The marine environment is highly complex, and conventional in situ measurements carried out, e.g., on research cruises, can only provide a series of snapshots. Therefore, uncovering the real dynamics requires complementation with data of higher spatiotemporal coverage, which could be provided by autonomous measurements (Zielinski et al., 2009; Chai et al., 2020). The Argo network of autonomous floats¹ has revolutionized the capability for operational observations (Roemmich, 2001) to be the main source of CTD measurements (Riser et al., 2016). Biogeochemical Argo floats (BGC-Argo floats) (Biogeochemical-Argo Planning Group, 2016), the extension of the Argo core mission, supply the optical and biogeochemical observations necessary to construct a global, full-depth, and multidisciplinary ocean-observing system (IOCCG, 2011; Bittig et al., 2019; Claustre et al., 2020). This type of float accommodates up to six essential biogeochemical and bio-optical variables (Figure 1): oxygen (O₂), pH, nitrate (NO₃), suspended particles, chlorophyll-a (chl-a), and radiometry.

The basic radiometric measurement is the spectral radiance, L(λ), representing the radiant flux from a defined direction through a defined area. From radiance, all other integral measures of the light field can be derived, such as irradiance E(λ),which is the total radiant flux per unit area of horizontal surface measured in the direction of the incoming light. The downwelling irradiance E_d(λ), thus the irradiance measured toward the sea surface, is the parameter that is regularly measured by the BGC-Argo floats (Organelli et al., 2016). However, some initiatives have started to look also at the upwelling radiance L_u (Gerbi et al., 2016; Leymarie et al., 2018; Wojtasiewicz et al., 2018). The propagation of E_d(λ) in the water column can be described approximately as an exponential function of depth (Mobley, 1994):

In case a layer-averaged K_d (from 0 to z) is used, light decline is calculated according to:

There, z [m] is water depth, E_d(λ,0⁻)[Wm⁻²sr⁻¹nm⁻¹] the irradiance right below the surface, and K_d(λ,z)[m⁻¹] the diffuse attenuation coefficient of downwelling irradiance, which can be deduced from measurements of E_d in different depths. Both E_d and K_d are spectral quantities, indicated by λ[nm] for wavelength.

The slope of this exponential decline, thus K_d(λ), depends on the absorption (a)and backscattering properties (b_b) of the water (Lee et al., 2005), i.e., termed the inherent optical properties (IOPs), which are in turn determined by the water molecules themselves as well as the properties and concentration optically active substances (OAS) within the water column, such as phytoplankton, detrital matter, and colored dissolved organic matter (CDOM; see also Kirk, 2011). From E_d(λ) and L_u(λ), other quantities, commonly known as apparent optical properties (AOPs), can be derived. In addition to K_d(λ), a frequently derived AOP is radiance reflectance, r_rs(λ), (or remote sensing reflectance, R_rs(λ), when measured just above the sea surface). It is defined as the ratio of L_u(λ)/E_d(λ) (see Morel and Gentili, 1996). A multiplication with the extraterrestrial solar spectral irradiance, gives another important parameter, the normalized water-leaving radiance, L_wn(λ) (see Gordon and Clark, 1981). K_d(λ), r_rs(λ), and R_rs(λ) are also the most commonly studied parameters of interest in remote sensing applications, notably satellite observation (Austin and Petzold, 1981; Werdell and Bailey, 2005). However, converting a radiometric signal captured by a spaceborne sensor to an end-user product carries uncertainties. High demands for accuracy in absolute calibration of satellite remote sensing data (IOCCG, 2013) and high quality field observations are required for satellite product validation/evaluation (Mueller and Austin, 1992; Zibordi and Voss, 2014). For both optical in situ and remote sensing observation, various bio-optical algorithms have been developed to relate features of the water signal to in-water quantities of interest (e.g., O’Reilly et al., 1998; Maritorena et al., 2002). Improving these retrieval algorithms (existing) and develop enhanced ones that exploit forthcoming technologies remains a priority for ocean optics community.

Ocean optics research has recently undergone a shift from multispectral (with broadband data) to hyperspectral instrumentation, both in situ and remotely (Chang et al., 2004). The higher number of available bands in hyperspectral observations allows a more detailed examination regarding spectral resolution of AOPs and IOPs within the water column. This makes applications requiring higher resolution feasible, e.g., the discrimination of phytoplankton groups, the determination of particle size, and more detailed determination of CDOM spectral properties. Conventional Argo floats already provide sustained high-quality data at unprecedented temporal and spatial resolution. Thus, the addition of hyperspectral observation capabilities will allow Argo platforms to meet demands regarding more sophisticated approaches from the field of optical models and remote sensing. The exploitation of such high-resolution data (in the spatial, temporal, and spectral dimension), measured by a single system that carries also IOP sensors, will certainly open new avenues for bio-optical research in the future. This exciting combination (Argo platform + hyperspectral radiometers) could be the next logical step for Argo floats network (Claustre et al., 2020). With this in mind, this review summarizes the current state-of-the-art of radiometric measurements on Argo floats, and identifies emerging future pathways for the development of hyperspectral Argo radiometry and data exploitation. In our search, we referred to the list of the BGC Argo mission bibliography,² from which we selected studies involving radiometric measurements (including papers describing data management approaches and quality-control procedures). We made additional research using Google Scholar and Scopus for verification.

Argo radiometric measurements are currently limited to observations measuring, depending on configuration, either irradiance or radiance at up to four channels. Acquisition of radiometric observations takes place in discrete spectral bands, selected to detect major variations in underwater optical properties, allowing these multispectral devices to be effective in deriving the relative concentration of in-water constituents over the water column. Taking the example of phytoplankton, the retrieval of more details regarding species composition (which is oftentimes called for) has been challenging because of the insufficient spectral resolution of the multispectral sensor (Bracher et al., 2017). Hyperspectral radiometers successfully achieved this discrimination (e.g., Isada et al., 2015; Xi et al., 2015), as well as the retrieval of phytoplankton functional types (Wolanin et al., 2016), as they provide measurements at hundreds of contiguous wavebands distributed equally from the ultraviolet to the near-infrared segment of the spectrum with a resolution of the order of 5 nm. This has led to more frequent use of hyperspectral instrumentation in ocean optics research (Chang et al., 2004). Bringing hyperspectral devices into play on robotic platforms such as Argo floats is therefore the next logical step (Claustre et al., 2020). In some, especially remote areas affected by regularly occurring (harmful) algal blooms, monitoring strategies require increased observing capabilities (Sellner et al., 2003). Whilst detection and characterization of the spatial and temporal development of phytoplankton biomass is already feasible using existing multispectral Argo floats, the transformation toward hyperspectral technology will provide additional opportunities of tracking its composition and predict harmful algal bloom development (Shen et al., 2012; Dierssen et al., 2020). Ultimately, hyperspectral data based on Argo float may pave the way toward harmful algal bloom forecasting programs that require nearly real time observations from a global communication network (Jochens et al., 2010).

Hyperspectral data is further advantageous for calibration and validation purposes (McClain and Meister, 2012) of current and future hyperspectral ocean color missions such as the Environmental Mapping and Analysis Program (EnMAP) (Guanter et al., 2015), the Hyperspectral Precursor of the Application Mission (PRISMA) (Loizzo et al., 2018), and the Pre-Aerosol, Clouds, and ocean Ecosystem (PACE) mission (Gregg and Rousseaux, 2017). Indeed, hyperspectral radiometric observations measured by Argo will provide high-resolution data (in the context of spatial, temporal and spectral resolution) that will be more appropriate to match the spectral bands of these advanced sensors (Werdell et al., 2013; Brewin et al., 2016). This would subsequently improve the assessment of derived products and thereby upgrade the validation metrics (Maritorena and Siegel, 2005; Maritorena et al., 2010; Seegers et al., 2018). A particular benefits of such datasets would be that they offer the possibility to develop new algorithms with more freedom in selecting spectral bands for bio-optical studies. Such developments would in turn improve both empirical remote-sensing (Gordon and Morel, 1983; Lee et al., 1998; Stramska et al., 2003) and semi-analytical approaches (Carder et al., 1999), including the evaluation of radiative transfer modeling simulations (Mobley and Sundman, 2008). This would also serve as a contribution to the tuning of existing (heritage) IOP retrieval approaches (Werdell et al., 2018).

Hyperspectral Argo data, on other hand, will have their own scientific value in lieu of just supporting satellite missions. The exploitation of high-resolution data (in the spatial, temporal, and spectral extension), measured by a single system that carries also IOP sensors, would potentially reduce uncertainties (due to temporal or spatial biases) in existing inverse algorithms (Rehm and McCormick, 2011). Thence, the availability of hyperspectral AOPs quantities will enable the derivation of high spectral resolution IOP data, and this will considerably reduce the ambiguities introduced by the relative contribution of different accessory phytoplankton pigments for example, as well as retrieving CDOM spectral optical properties which bring additional biogeochemical information (Gordon et al., 2009). The investigation of such contemporaneous measurements is demonstrated in the work of Rehm and Mobley (2013) where they estimated absorption and backscattering coefficients to an accuracy of ±0.01 m^–1 and ±0.0005 m^–1 respectively, based on hyperspectral vertical radiometric observations and concurrent measurements of the beam attenuation coefficient at 650 nm in optically simple oligotrophic waters. In the more optically complex waters where OAS do not covary with chl-a, Wollschläger et al. (2020) obtained a moderately complex parametrization of the underwater light field based on hyperspectral light measurements. Such parametrizations could be strongly improved by accounting for seasonal and regional variation based on data coming from hyperspectral Argo-floats. Increasing the number of bands will moreover allow testing of various wavelengths combinations based on the spectral diffuse attenuation to better link IOPs features (Brown et al., 2004), as well as to check for consistency of calibration of fluorescence observations (accompanied sensor on Argo float) (Xing et al., 2011).

The deployment of hyperspectral radiometers on unattended platforms like Argo floats applies higher demands on the capabilities of the sensors than conventional in-water radiometry. Some studies which performed hyperspectral light field measurements using drifting devices (Zielinski et al., 2006; Barnard et al., 2018) have already demonstrated stable short-term field deployment of autonomous hyperspectral radiance and irradiance measurements. However, both studies highlight the importance of technical features like radiometer power consumption, long-term stability and reliability, standardized interfaces and protocols, as well as intelligent data handling. At present, the first studies are performed to evaluate the use of hyperspectral radiometers on Argo floats. The sensor used for these studies is an improved, 1,000 m-depth rated RAMSES radiometer (TRIOS, Rastede, Germany) designed to collect the incoming irradiance in the spectral range of 320–950 nm, with a spectral resolution of 2.2 nm and a spectral accuracy of 0.2 nm. The sensor also features the new G2 Modbus-interface for easier platform integration (Trios, 2021). An important consideration for Argo-users is that the increased information content increases the demand for data transfer and processing capabilities. Handling huge amounts of data within a timeframe of 24 h as strived by the Argo community will be a demanding task. In case of the mentioned RAMSES radiometer, the data size is approximately six times that of the OCR-504. Together with the moderately higher power consumption (45 mA compared to 25 mA for the OCR-504), this constrains either the measurement frequency or deployment duration of an Argo float equipped with this sensor. Therefore, reducing power consumption of the sensor and the dimension of the data onboard the platform prior to transmission could be important future steps. This process would probably demand embedded computing power in the sensor and the platform itself, which poses challenges on the commercially available instrumentation.

This review captures the current state of light field measurements onboard Argo floats, describes their application in bio-optical ocean studies, and points to future directions in the transition from multispectral toward hyperspectral radiometry. This transition would allow an exploitation of light field measurements far beyond to what is covered to date.

Moving forward in operating Argo radiometry, hyperspectral observations would provide more detailed insights through direct hyperspectral investigation of the depth-resolved light distribution in different waters bodies and would be relevant to ocean color products. The development of novel approaches that exploit this information will open new avenues for bio-optical long-term studies of global ocean change and foster its relevance for carbon cycle investigation by addressing phytoplankton not only in terms of biomass, but also with respect to taxonomic composition. Critically, exploiting these advantages requires the development, in parallel, of new algorithms and procedures for quality assessment, data management, and on-board data reduction, all of which are intrinsically related to the technical features of both sensor and platform.

AJ made the draft version. JW and OZ participated on the outline of the manuscript. All authors contributed to manuscript revisions and agreed to the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.