SuperDove radiometric data assessment in coastal and inland waters

The use of high-resolution data in aquatic applications increased significantly in the last decade with the launch of decametre-scale optical sensors. More recently, commercial very-high resolution (VHR) sensors, offering finer spatial and temporal resolutions, have shown the potential of complementing data from high-resolution missions. Planet SuperDove (SD), with a band-setting similar to the Copernicus Sentinel-2 MultiSpectral Instrument (S2-MSI), a 3-m spatial resolution and quasi-daily revisiting time, show the potential for widening water monitoring applications to smaller water basins, and finer-scale phenomena. However, the uncertainties in SD products need to be quantified, to assess their fitness-for-purpose for these applications. This work aims to provide uncertainty estimates for SD-derived aquatic remote sensing reflectance (R_RS) in different water types, benefitting from the radiometric measurements of the AERONET-OC network. R_RS was derived from both Surface Reflectance (SR) products, distributed by Planet, or from data processed with ACOLITE. The comparability between SD and S2-MSI products was also assessed comparing R_RS and Rayleigh-corrected reflectance (R_RC) from S2-MSI and SD. The results indicate generally low performance across all bands for both SD R_RS products, except in the most turbid waters, and highlight the lack of a publicly available robust atmospheric correction processor for SD data for most optical water types. The comparison to S2-MSI shows promising results only when comparing R_RC values, but differences still suggest issues associated with calibration and radiometry of the SD sensors. The results also highlight the need for a harmonization strategy to ensure consistent integration of these datasets within multi-source monitoring systems.

1 Introduction

The use of high-resolution data in aquatic applications increased significantly in the last decade with the launch of optical sensors offering decameter-scale imagery (i.e., with a resolution of the order of a few tens of meters). These include Landsat-8 and 9, jointly developed by the National Aeronautics and Space Administration (NASA) and the United States Geological Survey (USGS), and the Copernicus Sentinel-2 (S2) A/B/C satellites. They host on-board the Operational Land Imager (OLI) and the MultiSpectral Instrument (MSI), respectively, acquiring at 30 m (OLI) and 10–60 m (MSI) resolution. Although originally designed for terrestrial applications, these sensors demonstrate good capabilities for inland-water and ocean-colour observations (e.g., Giardino et al., 2014; Bresciani et al., 2016; Pahlevan et al., 2017; 2020; Vanhellemont, 2019a). More recently, commercial very-high resolution (VHR) sensors offer even finer spatial and temporal resolutions, potentially widening the application of optical remote sensing data to smaller basins, and to phenomena characterized by fine temporal and spatial scales (Mansaray et al., 2021; Vatitsi et al., 2024).

The European Union Copernicus programme offers access to additional optical data from a few commercial missions, included in the so-called Copernicus Contributing Missions (CCM). These include commercial very-high resolution products from GeoSat (GEOSAT, 2023), WorldView (Vantor, https://vantor.com/product/worldview/), Pléiades and Pléiades Neo (AIRBUS, 2012), PlanetScope (Planet Labs PBC, 2023) and SPOT (AIRBUS, 2013) constellations. Their sensors usually offer very-high spatial resolution (down to submeter scale), and quasi-daily revisiting time allowed by a combination of various sensors belonging to each constellation. Among these commercial missions, PlanetScope additionally offers critical spectral capabilities for water quality monitoring applications through its SuperDove sensors. Planet SuperDove (SD) instruments (version PS2.SD, Planet Labs PBC, 2023), launched in flocks since 2020, are equipped with a set of 8 bands in the visible-NIR domain (centered at 443, 490, 531, 565, 610, 665, 705, and 865 nm), among which 6 are similar to those characterizing S2-MSI (i.e., those centered at 443, 490, 560, 665, 705, and 865 nm). These common bands may favor the integration of SD data in multi-sensor monitoring systems primarily based on S2-MSI data, to ensure frequent coverage over smaller water bodies. Additionally, the two spectral bands centered at 620 nm and 705 nm, respectively, offer a unique opportunity, particularly for phytoplankton and harmful algal bloom monitoring, being sensitive to the presence of pigments such as phycocyanin (characterizing cyanobacteria) and chlorophyll-a, and to the presence of high biomass concentrations (see Supplementary Figure S1 in the Supplementary Material for a full description of bands center-wavelengths, and spectral response functions, SRFs). Previous versions of these sensors (Dove Classic and Dove-R) were only equipped with 4 bands (centered at 490, 565, 665, and 865 nm). A few examples of the applications of Dove and SuperDove imagery include: identifying algal scum in riverine waters (Herrmann et al., 2024); cyanobacteria detection in riverine and estuarine waters (Yao et al., 2024); monitoring of small inland water bodies (Perin et al., 2021); detection of floating algal blooms (Ahn et al., 2024).

Uncertainties in these sensors derived products need to be quantified, to assess their fitness-for-purpose for water monitoring applications (Mélin et al., 2016; IOCCG Protocol Series, 2019). Moreover, an assessment of their added value and a thorough cost-benefit analysis should be undertaken, given the high monetary costs associated with commercial data purchase and the potential environmental costs related to the substantial storage and processing requirements of a systematic application of these data.

Recently, remote sensing reflectance (R_RS) derived from both Dove and SuperDove data—using either the operational Planet Surface Reflectance (SR) products (Planet Labs PBC, 2023) or outputs generated with the ACOLITE atmospheric-correction processor (Vanhellemont, 2019a)—have been evaluated over coastal and inland waters. Maciel et al. (2020) assessed the accuracy of Dove R_RS over the Amazon area, showing quite good accuracy in all visible and NIR bands in turbid waters, but unsatisfactory results in clear lakes. Conversely, Chasles et al. (2025) reported promising results obtained over a few Brazilian inland water bodies for SD, yet exhibiting varying degrees of accuracy in different water types. Vanhellemont (2020) and Vanhellemont (2023) compared Dove and SD data with hyperspectral in situ data in turbid Belgian coastal waters and in clearer conditions at the Acqua Alta Oceanographic Tower (AAOT) in the northern Adriatic Sea. These works reported good consistency among the satellites within the constellation and comparable, promising performances by both ACOLITE and SR products in turbid waters, whereas their suitability decreased in clearer conditions. SD products were also evaluated in very turbid waters of the Rio de la Plata by Dogliotti et al. (2024), showing a tendency to overestimate in situ reflectance, especially at low values, with reduced performances with respect to decameter sensors. Data from several coastal sites around the world belonging to the ocean component of the AERONET network, AERONET-OC (Zibordi et al., 2009; 2021) were exploited by Vanhellemont (2019b) to assess R_RS from Dove sensors, reporting adequate results for green and red bands in brighter waters, but poor performance over dark water targets, especially for the NIR, potentially due to Sun glint and adjacency contaminations but also due to poor sensor performance and calibration issues. SD radiometric data were finally evaluated comparing SD Top-of-Atmosphere (TOA) reflectance R_TOA and R_RS values with quasi-synchronous S2-MSI products (Kabir et al., 2025) across various sites globally distributed. The study showed differences ranging from 0.7% to 13% for R_TOA, and from 16% to 95% for R_RS. However, significant discrepancies were observed among SD sensors, suggesting the presence of sensor-to-sensor inconsistencies within the SD constellation.

When evaluating the added value of incorporating Dove (Classic or -R) and SD data into operational monitoring, reported results vary depending on the type of analysis and datasets used. For instance, Atton Beckmann et al. (2025) applying SD data, reported that SD performed well in comparison to S2-MSI for monitoring phytoplankton in a small, eutrophic lake, with ‘distinct advantages arising from higher spatial and temporal resolution’. Conversely, Liu et al. (2022), applying Dove data, showed that the accuracy of Harmful Algal Bloom (HAB) detection by decameter-scale sensors was higher than by Doves. Finally, Mansaray et al. (2021), also applying Doves, concluded that VHR data have comparable capabilities to those from Landsat-8-OLI and S2-MSI for sensing key water quality parameters. However, they also concluded that their finer spatial and temporal resolutions are counterbalanced by the fewer spectral bands and the cost of acquiring commercial data, suggesting they may be used only for applications with strong daily variability of water parameters and/or for very small waterbodies.

Paving the way for future analyses of the added value of VHR data in water quality monitoring programs, this work aims to provide uncertainty estimates for SD-derived aquatic reflectance in different water types, i.e., water characterized by different optical properties. It benefits of fiducial reference radiometric measurements from AERONET-OC and greater data availability since the first years of SD operations. Data from sites located in European coastal waters were considered, as they are broadly representative of the variability observed in European marine environments (Cazzaniga and Mélin, 2024). Additionally, AERONET-OC data from four inland-water sites were also included, to evaluate SD performances in inland waters and basins affected by HAB events. Finally, to increase data availability in highly turbid waters, an AERONET-OC site in the Río de la Plata estuary was also included.

Subsequently, a comparative analysis was carried out with respect to S2-MSI 10-m scale imagery, from all S2 satellites (A, B and C). This was done considering both R_RS and Rayleigh corrected reflectance (R_RC) values, quite often used to define indices applied to remote sensing of HABs. This enabled an evaluation of the comparability between the two sensor types and their potential for joint use in water quality monitoring systems.

2 Materials and methods

2.1 In situ data

AERONET-OC R_RS values were obtained from all European sites and from inland water sites installed in the U.S. (Figure 1). These include i. Casablanca Platform (CSP) in the western Mediterranean Sea, exhibiting occurrence of clear Case-1 waters (i.e., waters in which the optical properties are dominated by chlorophyll-a); ii. Acqua Alta Oceanographic Tower (AAOT) in the northern Adriatic Sea, Galata (GLT) and Section-7 (ST7) platforms in the Black Sea for optically-complex waters characterized by varying concentrations of suspended particulate matter and colored dissolved organic matter (CDOM); iii. Gustaf Dalen, Helsinki and Irbe lighthouse towers (GDLT, HLT, and IRLT) in the Baltic Sea, also characterized by optically-complex waters, with high concentrations of CDOM and cyanobacteria events during summer; iv. Zeebrugge-MOW and Thornton C-power sites (ZEE and TCP), in highly turbid waters off the Belgian coast; v. Inland water sites: Palgrunden (PAL) in the Swedish Lake Vänern, two sites in Lake Okeechobee (LOK and LOKN) and one in Lake Erie (LER). The PAL site is characterized by oligotrophic waters, optically dominated by CDOM (Philipson et al., 2016). Instead, Lake Erie and Lake Okeechobee are characterized by shallow waters, sediment resuspension and frequent summertime cyanobacteria blooms (Moore et al., 2019; Jiang et al., 2025). Additional data were obtained at the RdP-EsNM site (RDP), located in the Río de la Plata estuary, characterized by highly turbid waters, with very high concentration of inorganic suspended matter and cyanobacteria blooms events (Dogliotti et al., 2024). The exact location of the various sites is shown in Figure 1.

Figure 1

Figure 1. Locations of the AERONET-OC sites included in the analysis.

In situ R_RS values were obtained from Level-2 (Version 3) normalized water leaving radiance (L_WN) corrected for bidirectional effects (for bands up to 665 nm) according to Morel et al. (2002) and band-shifted to match SD center-wavelengths (Mélin and Sclep, 2015).

2.2 Satellite data

SuperDove Analytic Radiance (AR) products and surface reflectance products (SR), as distributed by Planet, were obtained from a 3 × 3-km window centered at the respective AERONET-OC sites locations. The 3 × 3 km dimension was selected to minimize quota consumption for SD data download, while still being suitable for ACOLITE processing (Vanhellemont, 2020). Planet Data Application Programming Interface (API) was used to initially filter the products for cloudiness. Only products available within 1 h from an AERONET-OC measurement, with a satellite viewing zenith angle lower than 60° and a sun zenith angle lower than 70° were selected. A visual check was also performed to further exclude data affected by clouds and strong Sun glint. For the comparison with MSI, additional cloud- and glint-free SD products were selected manually to increase the number of matchups (N).

TOA radiance L_TOA and reflectance R_TOA were obtained multiplying digital number (DN) values by the corresponding conversion factors provided with AR products. Planet derives SR from AR data using atmospheric correction based on the vector version of the Second Simulation of the Satellite Signal in the Solar Spectrum Vectorial Version (6SV2.1, Vermote et al., 1997) radiative transfer code with a fixed aerosol model (continental) and aerosol optical depth at 550 nm (AOD₅₅₀), water vapor and ozone values retrieved from MODIS near-real-time data (MOD09CMA, MOD09CMG and MOD08-D3) when available, or set to a fixed value otherwise, i.e., 0.226, 4.05 g cm^-2, 0.255 cm-atm, respectively (Planet Labs PBC, 2023). Alternatively, in this work, AR data were also atmospherically corrected through the ACOLITE software using the Dark Spectral Fitting (DSF) processor (Vanhellemont, 2019b; 2023; Vanhellemont and Ruddick, 2021), fully recognizing that DSF has been developed for atmospheric correction over turbid waters, which is not the case for all AERONET-OC sites here considered. R_RS and R_RC were obtained from both SD and S2-MSI products, at 3-m and 10-m resolution, respectively. R_RS from both DSF and SR products were fully normalized for bidirectional effects, applying the same correction used for AERONET-OC data.

2.3 Matchups criteria and statistics

At each site, SD matchups were identified with quasi-synchronous measurements from either in situ or S2-MSI products, considering a maximum time difference of 1 h between SD and in situ or S2-MSI acquisition time.

For SR products, valid pixels were identified applying the usable data mask (udm) provided along SR products, allowing to filter pixels affected by clouds, snow, shadow and haze. From udm, the classification confidence map was also used, excluding pixels classified with a level of confidence below 90%. To identify and filter non-water pixels, the same flagging criteria implemented in ACOLITE were used, i.e., excluding pixels with R_TOA(865)>0.1 and R_TOA exceeding 0.3 at any band. For DSF products, ACOLITE l2_flags product mask was applied. R_RC products were masked using only l2_flags bits 0, 2 and 4, to only exclude non-water (including clouds) and out-of-scene pixels. It is recognized however that the thresholds imposed on R_TOA may lead to the exclusion of pixels affected by strong cyanobacteria surface accumulation and a more effective masking criteria should be identified (see comments in the discussion section).

Matchups were discarded if more than 10% of the pixels within a 1 × 1 km window were identified as non-water pixels or contained invalid values. This larger window was used to reduce the effects of clouds. Even if a 100% of pixel validity is usually recommended when working with coarser resolution products, a looser criterium was preferred mainly as to avoid excluding matchups from sites with large platforms hosting AERONET-OC instruments or vessels, which surface area is not negligible with respect to pixel size. Afterwards, pixels were additionally masked band by band, when their values were outside the range defined by µ(λ) ± 1.5σ(λ), where µ(λ) and σ(λ) are the mean and standard deviation, respectively, calculated at each band λ from the pixels within a 500 × 500 m window.

The median value was finally calculated for each band from non-masked pixels within the 500 × 500 m window at all sites, except for RDP and ZEE. For these sites, a smaller window (100 × 100 m) was considered to avoid including different water masses, shifted by, respectively, 60 and 100 m in the northern direction to avoid any effect due to the site structures (Vanhellemont and Ruddick, 2018; Dogliotti et al., 2024).

Validation statistics included root mean squared difference (RMSD, Equation 1), mean relative difference (Ψ, Equation 2) and the mean absolute relative difference (|Ψ|, Equation 3).

$\text{RMSD}=\frac{1}{N}{\sum }_{i=1}^N{\left(x_i^{SD}-x_i^{is}\right)}^2(1)$

$\mathrm{\Psi }=100\frac{1}{N}{\sum }_{i=1}^N\frac{\left(x_i^{SD}-x_i^{is}\right)}{x_i^{is}}(2)$

$\left|\mathrm{\Psi }\right|=100\frac{1}{N}{\sum }_{i=1}^N\frac{\left|x_i^{SD}-x_i^{is}\right|}{x_i^{is}}(3)$

where N is the number of matchups, $x_i^{SD}$ is SD product value for the matchup i, and $x_i^{is}$ is the corresponding value from AERONET-OC measurements. The Spectral Angle Mapper values (SAM, Kruse et al., 1993, in degrees; Equation 4) were also calculated between SR or DSF products and in situ values in the 442–665 nm spectral range, to assess their performances in terms of spectral shape.

$\text{SAM}=\frac{180}{\pi }{\cos }^{-1}\left(\frac{{\sum }_{i=1}^N\left(x_i^{SD}{x}_i^{is}\right)}{\sqrt{{\sum }_{i=1}^N{x}_i^{SD}{\hspace{0.17em}}^2}\sqrt{{\sum }_{i=1}^N{x}_i^{is}{\hspace{0.17em}}^2}}\right)(4)$

RMSD, Ψ, and |Ψ| were also calculated when comparing SD and S2-MSI values, replacing $x_i^{is}$ with $x_i^{MSI}$, i.e., S2-MSI derived corresponding values, in Equations 1–3. Both R_RS and R_RC were considered for common center-wavelengths, but also for a few representative R_RC-derived indices used in EO water quality monitoring systems. These latter included:

1. the Normalized Difference Chlorophyll Index (NDCI, Mishra and Mishra, 2012; Equation 5):

$NDCI=\frac{R_{RC}\left(705\right)-R_{RC}\left(665\right)}{R_{RC}\left(705\right)+R_{RC}\left(665\right)}(5)$

2. the Normalized Difference Vegetation Index (NDVI, Rouse et al., 1974; Equation 6):

$NDVI=\frac{R_{RC}\left(865\right)-R_{RC}\left(665\right)}{R_{RC}\left(865\right)+R_{RC}\left(665\right)}(6)$

3. the Normalized Difference Water Index (NDWI, Oyama et al., 2015; Equation 7):

$NDWI=\frac{R_{RC}\left(490\right)-R_{RC}\left(865\right)}{R_{RC}\left(490\right)+R_{RC}\left(865\right)}(7)$

3 Results

3.1 SD validation at AERONET-OC sites

A total of 630 and 766 matchups between in situ data and, respectively, SD SR and DSF products were obtained (Figures 2, 3). Matchups were unevenly distributed among sites and no matchups were available for HLT and TCP (sites with an incomplete temporal coverage). However, besides AERONET-OC data availability, the number of matchups strongly depends on the distance from the shoreline. The SD constellation was primarily designed for land applications, resulting in partial coverage of coastal areas. Additionally, manual screening for glint may affect the number of available matchups, as the SD sensor design does not include a tilt strategy to minimize glint contamination. At certain sites, glint may occur frequently depending on latitude and season.

Figure 2

Figure 2. Scatterplots for SR R_RS against AERONET-OC R_RS at 442, 490, 531, 565, 610, 665, 705 and 865 nm. In the legend, numbers in brackets show the number of matchups and the number of sensors available at the respective site. RMSD is expressed in sr⁻¹, $\mathrm{\Psi }$ and |$\mathrm{\Psi }$| in %.

Figure 3

Figure 3. As Figure 2 but for ACOLITE DSF R_RS.

Matchups were obtained from 152 to 161 different sensors from the SD constellation, for SR and DSF products, respectively, with less than 10% of the sensors showing more than 10 matchups. The majority (49%) is encountered with a single matchup. At each site, none of the available sensors has more than 3 matchups, which prevents deriving sensor-specific statistics.

Both SR and DSF products overestimate on average in situ values, with best (still quite unsatisfactory) performance in the green. Less scattered matchups are shown by SR products, whose difference statistics are halved with respect to DSF ones, although based on a lower number of matchups and above 81% when considering |Ψ|. SR results are improved considerably when including only the most turbid-water sites, i.e., RDP and ZEE, even better for DSF products, with $\mathrm{\Psi }$ values between +10.5% at 610 nm and +32.4% at 442 nm (Ψ is between +18.5% and +72.4% for SR products), excluding results for 865 nm. The distribution of matchup points along the 1:1 line at longer wavelengths is more prominent for the RDP and ZEE sites compared to other sites (Figure 2). Less turbid, but certainly not clear, waters at the U.S. inland water sites do not show equal performance, with r² between 0.1 and 0.7 and a quite strong overestimation also at longer wavelengths, which are generally quite useful in HAB monitoring systems (i.e., 610, 665 and 705 nm, with Ψ equal to +95.9%, +99.7%, and +133.7% respectively, for DSF products).

The better performance over very turbid waters also applies to the spectral shape. Figure 4 reports all the spectra available from both SR and DSF products and the corresponding in situ spectra at the RDP and CSP sites. At RDP, except for a few cases, both products show quite reasonable spectral shapes, although slightly better for DSF. Conversely, at CSP (taken as example representative of fairly clear waters), the very high residual values at longer wavelengths make the spectra look much flatter than the spectral shape measured in situ. These differences are also reflected by the SAM values (Table 1), with the smallest average values shown for RDP and ZEE (5.4° and 4.5° at RDP, 2.1° and 3.0° at ZEE for SR and DSF, respectively) and the largest average values shown for CSP. The SAM values from SR and DSF are usually comparable.

Figure 4

Figure 4. In situ (top) and SuperDove-derived R_RS values from SR (middle) and DSF (bottom) products for RDP (left) and CSP (right) sites. Only common available spectra from DSF, SR and in situ measurements have been included. The colors are not unique but are consistent between plots in each column.

Table 1

Table 1. SAM values (in degrees) for the matchups at the various AERONET-OC sites for SR and DSF calculated for bands between 442 and 665 nm. The values at ZEE have been calculated without the band at 610 nm. The values for S2-MSI have been calculated only from 443, 490, 560 and 665 nm.

For completeness the results are reported for all bands. However, the results at 865 nm may not be meaningful, considering the relatively large uncertainties affecting AERONET-OC in the NIR across most water types (Zibordi et al., 2021).

3.2 SD and S2-MSI comparison

3.2.1 R_RS and AOD₅₅₀

A total of 198 matchups were obtained comparing SD and MSI products. Results from the comparison of both ACOLITE DSF R_RS and R_RC are here reported for completeness, despite the very low performance shown by R_RS products.

Quite large differences were observed for R_RS values (Figure 5). SD products show higher R_RS values with respect to S2-MSI in the blue, but lower in the green (Ψ = −12.8%). However, when considering each single site, the results may vary. SD values are lower at shorter wavelengths at the AAOT and PAL, while a positive bias is observed in the green and red (Figure 6) at these same sites. At the most turbid sites (ZEE and RDP) and at LER, SD R_RS values are generally lower than S2-MSI values at all bands, except for 865 nm.

Figure 5

Figure 5. Scatterplots for SD and S2-MSI R_RS products at 442, 490, 565, 665, 704 and 865 nm obtained using ACOLITE. No band-shifting is applied. Numbers in brackets show the number of matchups at each site. RMSD is expressed in sr⁻¹, $\mathrm{\Psi }$ and |$\mathrm{\Psi }$| values are in %.

Figure 6

Figure 6. Statistics obtained at each site comparing SD and S2-MSI R_RS values at 442, 490, 565, 665, and 704 nm obtained using ACOLITE. LOK values were obtained from a single matchup and are not reported in r² plot.

The results shown in Figure 5, and the pronounced overestimations of SD R_RS illustrated in Figure 3, indicate that R_RS data derived by ACOLITE for S2-MSI are likewise overestimated with respect to in situ data. This is further supported by a direct comparison of S2-MSI to AERONET-OC data (see Discussion section, and Pahlevan et al., 2021).

When comparing AOD₅₅₀ values applied during atmospheric correction, the results show considerable scatter. On average, SD values are lower than S2-MSI ones, with RMSD = 0.159, Ψ = −39.6% and r² = 0.2.

3.2.2 R_RC and R_RC-derived indices

The R_RC products comparison showed more promising results, mainly for bands at 560, 665 and 705 nm (r² = 0.9). SD values are slightly higher than S2-MSI ones, with Ψ generally below +26% (see Figure 7), with the exception of a few bands at IRLT, ZEE and RDP (see Figure 8), where SD values are lower when compared to S2-MSI. Considering all sites, the lowest agreement is observed at 865 nm with r² = 0.5 and |Ψ|> 70%.

Figure 7

Figure 7. As Figure 5 but for SD and S2-MSI R_RC products. RMSD is unitless, $\mathrm{\Psi }$ and |$\mathrm{\Psi }$| values are in %.

Figure 8

Figure 8. As Figure 6 but for SD and S2-MSI R_RC values.

These results are translated in a fairly low comparability for R_RC-derived indices, with SD showing lower NDCI and NDWI values and higher NDVI values compared to S2-MSI (Figure 9). Also in this case, results vary site by site (Table 2). The best results are reported for NDCI at LER (|Ψ|<27%, r² = 0.8), for NDWI at ZEE (|Ψ|<16%, r² = 0.5), and for NDVI at RDP (|Ψ| = 22%, r² = 0.8). These differences could lead to inconsistent diagnostics in terms of water quality.

Figure 9

Figure 9. Scatterplots for SD and S2-MSI indices from R_RC. The numbers in brackets show the number of matchups at each site. RMSD is unitless, $\mathrm{\Psi }$ and |$\mathrm{\Psi }$| values are in %.

Table 2

Table 2. Statistics obtained at each site comparing SD and S2-MSI R_RC-derived indices. RMSD is unitless, |Ψ| and Ψ values are in %.

It should be noted that S2-MSI and SD have different spectral response functions (Supplementary Figure S1 in the Supplementary Material), and no type of spectral adjustment has been applied for this comparison. Besides the differences in wavelengths, the R_RC S2-MSI and SD products are also not supposed to match perfectly as the geometry of illumination and observation slightly differs between the two sensors. Nevertheless, the 1-h time difference criterion effectively limits differences in solar zenith angles (on average 2.3° ± 2.4°), whereas the differences between the two sensors’ viewing zenith angle remains moderate (4.3° ± 3.2°), with mean values of 7.0° for S2-MSI and 3.1° for SD, respectively.

4 Discussion

The enhanced spatial and temporal resolution provided by the Planet SD constellation offers the possibility of monitoring smaller water basins and smaller scale phenomena at a time scale very useful for Earth Observation-based water quality monitoring systems. To assess the uncertainties related to these sensors data, SD radiometric products were assessed against in situ data from the AERONET-OC network at coastal and inland water sites. Both SR (atmospherically-corrected products distributed by Planet) and ACOLITE-derived R_RS products were evaluated, showing quite low performance at all bands (with very large overestimates), mainly at relatively clear water sites.

It should be, however, recalled that SD sensors have not been designed specifically for water applications. Their relatively low signal-to-noise ratio and the radiometric sensor-to-sensor inconsistencies observed across SD constellation (Kabir et al., 2025) may hinder their applicability in clearer waters. Sun glint contamination can also affect the performance. No glint correction was applied in this study, being difficult to address with no available bands in short-wave infrared (SWIR), particularly in non-clear waters where the hypothesis of null signal in the NIR may not be valid (Vanhellemont, 2023). It is however recalled that images affected by Sun glint were manually discarded. Additionally, the atmospheric correction applied by Planet has not been specifically designed for water applications, whereas ACOLITE has been developed mainly to address atmospheric correction over very turbid waters. Previous assessments of S2-MSI DSF products (see Supplementary Figure S2 in the Supplementary Material) also show overall overestimate with respect to in situ data (Ψ > +107% at all bands, except for R_RS (705), available only for a few sites, with Ψ = +54.7%), and a strong misalignment from the 1:1 line for clearer-water matchup points; conversely good results are reported for very turbid water sites (e.g., Dogliotti et al., 2023). Still, the results appeared less scattered than for SD, with better r² values (between 0.6 and 0.9). SAM mean values (see Table 1) for S2-MSI are instead comparable with respect to SD values for most sites, except for the relatively clear water CSP site. It should be considered however that SAM is calculated for S2-MSI without the bands at 531 and 610 nm, which are not available for the MSI sensors, and besides that, the uncertainty estimates rely on a different number of matchups for each site.

As mentioned in previous sections, difficulties are encountered when trying to define an efficient masking strategy, challenged by the lack of a SWIR band, and also by the inter-band misalignment sometimes observed in SD products over water. ACOLITE masks, based on fixed thresholds for TOA reflectance values, may sometimes not be sufficient to mask site structures and shadows. This was partially solved combining the use of the median, robust to outliers, when calculating matchup values, and the additional criteria based on the range µ(λ) ± 1.5σ(λ) applied band by band, generally allowing to exclude structure and shadow-affected pixels. However, both these masks may lead to the incorrect exclusion, for example, of pixels affected by cyanobacteria surface accumulation (as observed for a few dates at the LOK site).

Besides masking, the SD products inter-band misalignment may also limit the applicability of spectral indices, such as those used to identify HABs, vegetation or other phenomena characterized by inhomogeneous spatial distribution on the water surface.

Another point of discussion addressed here is associated with the potential added value provided by the more frequent availability of SD data with respect to S2-MSI. Besides advantages for monitoring applications, this also should translate in larger validation datasets. In fact, between 2021 and 2023, the total number of valid SD matchups is almost 50% larger than S2-MSI ones, with N equal to 707 and 484, respectively, considering DSF products. Excluding the CSP, IRLT and RDP sites, where less SD matchups are available, availability for SD raises between +18% at ST7 and +273% at LOKN with respect to S2-MSI. The operation of a constellation is beneficial for this increase in the number of matchups and coverage (Section 3.1) but, with respect to validation of traditional missions, it also raises new challenges to identify sensor-specific issues.

The results obtained for the SD products validation are not completely unexpected. However, they highlight, the current lack of a publicly-available robust atmospheric correction processor for SD sensors for a variety of water types. At the same time, in particular in very turbid or bright waters, they provide promising results, when considering not fully atmospherically corrected products (R_RC). Also in this case, the results are not really surprising, considering SD sensors are calibrated on orbit using S2-MSI data (Collison and Bourne, 2022). Yet, regardless of the relative agreement of SD and S2-MSI R_RC products, differences obtained when comparing representative spectral indices used for water quality monitoring suggest that efforts are still required in harmonizing these sensors data before merging them in operational systems for water quality monitoring.

In conclusion, this study provides a comprehensive assessment of Planet SuperDove (SD) radiometric data in coastal and inland waters using AERONET-OC reference measurements, covering a variety of optical water types. Results indicate that SD products performance remains limited over weakly scattering waters, likely due to sensor design, radiometric calibration issues and the lack of an atmospheric correction processor optimized for non-turbid waters. The findings underscore the need for improved calibration and harmonization strategies with S2-MSI to enable the consistent integration of SD data into multi-sensor monitoring systems.

Future work should first address the need for improving the on-orbit calibration of SD sensors over aquatic environments, either using in situ or modelled (Werdell et al., 2007) data, or through cross-calibration against other well-calibrated sensors, such as S2-MSI. This strategy has been applied by Kabir et al. (2025), who demonstrated considerable improvements in residuals when comparing SD data with quasi-synchronous S2-MSI observations. However, additional data and effort will be required to obtain further benefits from independent, sensor-by-sensor cross-calibration.

Additional support may come from adapting atmospheric-correction processors, successfully applied to decametre-scale sensors (see, for example, Pahlevan et al., 2021 and references therein), to also accommodate SD sensors. Nonetheless, limitations may persist due to SD sensors’ design and characteristics, which are not specifically optimized for aquatic applications.

Finally, due to the intrinsic uncertainties that also affect S2-MSI data, in situ measurements are needed to provide a robust validation of all these strategies. Future work will aim to address this by collecting additional in situ radiometric and bio-optical data in inland water environments.

Data availability statement

Publicly available datasets were analyzed in this study. This data can be found here: the AERONET-OC dataset at https://aeronet.gsfc.nasa.gov/cgi-bin/draw_map_display_seaprism_v3; Sentinel-2 MSI L1C products (utilized to derive R_RS and R_RC values) at https://browser.dataspace.copernicus.eu/. Planet SuperDove data are instead subject to access regulations and cannot be re-distributed by the authors.

Author contributions

IC: Writing – review and editing, Formal Analysis, Methodology, Data curation, Writing – original draft, Investigation, Visualization, Software, Conceptualization, Validation. AD: Funding acquisition, Writing – review and editing. SK: Funding acquisition, Writing – review and editing. FM: Software, Supervision, Project administration, Conceptualization, Writing – review and editing.

Funding

The author(s) declared that financial support was received for this work and/or its publication. This work benefitted from support from the Joint Research Centre through the Wat€rs project. The support provided by DG DEFIS, i.e., the European Commission Directorate General for Defence Industry and Space, and by the Copernicus Programme is also gratefully acknowledged. Thanks also to the Swedish National Space Agency, projects 2021-00050 and 2021-00064.

Acknowledgements

The authors would like to thank the AERONET-OC PIs Giuseppe Zibordi, Barbara Bulgarelli, Dimitry Van der Zande, Elena Lind, Francesca Ortenzio, Marco Talone, Menghua Wang, Nima Pahlevan, Pawan Gupta, Steve Ruberg, and Timothy Moore for their effort in establishing and maintaining the sites included in this study. They would also like to thank the European Space Agency for SuperDove data access and the Copernicus programme for Sentinel-2 data access. They would like to thank the two reviewers for their constructive and insightful comments.

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

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

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

Publisher’s note

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

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/frsen.2025.1753296/full#supplementary-material

References

Ahn, J., Kim, K., Kim, Y., Kim, H., and Lee, Y. (2024). Detection of floating algae blooms on water bodies using PlanetScope images and shifted windows transformer model. Remote Sens. 16, 3791. doi:10.3390/rs16203791

CrossRef Full Text | Google Scholar

AIRBUS (2012). Pléiades imagery - user guide. Available online at: https://storage.googleapis.com/p-ssp-iep-prod-8ff-strapi-uploads/210415_Airbus_Pleiades_Imagery_user_guide_be12b8f35b/210415_Airbus_Pleiades_Imagery_user_guide_be12b8f35b.pdf.

Google Scholar

AIRBUS (2013). SPOT 6/7 imagery - user guide. Available online at: https://storage.googleapis.com/p-ssp-iep-prod-8ff-strapi-uploads/SPOT_User_Guide_2021_09_low2_d145a2a31b/SPOT_User_Guide_2021_09_low2_d145a2a31b.pdf.

Google Scholar

Atton Beckmann, D., Spyrakos, E., Hunter, P., and Jones, I. D. (2025). Widespread phytoplankton monitoring in small lakes: a case study comparing satellite imagery from planet SuperDoves and ESA sentinel-2. Front. Remote Sens. 6, 1549119. doi:10.3389/frsen.2025.1549119

CrossRef Full Text | Google Scholar

Bresciani, M., Giardino, C., Lauceri, R., Matta, E., Cazzaniga, I., Pinardi, M., et al. (2016). Earth observation for monitoring and mapping of Cyanobacteria blooms. Case studies on five Italian lakes. J. Limnol. 76, 127–139. doi:10.4081/jlimnol.2016.1565

CrossRef Full Text | Google Scholar

Cazzaniga, I., and Mélin, F. (2024). How representative are European AERONET-OC sites of European marine waters? Remote Sens. 16, 1793. doi:10.3390/rs16101793

CrossRef Full Text | Google Scholar

Chasles, R. G., Maciel, D. A., Barbosa, C. C. F., Novo, E. M. L., Martins, V. S., Paulino, R., et al. (2025). Accuracy assessment of PlanetScope SuperDove products for aquatic reflectance retrieval over Brazilian inland and coastal waters. ISPRS J. Photogramm. Remote Sens. 227, 678–690. doi:10.1016/j.isprsjprs.2025.06.036

CrossRef Full Text | Google Scholar

Collison, A., and Bourne, H. (2022). On-orbit radiometric calibration of the planet satellite fleet. San Francisco, CA, United States. Available online at: https://assets.planet.com/docs/radiometric_calibration_white_paper.pdf.

Google Scholar

Dogliotti, A. I., Merlo, R., Yema, L., and O’Farrell, I. (2023). Validation of Sentinel-2/MSI water reflectance and water quality products in the turbid waters of Río de la Plata estuary using fixed automated hyperspectral in situ observations, in Remote Sensing of the Ocean, Sea Ice, Coastal Waters, Editors C. R. Bostater, and X. Neyt, 1, 1. SPIE. doi:10.1117/12.2684141

CrossRef Full Text | Google Scholar

Dogliotti, A. I., Piegari, E., Rubinstein, L., Perna, P., and Ruddick, K. G. (2024). Using the automated HYPERNETS hyperspectral system for multi-mission satellite ocean colour validation in the Río de la Plata, accounting for different spatial resolutions. Front. Remote Sens. 5, 1354662. doi:10.3389/frsen.2024.1354662

CrossRef Full Text | Google Scholar

GEOSAT (2023). GEOSAT-2 products and services. Available online at: https://earth.esa.int/eogateway/documents/20142/37627/GEOSAT-2-Imagery-User-Guide.pdf.

Google Scholar

Giardino, C., Bresciani, M., Cazzaniga, I., Schenk, K., Rieger, P., Braga, F., et al. (2014). Evaluation of multi-resolution satellite sensors for assessing water quality and bottom depth of Lake Garda. Sensors Switz. 14, 24116–24131. doi:10.3390/s141224116

PubMed Abstract | CrossRef Full Text | Google Scholar

Herrmann, M., Brehm, T., Kleinteich, J., Klotz, F., and Baschek, B. (2024). “Detecting algal scum in an inland river using planet and Sentinel-2 multispectral imagery,” in Remote sensing for agriculture, ecosystems, and hydrology XXVI. Editors C. M. Neale, A. Maltese, C. Nichol, and C. R. Bostater, 38. SPIE. doi:10.1117/12.3030011

CrossRef Full Text | Google Scholar

IOCCG Protocol Series (2019). IOCCG report number 18 - uncertainties in ocean colour remote sensing.

Google Scholar

Jiang, M., Cannizzaro, J., McFarland, M. N., Wistort, Z., Beckler, J. S., Hu, C., et al. (2025). Modeling water quality and Cyanobacteria blooms in Lake Okeechobee: II. Dynamics of diurnal cycles and impacts of Cyanobacteria diel vertical migration. Ecol. Modell. 505, 111107. doi:10.1016/j.ecolmodel.2025.111107

CrossRef Full Text | Google Scholar

Kabir, S., Saranathan, A. M., Barnes, B. B., Ashapure, A., O’Shea, R. E., and Stengel, V. G. (2025). Feasibility of PlanetScope SuperDove constellation for water quality monitoring of inland and coastal waters. Front. Remote Sens. 6, 1624783. doi:10.3389/frsen.2025.1624783

CrossRef Full Text | Google Scholar

Kruse, F. A., Lefkoff, A. B., Boardman, J. W., Heidebrecht, K. B., Shapiro, A. T., Barloon, P. J., et al. (1993). The spectral image processing system (SIPS)—Interactive visualization and analysis of imaging spectrometer data. Remote Sens. Environ. 44, 145–163. doi:10.1016/0034-4257(93)90013-N

CrossRef Full Text | Google Scholar

Liu, S., Glamore, W., Tamburic, B., Morrow, A., and Johnson, F. (2022). Remote sensing to detect harmful algal blooms in inland waterbodies. Sci. Total Environ. 851, 158096. doi:10.1016/j.scitotenv.2022.158096

PubMed Abstract | CrossRef Full Text | Google Scholar

Maciel, D. A., Novo, E. M. L. D. M., Barbosa, C. C. F., Martins, V. S., Flores Júnior, R., Oliveira, A. H., et al. (2020). Evaluating the potential of CubeSats for remote sensing reflectance retrieval over inland waters. Int. J. Remote Sens. 41, 2807–2817. doi:10.1080/2150704X.2019.1697003

CrossRef Full Text | Google Scholar

Mansaray, A. S., Dzialowski, A. R., Martin, M. E., Wagner, K. L., Gholizadeh, H., and Stoodley, S. H. (2021). Comparing PlanetScope to Landsat-8 and Sentinel-2 for sensing water quality in reservoirs in agricultural watersheds. Remote Sens. 13, 1847. doi:10.3390/rs13091847

CrossRef Full Text | Google Scholar

Mélin, F., and Sclep, G. (2015). Band shifting for ocean color multi-spectral reflectance data. Opt. Express 23, 2262–2279. doi:10.1364/OE.23.002262

PubMed Abstract | CrossRef Full Text | Google Scholar

Mélin, F., Sclep, G., Jackson, T., and Sathyendranath, S. (2016). Uncertainty estimates of remote sensing reflectance derived from comparison of ocean color satellite data sets. Remote Sens. Environ. 177, 107–124. doi:10.1016/j.rse.2016.02.014

CrossRef Full Text | Google Scholar

Mishra, S., and Mishra, D. R. (2012). Normalized difference chlorophyll index: a novel model for remote estimation of chlorophyll-a concentration in turbid productive waters. Remote Sens. Environ. 117, 394–406. doi:10.1016/j.rse.2011.10.016

CrossRef Full Text | Google Scholar

Moore, T. S., Feng, H., Ruberg, S. A., Beadle, K., Constant, S. A., Miller, R., et al. (2019). SeaPRISM observations in the western basin of Lake Erie in the summer of 2016. J. Gt. Lakes. Res. 45, 547–555. doi:10.1016/j.jglr.2018.10.008

CrossRef Full Text | Google Scholar

Morel, A., Antoine, D., and Gentili, B. (2002). Bidirectional reflectance of oceanic waters: accounting for raman emission and varying particle scattering phase function. Appl. Opt. 41, 6289–6306. doi:10.1364/AO.41.006289

PubMed Abstract | CrossRef Full Text | Google Scholar

Oyama, Y., Matsushita, B., and Fukushima, T. (2015). Distinguishing surface cyanobacterial blooms and aquatic macrophytes using Landsat/TM and ETM + shortwave infrared bands. Remote Sens. Environ. 157, 35–47. doi:10.1016/j.rse.2014.04.031

CrossRef Full Text | Google Scholar

Pahlevan, N., Sarkar, S., Franz, B. A., Balasubramanian, S. V., and He, J. (2017). Sentinel-2 MultiSpectral instrument (MSI) data processing for aquatic science applications: demonstrations and validations. Remote Sens. Environ. 201, 47–56. doi:10.1016/j.rse.2017.08.033

CrossRef Full Text | Google Scholar

Pahlevan, N., Smith, B., Schalles, J., Binding, C., Cao, Z., Ma, R., et al. (2020). Seamless retrievals of chlorophyll-a from Sentinel-2 (MSI) and Sentinel-3 (OLCI) in inland and coastal waters: a machine-learning approach. Remote Sens. Environ. 240, 111604. doi:10.1016/j.rse.2019.111604

CrossRef Full Text | Google Scholar

Pahlevan, N., Mangin, A., Balasubramanian, S. V., Smith, B., Alikas, K., Arai, K., et al. (2021). ACIX-Aqua: a global assessment of atmospheric correction methods for Landsat-8 and Sentinel-2 over lakes, rivers, and coastal waters. Remote Sens. Environ. 258, 112366. doi:10.1016/j.rse.2021.112366

CrossRef Full Text | Google Scholar

Perin, V., Roy, S., Kington, J., Harris, T., Tulbure, M. G., Stone, N., et al. (2021). Monitoring small water bodies using high spatial and temporal resolution analysis ready datasets. Remote Sens. 13, 5176. doi:10.3390/rs13245176

CrossRef Full Text | Google Scholar

Philipson, P., Kratzer, S., Mustapha, S. B., Strömbeck, N., and Stelzer, K. (2016). Satellite-based water quality monitoring in Lake Vänern, Sweden. Int. J. Remote Sens. 37, 3938–3960. doi:10.1080/01431161.2016.1204480

CrossRef Full Text | Google Scholar

Planet Labs, P. B. C. (2023). PlanetScope product specifications. Available online at: https://assets.planet.com/docs/Planet_PSScene_Imagery_Product_Spec_letter_screen.pdf.

Google Scholar

Rouse, J. W., Haas, R. H., Deering, D. W., Schell, J. A., and Harlan, J. C. (1974). Monitoring the vernal advancement and retrogradation (green wave effect) of natural vegetation. Available online at: https://ntrs.nasa.gov/citations/19750020419.

Google Scholar

Vanhellemont, Q. (2019a). Adaptation of the dark spectrum fitting atmospheric correction for aquatic applications of the landsat and Sentinel-2 archives. Remote Sens. Environ. 225, 175–192. doi:10.1016/j.rse.2019.03.010

CrossRef Full Text | Google Scholar

Vanhellemont, Q. (2019b). Daily metre-scale mapping of water turbidity using CubeSat imagery. Opt. Express 27, A1372. doi:10.1364/OE.27.0A1372

PubMed Abstract | CrossRef Full Text | Google Scholar

Vanhellemont, Q. (2020). Sensitivity analysis of the dark spectrum fitting atmospheric correction for metre- and decametre-scale satellite imagery using autonomous hyperspectral radiometry. Opt. Express 28, 29948–29965. doi:10.1364/OE.397456

PubMed Abstract | CrossRef Full Text | Google Scholar

Vanhellemont, Q. (2023). Evaluation of eight band SuperDove imagery for aquatic applications. Opt. Express 31, 13851–13874. doi:10.1364/OE.483418

PubMed Abstract | CrossRef Full Text | Google Scholar

Vanhellemont, Q., and Ruddick, K. (2018). Atmospheric correction of metre-scale optical satellite data for inland and coastal water applications. Remote Sens. Environ. 216, 586–597. doi:10.1016/j.rse.2018.07.015

CrossRef Full Text | Google Scholar

Vanhellemont, Q., and Ruddick, K. (2021). Atmospheric correction of Sentinel-3/OLCI data for mapping of suspended particulate matter and chlorophyll-a concentration in Belgian turbid coastal waters. Remote Sens. Environ. 256, 112284. doi:10.1016/j.rse.2021.112284

CrossRef Full Text | Google Scholar

Vatitsi, K., Siachalou, S., Latinopoulos, D., Kagalou, I., Akratos, C. S., and Mallinis, G. (2024). Monitoring water quality parameters in small Rivers using SuperDove imagery. Water 16, 758. doi:10.3390/w16050758

CrossRef Full Text | Google Scholar

Vermote, E. F., Tanre, D., Deuze, J. L., Herman, M., and Morcette, J.-J. (1997). Second simulation of the satellite signal in the solar spectrum, 6S: an overview. IEEE Trans. Geosci. Remote Sens. 35, 675–686. doi:10.1109/36.581987

CrossRef Full Text | Google Scholar

Werdell, P. J., Bailey, S. W., Franz, B. A., Morel, A., and McClain, C. R. (2007). On-orbit vicarious calibration of ocean color sensors using an ocean surface reflectance model. Appl. Opt. 46, 5649–5666. doi:10.1364/AO.46.005649

PubMed Abstract | CrossRef Full Text | Google Scholar

Yao, Y., Hu, C., Cannizzaro, J. P., Zhang, S., Barnes, B. B., Xie, Y., et al. (2024). Detecting cyanobacterial blooms in the caloosahatchee river and Estuary using PlanetScope imagery and deep learning. IEEE Trans. Geosci. Remote Sens. 62, 1–13. doi:10.1109/TGRS.2024.3354211

CrossRef Full Text | Google Scholar

Zibordi, G., Mélin, F., Berthon, J.-F. F., Holben, B., Slutsker, I., Giles, D., et al. (2009). AERONET-OC: a network for the validation of ocean color primary products. J. Atmos. Ocean. Technol. 26, 1634–1651. doi:10.1175/2009JTECHO654.1

CrossRef Full Text | Google Scholar

Zibordi, G., Holben, B. N., Talone, M., D’Alimonte, D., Slutsker, I., Giles, D. M., et al. (2021). Advances in the ocean color component of the aerosol robotic network (AERONET-OC). J. Atmos. Ocean. Technol. 38, 725–746. doi:10.1175/jtech-d-20-0085.1

CrossRef Full Text | Google Scholar