance Ratio and Lookup Table Algorithms

Early ALH retrievals from oxygen absorption measurements mainly used radiance ratio diagnostics combined with pre-computed radiative-transfer LUTs. For multispectral imaging radiometers such as MERIS, OLCI, and DSCOVR/EPIC [73], the basic observable is the contrast between an O₂-absorbing channel and a nearby weakly absorbing or continuum channel:

$$R({\mathsf{\lambda }}_{\mathit{in}},{ \mathsf{\lambda }}_{\mathit{out}}) = {\displaystyle \frac{I({\mathsf{\lambda }}_{\mathit{in}})}{I({\mathsf{\lambda }}_{\mathit{out}})}},$$

(11)

where $I\left({\mathsf{\lambda }}_{\mathit{in}}\right)$ is the TOA radiance at an O₂-absorbing wavelength (e.g., 760.8 nm in the O₂ A-band or 687 nm in the O₂ B-band), and $I\left({\mathsf{\lambda }}_{\mathit{out}}\right)$ is the radiance in a nearby weakly absorbing continuum channel. This approach is computationally efficient and suitable for operational imaging instruments, but $R({\mathsf{\lambda }}_{\mathit{in}},{ \mathsf{\lambda }}_{\mathit{out}})$ is not uniquely sensitive to ALH. Reliable retrievals require constraints on AOD and aerosol optical properties, and performance degrades for low AOD, bright surfaces, unresolved clouds, and vertically extended plumes [74].

4.2.2. Optimal Estimation Algorithms

High-spectral-resolution O₂ A-band measurements can exploit the different altitude sensitivities of weak and strong absorption lines to retrieve ALH [75]. Full-physics OE retrievals fit the measured spectrum with a multiple-scattering radiative-transfer model [76]. The cost function used for this inversion has the same general form as that mentioned above for cloud-parameter retrievals in Equation (10). A compact aerosol extinction profile is often used to reduce the dimensionality of the inverse problem [77]. A commonly used Gaussian parameterization is:

$${\mathsf{\beta }}_{\mathit{ext}}\left(z\right)={\displaystyle \frac{{\mathsf{\tau }}_{\mathit{aer}}}{w\sqrt{2\pi }}}\mathit{exp}\left[-{\displaystyle \frac{(z-z_{\mathit{aer}}{)}^2}{2w^2}}\right],$$

(12)

where ${\mathsf{\tau }}_{\mathit{aer}}$ is AOD, $z_{\mathit{aer}}$ is the aerosol layer mid-height (representing the ALH), and $w$ represents the half-width or geometric thickness of the aerosol layer. The retrieval state vector typically includes $z_{\mathit{aer}}$, ${\mathsf{\tau }}_{\mathit{aer}}$, $w$, and surface-reflectance-related parameters. The TROPOMI operational ALH product is a representative O₂ A-band OE application. Its retrieval accuracy depends strongly on assumptions about aerosol type, single-scattering albedo, surface albedo, cloud screening, and the vertical shape of the aerosol layer [78].

4.2.3. Machine Learning and Hybrid Algorithms

Machine learning methods are increasingly used to support ALH retrieval from O₂-band measurements, particularly where high-resolution multiple-scattering radiative-transfer calculations are computationally expensive. On the one hand, neural-network or principal-component-based emulators are trained on radiative transfer simulations from models such as VLIDORT or SCIATRAN to provide fast radiance calculations and, in some cases, Jacobians for OE retrievals [79,80]. On the other hand, machine learning models, including random forests, gradient-boosting methods, convolutional neural networks, and multilayer perceptron, can directly exploit collocations with CALIOP, EarthCARE, or ground-based lidar to estimate ALH from satellite spectra and auxiliary variables. Figure 6 presents comparisons between the ATLID extinction-weighted ALH and the SPEXone L2 ALH retrievals for two representative collocated cases. Two cases were evaluated: an African dust case on 25 January 2026 (14.77–16.00°E, 1.00–7.46°N) and a Central Asian dust case on 22 September 2025 (68.00–69.00°E, 33.50–39.00°N). The EarthCARE ATLID 355 nm aerosol extinction profile product (ECA_EXBA_ATL_AER_2A) and the SPEXone L2 product (PACE_SPEXONE.L2.RTAP_LD.V3_0) were used for comparison. The ATLID-derived reference ALH was calculated as the extinction-weighted mean altitude using collocation criteria of no more than 5.0 km in spatial distance and 30 min in temporal difference. Because ATLID and SPEXone use different wavelengths, footprints, vertical metrics, and sensitivities, the comparison demonstrates collocated behavior but is not presented as an independent validation or accuracy estimate. Such methods can capture complex nonlinear relationships and heterogeneous aerosol conditions, but their reliability depends strongly on training-set representativeness and out-of-distribution robustness [32,81].

Figure 6. Intercomparison of vertical ALH between active-lidar extinction profiles (ATLID L2) and passive polarimetric retrievals (SPEXone L2) for selected dust transport events over Africa (25 January 2026) and Central Asia (22 September 2025).

4.3. Oxygen-Band-Assisted Greenhouse Gas Retrievals

Satellite retrievals of column-averaged greenhouse gas abundances, especially XCO₂ and XCH₄, require accurate separation of gas absorption from surface reflection and atmospheric scattering. Therefore, the O₂ A-band and, increasingly, the 1.27 µm O₂ band are used to constrain surface pressure or apparent scattering pressure, normalize target-gas columns by the dry-air column, and diagnose photon-path changes caused by aerosols and thin clouds [44,82]. These roles are incorporated into retrieval approaches of different complexities, from proxy-based normalization and photon path-length distribution methods to full-physics optimal-estimation frameworks.

4.3.1. Surface Pressure Retrieval Method

Under clear-sky conditions with known surface reflectance and atmospheric state, oxygen absorption can constrain surface pressure. A generic spectral fitting problem is

$$\mathit{y}=\mathit{F}\left[p_s,A_s\left(\lambda \right),{\mathit{x}}_{atm},{\mathit{x}}_{sca},\Theta \right]+\mathit{ϵ},$$

(13)

where $p_s$ is the target surface pressure,${ A}_s\left(\mathsf{\lambda }\right)$ represents spectral surface reflectance, ${ \mathbf{x}}_{\mathit{atm}}$ includes atmospheric temperature and humidity profiles, ${\mathbf{x}}_{\mathit{sca}}$ denotes cloud and aerosol scattering parameters, $\mathsf{\Theta }$ represents the solar-viewing geometry, and $\mathsf{ϵ}$ is measurement error. Under clear-sky conditions and well-characterized surface reflectance, it approaches the true surface pressure. In the presence of clouds, aerosols, or strong surface anisotropy, however, the measurement may instead constrain an effective scattering pressure or photon-path-weighted pressure.

4.3.2. Dry-Air Normalization Method

A major application of reflected-solar O₂ absorption is the normalization of trace gas columns by dry-air mass. Column-averaged greenhouse gas abundances are defined as the ratio between the retrieved gas column and the dry-air column. Using O₂ as a reference absorber, this relationship can be approximated as

$${\displaystyle \frac{N_{\mathit{gas}}}{N_{\mathit{dry}}}}\approx {\mathsf{\chi }}_{O_2}{\displaystyle \frac{N_{\mathit{gas}}}{N_{O_2}}},$$

(14)

where $N_{\mathit{gas}}$ denotes the retrieved column abundance of CO₂, CH₄, or another target gas. This normalization is only valid when the O₂ and greenhouse gas absorption bands sample a consistent atmospheric light path.

4.3.3. Proxy Method for GHG Retrievals

The proxy method assumes that the fractional change in the photon path length due to scattering is nearly identical for the target gas and a spectrally adjacent proxy gas. XCH₄ is often retrieved using CO₂ as a proxy, or XCO₂ is normalized directly using the retrieved O₂ column. The generic formulation is expressed as

$$X_{\mathit{target}}={\displaystyle \frac{V_{\mathit{target}}^{\mathit{retrieved}}}{V_{\mathit{proxy}}^{\mathit{retrieved}}}}\times X_{\mathit{proxy}}^{\mathit{model}},$$

(15)

where $V_{\mathit{target}}^{\mathit{retrieved}}$ and $V_{\mathit{proxy}}^{\mathit{retrieved}}$ are the vertical column densities retrieved under a non-scattering assumption, and $X_{\mathit{proxy}}^{\mathit{model}}$ is an a priori proxy-gas mole fraction. While the proxy method efficiently cancels out first-order scattering errors and instrument-related radiometric biases, its foundational assumption degrades when the spectral distance between the target and proxy bands is large. The significant wavelength dependence of aerosol scattering phase functions and surface albedos (e.g., between the 0.76 μm O₂ A-band and the 1.6 μm CO₂ band) inherently limits the precision of simple ratio-based corrections over complex heterogeneous scenes.

4.3.4. Photon Path Length Probability Density Function Method

To explicitly account for complex multiple scattering without deploying a full-physics state vector, the photon path length probability density function (PPDF) method was developed as an alternative operational algorithm for the GOSAT mission. The PPDF method parameterizes scattering effects statistically rather than retrieving explicit aerosol and cloud layers [83]. The equivalent transmittance $T_{\mathit{eq}}\left(\mathsf{\lambda }\right)$ is

$$T_{\mathit{eq}}(\mathsf{\lambda })={\int }_0^{\infty }P(l)\mathit{exp}(-{\mathsf{\tau }}_{\mathit{abs}}(\mathsf{\lambda })·l) \mathit{dl},$$

(16)

where $P\left(l\right)$ is the probability density of relative photon path length and ${\mathsf{\tau }}_{\mathit{abs}}$ is gas absorption optical depth along a reference path $l$. O₂ A-band spectra constrain PPDF parameters, which are then transferred to CO₂ or CH₄ absorption bands using physical scaling factors, effectively correcting the gas absorption lines for complex path length deformations.

4.3.5. Optimal Estimation GHG Retrieval Algorithms

To meet the extremely rigorous requirements of modern global carbon observatories (e.g., OCO-2/3 and TanSat missions), XCO₂ and XCH₄ retrievals generally use full-physics OE approaches, such as the NASA ACOS (Atmospheric CO₂ Observations from Space) algorithm [84]. O₂, CO₂, CH₄, and H₂O spectral windows are jointly fitted, allowing gas abundance, surface pressure, surface reflectance, aerosol and thin-cloud parameters, temperature, water vapor, and calibration terms to be estimated or constrained consistently. The O₂ A-band is especially valuable because its pressure-broadened line structure provides differential sensitivity to scattering altitude and dry-air mass. When combined with SWIR GHG bands, this information helps reduce the aliasing of aerosol and thin-cloud effects into retrieved gas columns [85]. Nevertheless, residual biases remain over bright surfaces, complex terrain, snow/ice, thin cirrus, and elevated aerosol plumes, where simple aerosol parameterizations cannot represent real scattering complexity. Figure 7 demonstrates the constraining effect of the oxygen absorption band on greenhouse gas retrievals.

Figure 7. Oxygen-band constraints in greenhouse gas retrievals.

5. Applications in Climate and Environmental Monitoring

O₂-band remote sensing provides a set of atmospheric constraints that are central to climate research and environmental monitoring. In the reflected-solar UV–VIS–SWIR spectral range, O₂ and O₄ absorption measurements are used to infer cloud pressure and height, effective cloud fraction, aerosol layer height, surface or apparent pressure, dry-air column abundance, and photon-path modifications caused by scattering [86]. These quantities are not independent retrieval targets, but are linked through the common sensitivity of oxygen absorption to atmospheric mass, scattering altitude, and effective light path [87]. As a result, O₂-band products support a wide range of applications, from cloud and aerosol climatology to air-quality assessment, GHG retrieval, carbon flux inversion, and multi-sensor atmospheric correction (Figure 8). To provide a consolidated and quantitative synthesis across these diverse applications, Table 4 summarizes the representative satellite missions, operational products, typical retrieval precisions, dominant uncertainty sources, and primary failure scenarios for oxygen-band observations.

Figure 8. Synergistic use of oxygen-band atmospheric products from multiple missions for climate, air-quality, and carbon-cycle applications.

Table 4. Quantitative performance, main error sources, and retrieval constraints across major oxygen-band products.

Application	Representative Satellite	Product and Algorithm	Typical Quantitative Accuracy	Uncertainty and Error Sources	Operational Limitations
Cloud Characterization	TROPOMI	CTP/FRESCO-S	Precision: ~20–50 hPa (~0.5–1.0 km in CTH) [53,55]	Surface albedo coupling, 3D cloud boundary effects, broken clouds	High bias over bright snow/ice surfaces; reduced sensitivity for thin cirrus
Aerosol Profile	TROPOMI	ALH/optimal estimation	Accuracy: ~0.5–1.0 km (vs. CALIOP/ATLID) [32,75]	Prescribed SSA, phase function, land albedo	Degrades under low aerosol loading (AOD < 0.2); dark water sensitivity
GHG Light-Path Correction	OCO-2/ OCO-3	XCO2 light-path and surface pressure/ACOS full-physics	Single-sounding precision: <1.0 ppm (~0.25%) [82,85]	Spectroscopy database errors, spectral optical property mismatch (O2 A-band vs. SWIR)	Residual biases over heavy aerosol plumes, thin cirrus, and complex terrain
SWIR Light-Path Reference	MicroCarb/ MethaneSAT	1.27 um O2 band light-path constraint	Surface pressure precision: ~2–4 hPa [44]	Mesospheric dayglow/airglow emission contamination	Sensitivity to airglow correction errors at high solar zenith angles

5.1. Cloud Climatology and Radiative Forcing Assessment

Accurate global characterization of CTH, CTP, ECF, and COD from O₂-band retrievals provides a critical basis for quantifying cloud radiative forcing, which remains one of the largest sources of uncertainty in climate sensitivity estimates. Multi-decadal cloud property records derived from GOME (1995~2011), SCIAMACHY (2002~2012), MERIS (2002~2012), OMI (2004~present), GOME-2 (2006~present), and TROPOMI (2017~present) provide long-term climatological baselines required to detect trends in cloud-top altitude and fractional coverage associated with large-scale atmospheric circulation changes [88]. The transition from polar-orbiting to geostationary platforms (e.g., GEMS, TEMPO, and Sentinel-4) further enables the resolution of diurnal cycles in cloud development and dissipation that are systematically aliased in sun-synchronous observations, providing more accurate estimates of the time-integrated cloud radiative effect.

The combination of O₂ A-band-derived CTH with simultaneous COD retrievals from visible channels enables the classification of cloud regimes (e.g., low marine stratocumulus, mid-level altocumulus, deep convective anvils) and the estimation of cloud liquid water path and ice water path. These cloud regime classifications are directly relevant to the parameterization of convective and boundary-layer cloud processes in general circulation models (GCMs), where persistent biases in simulated low-cloud fractions over subtropical oceans remain a leading source of inter-model spread in equilibrium climate sensitivity [89].

Beyond instantaneous cloud characterization, consistent O₂ A-band records have been used to assess multi-decadal CTH variability. Applying SACURA to GOME, SCI-AMACHY, and GOME-2 observations from June 1996 to May 2012, Lelli et al. reported a deseasonalized global (60°S–60°N) trend of −1.78 ± 2.14 m yr⁻¹ and an ENSO-corrected trend of −0.49 ± 2.22 m yr⁻¹; neither differs significantly from zero at the global scale, although regional tendencies occur. This example illustrates both the climate value of harmonized oxygen-band records and the need to control inter-sensor calibration, sampling, and algorithm consistency before interpreting long-term trends. Using physically consistent algorithms such as SACURA, multi-mission satellite series (e.g., GOME, SCI-AMACHY) have provided global evaluations of linear trends in cloud-top height, offering valuable observational constraints on cloud-altitude feedback mechanisms under a warming climate [90,91].

5.2. Aerosol Vertical Distribution and Air-Quality Monitoring

ALH retrievals from O₂ absorption provide observational constraints on the three-dimensional distribution of aerosol scattering that are not accessible from column-integrated AOD measurements alone. The vertical placement of aerosol layers critically determines their radiative forcing efficiency, their interaction with boundary-layer meteorology, and their impact on surface air quality [92]. Elevated biomass-burning smoke plumes, Saharan dust layers, and volcanic ash clouds are often transported thousands of kilometers from their source regions at altitudes of 2–8 km, and can produce radiative and chemical effects that differ substantially from those of near-surface aerosols with comparable column loading.

Operationally, TROPOMI ALH products have been systematically applied to improve the accuracy of tropospheric NO₂, SO₂, HCHO, and other tropospheric trace gas AMF calculations over heavily polluted regions in East Asia, South Asia, and sub-Saharan Africa, where elevated aerosol layers can cause AMF errors of 20~50% if treated as surface-layer scatterers. The synergistic combination of TROPOMI ALH with CALIOP vertical profiles has enabled the validation and refinement of aerosol transport model parameterizations, particularly for the representation of pyroconvective injection heights and long-range dust transport pathways. The emerging spectro-polarimetric capabilities of PACE/SPEXone and TanSat-2 promise to further advance aerosol characterization by simultaneously constraining ALH, AOD, single-scattering albedo, and particle-size distribution within a unified multi-angle polarimetric retrieval framework [93].

5.3. Greenhouse Gas Monitoring and Carbon Flux Inversion

The most societally consequential application of O₂ band remote sensing is the provision of corrected photon-path XCO₂ and XCH₄ retrievals for global carbon-cycle monitoring and anthropogenic emission quantification [94]. The O₂ A-band serves as the physical cornerstone of this application by simultaneously constraining the dry-air column, diagnosing the effective photon path length, and providing an independent surface pressure estimate.

The OCO-2 mission has demonstrated that O₂ A-band-anchored full-physics retrievals can achieve XCO₂ precisions of ~1 ppm (0.25%) on individual soundings, sufficient to resolve regional-scale carbon flux anomalies through atmospheric inversion modeling [95,96]. The GOSAT and GOSAT-2 time series spanning from 2009 to the present provide the longest continuous spaceborne record of column-averaged GHG abundances, enabling the detection of long-term trends in XCO₂ growth rates and interannual variability driven by El Niño-Southern Oscillation (ENSO)-related ecosystem responses. The new-generation GHG missions (e.g., GOSAT-GW, CO2M, and TanSat-2) are designed to detect and quantify anthropogenic CO₂ emissions from individual power plants, industrial facilities, and urban areas at the city scale through wide-swath imaging and dense spatial sampling [97].

5.4. Multi-Mission and Multi-Modal Integration

The combination of high-accuracy, high-spatial-resolution LEO retrievals (e.g., TROPOMI, OCI) with high-temporal-resolution GEO observations (e.g., GEMS, TEMPO, Sentinel-4) enables the construction of spatiotemporally complete atmospheric state descriptions that are individually inaccessible from either platform alone [98]. Optimal interpolation and data fusion frameworks that exploit the complementary sampling characteristics of LEO and GEO O₂ band products are an active area of methodological development [99].

The combination of passive O₂ band retrievals with active-lidar measurements from CALIPSO/CALIOP, EarthCARE/ATLID, and future spaceborne lidar missions provides mutually constraining information on aerosol and cloud vertical structure. While lidar provides direct, high-vertical-resolution extinction profiles along a narrow curtain track, O₂ band retrievals provide spatially continuous but vertically integrated constraints that can be used to extend and validate the lidar-derived vertical distributions across broader spatial domains [100].

The integration of continuous O₂ A-band spectroscopy with multi-angle polarimetric measurements, as implemented in PACE/SPEXone and planned for TanSat-2, maximizes the information content available for simultaneous retrieval of aerosol microphysics and vertical distribution. Polarimetric channels constrain particle size, particle shape, and refractive index, while spectrally resolved O₂ absorption constrains scattering altitude [101]. Together, these complementary measurements help reduce the degeneracy that limits retrievals based solely on scalar intensity or individual spectral features.

6. Challenges and Future Directions

6.1. Outstanding Challenges

Despite substantial advances, several fundamental challenges continue to limit the accuracy and utility of O₂ band retrievals:

• Residual errors in O₂ line parameters, including line positions, intensities, pressure-broadening coefficients, and non-Voigt line shape parameters, propagate directly into systematic biases in surface pressure and photon-path retrievals. The O₂ A-band spectroscopy encoded in current databases (HITRAN, GEISA) has been substantially improved through laboratory measurements and satellite-constrained fitting, but sub-percent-level discrepancies between different spectroscopic datasets remain a leading source of inter-mission bias in XCO₂ retrievals [38]. Continued investment in high-accuracy laboratory spectroscopy and in-orbit spectroscopic calibration is essential for achieving the long-term stability required for climate trend detection.
• Full-physics OE retrievals of ALH and GHG columns remain sensitive to prescribed aerosol optical properties, including single-scattering albedo, phase function, and spectral refractive index, because these parameters cannot be independently constrained from O₂-band measurements alone. The spectro-polarimetric architectures of PACE/SPEXone and TanSat-2 represent the most promising near-term approach to resolving this degeneracy, but the development of retrieval algorithms that fully exploit the joint information content of spectral absorption and polarimetric scattering remains an active research frontier [93,102].
• Partially cloudy scenes remain a major source of systematic error in both cloud property and GHG column retrievals, particularly when optically thin clouds, broken cloud fields, and cloud shadows occur within the same satellite footprint. The independent pixel approximation (IPA) employed in most operational algorithms neglects horizontal photon transport across cloud boundaries, introducing biases that scale with cloud fraction gradients and are particularly severe at coarse spatial resolutions [103]. Higher spatial resolution observations and three-dimensional radiative-transfer treatments are needed to address this limitation.
• The fundamental assumption underlying O₂ A-band photon-path correction for GHG retrievals is that the effective photon path sampled by the O₂ A-band near 760 nm is representative of the path sampled by the target GHG bands near 1.6 or 2.0 µm. This assumption can break down when aerosol scattering and surface reflectance vary strongly with wavelength [104]. The adoption of the 1.27 µm O₂ band in MicroCarb and MethaneSAT partially addresses this limitation, but a comprehensive solution requires the simultaneous characterization of aerosol spectral optical properties across the full NIR–SWIR range [105].

6.2. Future Outlook

The next decade is likely to bring substantial advances in O₂-band atmospheric remote sensing, driven by improvements in satellite instrumentation, retrieval methodology, and multi-mission integration. These developments will expand the role of O₂ absorption measurements from cloud and aerosol correction toward a broader framework for constraining atmospheric mass, photon path length, and scattering altitude in climate, air-quality, and greenhouse gas applications.

• The transition from narrow-swath to wide-swath high-resolution spectrometers (e.g., CO2M and TanSat-2) will dramatically increase the spatial density of O₂ A-band soundings, enabling the detection and attribution of anthropogenic GHG emissions at the facility and city scale [106,107]. Maintaining instrument line shape uniformity and radiometric stability across wide fields of view represents the primary technological challenge for this generation of instruments.
• The coordinated operation of GEMS over the Asia-Pacific region, TEMPO over North America, and Sentinel-4 over Europe will establish a geostationary air-quality observational system for hourly monitoring of tropospheric composition [108]. By using O₂ or O₄ band measurements to constrain cloud and aerosol effects in air mass factor calculations, these missions can improve the consistency of regional air-quality retrievals. To use these three independently developed missions together, their cloud and aerosol products need to be made consistent in retrieval assumptions, validation standards, and uncertainty characterization.
• The integration of machine learning architectures with physical radiative-transfer constraints promises to simultaneously improve retrieval accuracy, computational efficiency, and uncertainty quantification [109]. These approaches are particularly well-suited to the high-dimensional, nonlinear, inverse problems posed by spectro-polarimetric O₂ band retrievals, where the information content of the measurement vector substantially exceeds what can be exploited by current analytical OE frameworks.

In conclusion, O₂ absorption-band remote sensing has developed from an early method for CTH estimation into an important component of passive atmospheric remote sensing in the reflected-solar spectrum. Its value lies in the ability to constrain atmospheric mass, photon path length, and scattering altitude, which are key to cloud, aerosol, trace gas, and GHG retrievals [110]. Continued improvements in spectroscopy, radiative-transfer modeling, retrieval algorithms, and satellite instrumentation, together with closer integration of lidar, passive imaging spectroscopy, and polarimetric observations, will further strengthen the role of O₂-band measurements in climate and environmental monitoring.

7. Conclusions

Satellite observations in oxygen absorption bands (O₂ A-band, O₂ B-band, 1.27 um band, and O₄ collision-induced bands) provide fundamental, physically grounded constraints on atmospheric mass, pressure, and photon path lengths. Over recent decades, oxygen-band remote sensing has evolved from a basic cloud-height estimation method into a critical, multi-disciplinary tool supporting cloud climatology, aerosol vertical profile detection, air-quality AMF corrections, and high-precision greenhouse gas (XCO₂/XCH₄) retrievals.

A central message of this review is that passive oxygen-band parameters (such as cloud-top pressure or aerosol layer height) are essentially effective radiative metrics rather than pure geometrical heights. The effective optical centroid retrieved by passive spectrometers depends heavily on optical depth, vertical extinction profiles, and in-cloud multiple scattering. Recognizing this physical distinction is crucial for proper data assimilation, active–passive cross-validation, and climate model applications.

Despite remarkable progress, systematic uncertainties persist. Spectroscopic parameter errors, line-shape modeling approximations (e.g., non-Voigt line shapes and Rotational Raman Scattering filling-in), surface BRDF coupling, 3D radiative-transfer effects in broken clouds, and wavelength-dependent path mismatches between NIR oxygen bands and SWIR GHG bands remain leading sources of retrieval bias.

Future advances will require improved O₂ and O₄ spectroscopy, more realistic radiative-transfer modeling, better instrument characterization, and stronger active–passive validation. Overall, O₂ and O₄ absorption measurements should be viewed not as auxiliary correction channels, but as a unifying framework for constraining atmospheric mass, photon transport, and scattering altitude. As satellite observation systems move toward wider swaths, higher spatial resolution, geostationary sampling, and multi-angle polarimetric capability, oxygen-band remote sensing will remain central to improving the physical consistency and scientific value of atmospheric satellite retrievals.