Standardizing photographic measurements of coral albedo with RAW imaging and calibrated reflectance targets

Abstract

Coral pigment condition is routinely assessed from photographs, yet most commonly used image-based “brightness” metrics are derived from rendered images optimized for visual appearance rather than physical measurement. This limits comparability across cameras, workflows, and studies, and complicates interpretation of photographic signals in terms of intrinsic coral properties. Here we evaluated a simple, reference-based workflow for estimating coral albedo from digital images using unrendered camera measurements scaled to a calibrated reflectance standard. Coral specimens were imaged outdoors under diffuse illumination alongside a Spectralon 99% reflectance target, with paired RAW and JPEG images acquired for each scene. Broadband grayscale albedo estimates derived from these images were copared with independent spectrally derived albedo estimates computed from measured reflectance spectra convolved to the camera response. RAW-derived albedo closely matched the spectrally derived benchmark across coral and a ColorChecker target, whereas JPEG-derived albedo exhibited systematic albedo-dependent positive bias and substantially greater error. A brief sensitivity analysis further shows that the use of uncalibrated “white” reference materials produces predictable inflation in estimated albedo when their true reflectance differs from the assumed value. Additional comparisons across multiple coral species showed that the divergence between rendered and unrendered image-derived albedo estimates was consistent across taxa. Together, these results demonstrate that rendered image products are unsuitable for quantitative albedo estimation and that physically interpretable, repeatable photographic measurements require RAW image acquisition, reference-based normalization, and calibrated reflectance standards.

1 Introduction

Quantifying photosynthetic pigments is central to understanding coral physiological condition and ecological performance. In symbiotic reef corals, pigment content within Symbiodiniaceae varies in response to ambient light regimes (Iglesias-Prieto and Trench, 1997) and is influenced by the skeletal architecture and tissue structure of the host (Scheufen et al., 2017). Concentrations of key pigments, especially chlorophyll a and peridinin, are strongly linked to photosynthetic potential and therefore to the proportion of metabolic energy derived from autotrophy (Muscatine et al., 1981; Falkowski et al., 1984). Changes in environmental conditions can alter pigment abundance, intracellular pigment packaging, and symbiont population density (Brown et al., 1999; Apprill et al., 2007). Under acute stress, most notably during thermal anomalies, disruption of the coral-symbiont association can occur, resulting in substantial losses of symbionts and associated pigments (van Woesik et al., 2022).

Direct determination of coral pigment content has traditionally relied on spectrophotometric methods (e.g., Iglesias-Prieto and Trench, 1997; Nakamura et al., 2005; Krämer et al., 2022; Gantt et al., 2023) and high-performance liquid chromatography (e.g., Venn et al., 2006; Apprill et al., 2007; Torres-Pérez et al., 2015; Rosic et al., 2024). These approaches are time-intensive and depend on precise laboratory procedures involving tissue removal and chemical extraction. As a consequence, they necessitate destructive sampling of coral colonies. Such practical constraints have limited the feasibility of routine or large-scale pigment measurements. Higher-technology reflectance spectroscopy provides a powerful non-destructive means of estimating pigment concentrations (Hochberg et al., 2025), but its technical and logistical demands have prevented it from becoming broadly accessible for widespread monitoring applications.

Motivated by the need for scalable monitoring tools, and enabled by advances in technology, there has been increasing interest in adopting digital photography to assess coral pigmentation. For example, in developing the CoralWatch Coral Health Chart, Siebeck et al. (2006) related photographically derived hue-saturation-brightness values to relative chlorophyll a concentrations and zooxanthellae densities. Recently, Ferrara et al. (2026) expanded on the concept, exploring abilities of different combinations of red-green-blue (RGB; Table 1 lists all symbols and abbreviations in this paper) image channels to quantify those same parameters. Another common approach is to collapse the RGB image into a single brightness or grayscale channel (Winters et al., 2009; Chow et al., 2016; Amid et al., 2018; McLachlan and Grottoli, 2021). These approaches have proven useful for evaluating coral condition in their respective case studies. Because they can also be implemented with widely available image analysis software, they present a relatively low technical barrier and can be applied efficiently across larger datasets (McLachlan and Grottoli, 2021).

A central implementation problem in photographic assessments of coral pigmentation is that standard camera-produced images are designed for human viewing rather than physical measurement. Human vision does not perceive brightness linearly with radiant intensity (Gilchrist et al., 1999). Instead, perceived differences are weighted unevenly across the light intensity range, with greater sensitivity to darker tones than to brighter ones. Consumer digital cameras are built around this perceptual framework (Sharma and Trussell, 1997). First, they sample reflected light in three broad RGB channels chosen to approximate human cone sensitivities (Süsstrunk et al., 1999). The recorded sensor responses are then typically rendered into a standard output color space by applying a non-linear transfer function defined for image display and reproduction (Süsstrunk et al., 1999). In practice, this rendering commonly includes gamma encoding, white balancing, tone mapping, contrast adjustment, and related steps that reshape the sensor’s linear signal so that the final image appears visually natural to a human observer rather than remaining proportional to scene radiance or surface reflectance. Consequently, the pixel values in a standard digital photograph already contain an embedded perceptual bias: They are numerical representations of a visually rendered scene. Any metrics calculated from such rendered RGB data are themselves derived from an already non-linear representation and therefore non-linear with respect to reflected light. This is the core implementation problem for coral applications, because any biological interpretation of such metrics is then confounded by the camera’s perceptual rendering pipeline. This is further exacerbated by additional downstream nonlinear image processing, for example white balance adjustments (e.g., Ferrara et al., 2026).

A straightforward way to avoid these perceptual and encoding biases is to operate on unrendered image data (Stevens et al., 2007). At the sensor level, digital cameras respond approximately linearly to incident light intensity (Jiang et al., 2013), such that recorded digital counts are proportional to scene radiance. Many commercial-off-the-shelf (COTS) cameras offer the ability to save these linear data in a raw image format, often denoted “RAW”. Metrics derived from these linear data therefore have the potential to track biologically meaningful variation in coral optical properties more directly than rendered image values. However, the measured signal still depends on illumination, which can vary across scenes, time, and imaging conditions. As a result, pixel values alone do not represent an intrinsic property of the coral surface. To isolate the coral signal, it must be referenced to a stable, calibrated reflectance target within the scene. By scaling the coral response relative to a known standard, the measurement is normalized with respect to incident light, yielding a quantity that reflects the fraction of light returned from the coral surface rather than the absolute brightness of the image. In this form, the measurement becomes independent of viewing and illumination conditions and instead represents an intrinsic optical property of the coral.

This framing aligns with established coral optics approaches that use reflectance, often measured spectrally, to quantify coral condition and benthic composition (Hedley and Mumby, 2002; Hochberg et al., 2003; Yamano and Tamura, 2004; Hochberg et al., 2025). The difference is that digital cameras do not acquire high-resolution (<10 nm-wide) spectral data, but rather RGB images, where the R, G, and B channels are each 50–100 nm wide. In this context, reflectance is defined as the light flux reflected from an object in a given channel, ϕ_r, divided by the light flux incident to the object in the same given channel, ϕ_i. In photographic terms, ϕ_i is the stable, calibrated reflectance target within the scene, and ϕ_r is derived from the coral region of interest. In formal radiometric terms, the quantity measured by a camera is most precisely described as a hemispherical-directional reflectance factor. We propose to use the term albedo because it emphasizes the broadband nature of the metric and is more intuitive for the coral ecology audience. Ultimately, albedo is a physical quantity, bounded between 0 and 1, and directly linked to the optical properties of the coral tissue–skeleton system.

A tangential consideration is the choice of reference target used to estimate ϕ_i. In existing photographic protocols, normalization is typically performed using a visually “white” surface within the scene (e.g., McLachlan and Grottoli, 2021). However, common materials such as paper or plastic are neither spectrally neutral nor radiometrically stable. Further, their reflectance properties can vary substantially among source materials and over time due to manufacturing differences, optical brighteners, and aging or oxidation. As a result, the use of informal “white” references can introduce additional uncontrolled variability between imaging sessions or studies. In contrast, calibrated reflectance standards (e.g., Spectralon) provide well-characterized and temporally stable reflectance properties, enabling consistent scaling of ϕ_r relative to ϕ_i and improving the comparability of derived albedo estimates across deployments and time periods.

The objective of this study is to evaluate how photographic processing and reference selection influence the quantitative interpretation of coral “brightness” metrics commonly used in the literature. We focus specifically on grayscale or “whiteness”-based workflows, which collapse RGB image data into a single intensity channel and are widely applied due to their simplicity and scalability (Winters et al., 2009; Chow et al., 2016; Amid et al., 2018; McLachlan and Grottoli, 2021; Reichert et al., 2021). Using a standardized imaging protocol, we compare results derived from rendered (JPEG) and unrendered (RAW) image data to assess the impact of non-linear camera processing on inferred coral albedo. In addition, we quantify the sensitivity of normalized measurements to the choice of reference target by evaluating how deviations from an assumed reflectance standard propagate into derived values. Through these analyses, we aim to determine whether commonly applied photographic approaches yield physically interpretable metrics and to establish a framework for deriving more robust, repeatable estimates of coral optical properties from digital images.

2 Methods

2.1 Coral specimens

This study included 19 coral specimens, all identified as Montastraea cavernosa. Specimens were approximately 5 cm × 5 cm and were selected to provide a consistent set of coral surfaces for comparison with spectrally derived estimates. 17 appeared (visually) fully pigmented; one exhibited paling in a very small (<1 cm²) edge area; and one had distinct pigmented and pale halves, which were treated as effectively separate specimens in the data analysis.

2.2 Image collection

All imaging was conducted outdoors under natural daylight. A translucent white sheet was suspended above and around the collection area to diffuse illumination and reduce glare artifacts. Individual coral specimens were placed on a white waterproof paper background (Rite in the Rain DuraCopy synthetic paper) and imaged alongside a calibrated Spectralon 99% reflectance target (Figure 1).

Figure 1

All images were collected using a Nikon D5500 digital single-lens reflex (DSLR) camera equipped with a 10.0–24.0 mm f/3.5–4.5 lens set to a 17 mm focal length. The D5500 used a 23.5 × 15.6 mm complementary metal-oxide-semiconductor (CMOS) image sensor (24.78 MP) and was operated at its maximum spatial resolution (6000 x 4000 pixels). Images were recorded simultaneously in RAW and JPEG formats using the camera’s NEF (RAW) + JPEG Fine setting. RAW images were stored as 14-bit compressed NEF files, and JPEGs were stored using the JPEG-baseline standard with Fine compression (approximately 1:4). Recording both formats for each scene enabled direct comparison of albedo estimates derived from linear sensor data versus imagery rendered through the camera’s JPEG processing pipeline.

2.3 Image processing

Rendered JPEG images were processed explicitly following the workflow described by McLachlan and Grottoli (2021) to reflect common practice in coral brightness assessments. JPEG Fine images were analyzed in ImageJ, where each image was converted to 8-bit grayscale by averaging the R, G, and B channel values for each pixel.

Unrendered RAW (NEF format) images were read into MATLAB using the rawread function, which returns an unsigned 16-bit (D5500 pixel values are 14-bit), 2-dimensional color filter array (CFA) image. The rawread function also applies the required reversal of nonlinear range compression implemented by the D5500 when saving NEF files. (Many cameras apply similar range compressions when saving RAW images; their linearization tables are stored in image EXIF metadata.) Manufacturer-provided electrical dark levels (600 digital counts; stored in EXIF metadata) were subtracted from the CFA image, which was then demosaicked using MATLAB’s demosaic function. This function applies gradient-corrected linear interpolation to convert a two-dimensional Bayer CFA image into a truecolor image. The demosaicked result was an RGB image with pixel values in unsigned 16-bit digital numbers (DNs). Critically, the DNs followed the linear response of the CMOS sensor. (This image had pixel dimensions 6016 × 4016, and so was cropped to match the JPEG image dimensions. This crop was determined empirically by overlaying a test JPEG on its corresponding RAW version.) The RAW DNs were averaged across the RGB channels to form a 16-bit grayscale image, matching the JPEG processing step. This grayscale image was then exported as an unsigned 16-bit TIFF file, suitable for input to ImageJ.

For each coral specimen, 10 regions of interest (ROIs) were manually drawn using the ellipse tool in ImageJ over interior tissue regions. ROIs were drawn explicitly avoiding glare, edge effects, and localized surface irregularities, while also capturing as many different areas of the specimen as possible. For the single specimen with pigmented and pale halves, only five ROIs were selected in each half. A single ROI was defined over the Spectralon. Median ROI size was 1.1% of coral specimen area, with an interquartile range of 0.9%–1.5%. Thus, within a specimen, ROIs had comparable sizes and collectively covered ~10% of the coral area.

Again using ImageJ, mean grayscale values were extracted for each ROI from both the rendered JPG and unrendered RAW grayscale images. Identical ROIs were used for both the rendered and unrendered images, ensuring that differences between JPEG and RAW reflectance estimates were attributable to image encoding and processing rather than spatial sampling.

We defined coral albedo, α^coral, as broadband (grayscale) light flux reflected from the coral, , divided by broadband light flux incident to the coral, . The light flux reflected from the Spectralon reference, , divided by 0.99 to compensate for the Spectralon’s known reflectance, was used as the estimate for , such that

This value was calculated for each ROI in a coral specimen, then the resulting ROI-level values were averaged to provide an overall α^coral for the specimen. These calculations were applied to both JPEG and RAW data sets, providing and .

2.4 Spectral reflectance

To provide an independent reference for evaluating camera-derived albedo estimates, we measured spectral reflectance ρ(λ), where λ represents wavelength, for each of the targets. Flux spectra were acquired for coral specimens, , and the Spectralon reference target, , using a fiber-optic cable (ThorLabs FG550LEC, 550 µm diameter, 0.22 NA) attached to a handheld spectrometer (Analytical Spectral Devices FieldSpec HandHeld Pro-2). For each coral specimen, five replicate measurements were collected from the Spectralon reference, and ten measurements were collected at haphazardly selected locations across the coral surface. Selection was informal rather than formally randomized, because measurement locations were chosen to avoid glare and obvious artifacts, with the aim of capturing colony-scale variability. To approximate camera viewing geometry, spectral measurements were acquired at nadir, but care was taken to avoid self-shading. Observation distance was ~2 cm. Spectral reflectance was calculated as

Spectral measurements were collected immediately following image acquisition to maintain consistency in illumination and specimen condition between the two observation modalities.

We simulated from ρ(λ). First, the D5500 relative spectral response curves (RSRs, see ColorChecker Measurements below) were applied to ρ(λ) to calculate convolved reflectance for each camera channel:

where S_c(λ) is the RSR for channel c, c ∈{R,G,B}, and the integral is taken over the wavelength range λ = [400-700]. Then, following the same procedure as for image-derived α, we calculated the mean of {ρ_R,ρ_G,ρ_B} (analogous to averaging across image channels), and finally calculated the mean of the 10 individual point measurements (analogous to averaging ROIs) to find .

2.5 ColorChecker measurements

To supplement the coral measurements and extend the range of surface reflectances considered, additional imaging and spectral measurements were conducted using a Calibrite ColorChecker Classic target. The ColorChecker provided 24 standardized color patches spanning a broad range of brightness and chromaticity, including six neutral gray levels. It was used for two purposes: (1) to evaluate camera-derived albedo estimates across a wider dynamic range than that represented by the relatively dark coral specimens alone, and (2) to enable estimation of the camera’s effective relative spectral response functions (RSRs) by pairing known reflectance spectra with measured RGB values. The ColorChecker image (both RAW and JPEG Fine) was acquired outdoors under natural, diffuse illumination provided by an overcast sky and included the Spectralon reference target within the field of view. For each panel in the ColorChecker, we calculated for each channel c by normalizing the mean RAW digital number of the panel to the mean digital number of the Spectralon reference target, then multiplying by 0.99.

Corresponding reflectance spectra were collected indoors using a ThorLabs OSL2 Fiber Illuminator coupled to a bifurcated fiber-optic cable (ThorLabs BF19Y2HSO2, 200 µm diameter, 0.22 NA, 250–1200 nm) and an Ocean Optics USB2000+ spectrometer. During spectral acquisition, the fiber-optic collection tip was held fixed using a laboratory stand, and the ColorChecker and Spectralon targets were translated beneath it to maintain constant illumination and viewing geometry. We calculated for each panel in the same manner as for the corals.

The paired and measurements were used to estimate the camera’s effective RSRs using nonlinear least-squares optimization (lsqnonlin in MATLAB). For each RGB channel independently, the RSR was parameterized as a mixture of Gaussian functions, and the optimizer adjusted these parameters to minimize the mismatch between camera-derived and predicted from . The fitted RSRs represented the spectral weighting functions S_c that best mapped ρ(λ) to camera RGB reflectance space. These fitted functions did not recover the camera’s true spectral sensitivities but instead provided effective response functions sufficient to map between measured reflectance spectra and camera-band reflectance for the purposes of spectral-derived albedo estimation.

We calculated albedo values for the ColorChecker panels, , , and , following the same procedures used for the corals. The final dataset of 20 values for {, , } and 24 values for {, , } provided a consistent framework for evaluation of the influence of file format and encoding on quantitative albedo estimation.

2.6 Simple albedo simulation

Finally, to assess the impact of non-standard/uncalibrated reference targets, we performed a simple sensitivity analysis. For true coral α = {0.05, 0.06, 0.07,…, 0.15}, we calculated the α that would be estimated if using a reference with ρ = {1, 0.99, 0.98, …, 0.9}, but assuming reference ρ = 1. That is, we calculated α without compensating for variability in reference ρ.

3 Results

Camera-derived albedo estimates from RAW imagery closely tracked spectrally derived albedo across both ColorChecker panels and coral specimens (Figure 2). RAW-based albedo values clustered near the 1:1 line, indicating good agreement with spectroscopic estimates over the full dynamic range examined (Figure 2A). In contrast, albedo derived from JPEG images systematically deviated from spectroscopic values. Bias emerged rapidly even at low albedo values, corresponding to the range occupied by most coral specimens. In addition, JPEG-derived albedo estimates exhibited substantially greater variability among coral specimens than RAW-derived estimates, indicating reduced precision as well as reduced accuracy. RAW-derived albedo exhibited small, approximately zero-centered errors across the observed range, whereas JPEG-derived errors were already substantial at low true albedo (Figure 2B). These patterns were observed for both the ColorChecker and coral datasets, demonstrating that RAW-based processing preserves a near-linear relationship with spectrally derived albedo, while JPEG-based processing introduces pronounced, albedo-dependent distortion.

Figure 2

Modeled coral albedo estimates varied systematically with the assumed reflectance of the reference target (Figure 3). When the reference target reflectance was correctly specified as ρ = 1, estimated coral albedo matched the true value, whereas decreasing reference reflectance produced progressively larger overestimation of coral albedo when this deviation was not accounted for (Figure 3A). This bias occurred across all simulated coral albedos, indicating that uncertainty or variability in reference target reflectance can introduce systematic error even when the coral signal itself is unchanged. The magnitude of this error depended on both reference reflectance and coral albedo: errors increased as reference reflectance decreased, and errors were larger in absolute terms for higher-albedo corals (Figure 3B). Together, these results demonstrate that use of non-ideal or uncharacterized reference materials can lead to predictable, albedo-dependent inflation of estimated coral albedo.

Figure 3

4 Discussion

Taken together, our results demonstrate that rendered imagery is unsuitable for quantitative estimates of coral albedo. Rendering pipelines are designed to map sensor data into perceptually meaningful color spaces, applying non-linear transformations that distort the relationship between pixel values and reflected light. Consequently, common image formats that store rendered data (e.g., JPEG, PNG, BMP, and most TIFF encodings) introduce systematic bias into albedo estimates. This bias arises because intensity values are redistributed in a brightness-dependent manner, such that equal changes in albedo are not preserved across the camera’s dynamic range (Stevens et al., 2007; Jiang et al., 2013). The systematic divergence of JPEG-derived albedo estimates from the 1:1 line, even at low values (Figure 2), is consistent with these non-linear transformations, which are optimized for visual appearance rather than physical proportionality.

In contrast, operating on unrendered RAW image data retains the sensor’s approximately linear response to incident radiance (Jiang et al., 2013), allowing pixel values to scale proportionally with reflected light. The close agreement between RAW-derived estimates and spectrometer-derived albedo across targets (Figure 2) demonstrates that this approach supports quantitative albedo estimation without relying on perceptual color encoding. As a result, COTS cameras can be used as broadband, multi-channel sensors to derive physically interpretable albedo metrics when appropriate workflows are applied. This provides a practical balance between rigor and accessibility by leveraging widely available imaging systems while avoiding the distortions introduced by rendering.

The implications of rendering bias are non-trivial. JPEG-based “brightness” metrics systematically exaggerate subtle albedo changes (Figure 2), particularly across the range from normally pigmented (low albedo) to partially depigmented (mid-level albedo) corals. In physiological terms, this non-linearity inflates the apparent magnitude of change in coral condition, potentially overstating the severity or rate of pigment loss when interpreted quantitatively. This interpretation bias is substantially reduced when analyses are conducted on unrendered RAW image data, where the linear relationship between pixel values and reflected light is preserved.

A practical limitation is that the albedo metric implemented here is defined as an equal-weight average across the three camera channels rather than a spectrally weighted integral over channel bandwidths and response functions. For , the RGB DNs were averaged directly after reference scaling, and for , the convolved channel reflectances were likewise averaged across R, G, and B so that the spectrally-derived and camera-derived estimates were defined on the same basis. This approach was intentional, as it produced a broadband grayscale metric directly analogous to the JPEG-based workflow for while avoiding additional camera-specific weighting assumptions. As such, the resulting albedo estimates should be interpreted as a practical broadband reflectance metric rather than a fully spectrally resolved quantity. Future work could refine this framework by incorporating channel-specific spectral weighting, but that level of radiometric customization is beyond the scope of the generally applicable protocol developed here.

A primary limitation of this evaluation is that it used a single camera system (Nikon D5500), which may constrain generalization to all COTS imaging platforms. However, many DSLR and mirrorless cameras share similar CMOS sensor architectures and approximately linear sensor responses. They also share similar image-processing pipelines. As a result, the direction of the rendering-induced bias and the benefits of preserving unrendered RAW data are expected to be consistent across systems (Jiang et al., 2013). The magnitude of rendering bias may vary with camera model, settings, and processing parameters, reinforcing the importance of reporting full acquisition details. However, the underlying distinction between rendered and unrendered data is inherent to digital imaging systems and is not specific to a single instrument.

A key practical consideration of the workflow is the requirement for a calibrated reflectance target within each scene. Accurate albedo estimation depends on scaling the coral signal relative to a reference of known reflectance, such that variability in illumination is properly constrained. As demonstrated in the sensitivity analysis (Figure 3), deviations between the assumed and actual reflectance of the reference target introduce systematic, albedo-dependent bias in the estimated values. This effect is not specific to the present method but is inherent to any reference-based normalization approach. Accordingly, reference targets should be spectrally stable, calibrated, and deployed under consistent imaging geometry. Materials such as Spectralon satisfy these criteria but introduce practical constraints, including handling, fouling, and positioning within the scene. While these considerations may complicate routine field deployment, they do not preclude adoption of RAW-based workflows; rather, they represent necessary elements of experimental design to ensure quantitative accuracy and comparability.

An additional limitation is that camera-derived albedo represents view-dependent reflectance scaled by a reference target, rather than true hemispherical-directional albedo in a radiative-transfer sense. Coral tissues and skeletons exhibit anisotropic scattering due to microtopography, skeletal architecture, and internal refractive index contrasts. As a result, measured reflectance (and thus estimated albedo) may vary with illumination and viewing geometry. However, under consistent imaging conditions, the reference-based RAW workflow provides a stable and repeatable estimate of broadband albedo for a given specimen. In this context, the primary limitation is not in the measurement of albedo itself, but in its interpretation across taxa and morphologies, where differences in tissue structure, skeletal scattering, and optical path length may influence how albedo relates to underlying physiological properties. In this study, geometric effects were minimized by maintaining consistent camera positioning and illumination, and the inclusion of the ColorChecker target supports the robustness of the workflow across a broader range of surface reflectances. Future applications should consider these factors when interpreting albedo values among different coral types or imaging configurations.

A further point of clarification is the role of camera characterization in quantitative imaging. Each model of COTS RGB camera has its own spectral response function. The reference-based workflow applied here mitigates that factor by expressing coral reflectance relative to a co-imaged standard under near-identical conditions. Because both the coral and reference target are subject to functionally the same sensor response and optical pathway, the referencing approach provides a robust estimate of relative albedo without requiring explicit calibration of spectral sensitivity.

Spatial non-uniformities (e.g., vignetting) within the camera lens-sensor system are also of concern. These may add small and unequal constant offsets to ϕcoral and ϕref, which in turn can lead to errors in calculated αcoral. This study benefits from utilizing pixels in the central region of the images and achieves accurate αcoral without explicit spatial compensations. Best practice would be to avoid pixels near image edges, especially corners. Alternatively, the practitioner can create a camera model to enable explicit vignetting corrections. However, residual uncertainties that are not uniform across the sensor or stable under all conditions, such as dark current variation from the manufacturer’s reported value, are generally small relative to the signal magnitude and largely cancel in the normalized measurement. In this context, COTS cameras function effectively as broadband, multi-channel sensors for quantitative albedo estimation, without requiring explicit correction for sensor-specific characteristics. The accuracy of the approach derives from preserving linear sensor response and applying consistent normalization, rather than from detailed characterization of the imaging system itself.

Given these considerations, to evaluate the generality of this pattern across coral taxa, we extended the comparison of rendered and unrendered albedo estimates to additional species (Figure 4). While spectroscopic validation was limited to M. cavernosa, photographic comparisons across multiple species—Diploria labyrinthiformis, Pseudodiploria strigosa, Porites astreoides, and Orbicella franksi—show that the relationship between JPEG-derived and RAW-derived albedo is consistent across taxa. Specifically, all species follow the same systematic deviation from the 1:1 relationship observed in Figure 2, with JPEG-derived values increasingly overestimating albedo relative to RAW-derived estimates across the observed range. The subset of M. cavernosa samples for which spectroscopic measurements were available (black crosses) fall along the same trend, providing an anchor between the cross-species photographic dataset and the spectrally validated results. Together, these observations indicate that the rendering-induced bias is not species-specific, but instead reflects a general property of image encoding, and that RAW-based workflows provide consistent albedo estimates across a range of coral morphologies and optical characteristics.

Figure 4

The cross-taxon consistency shown in Figure 4 indicates that the bias introduced by rendered imagery arises from the imaging pipeline itself rather than from species-specific coral optics. This distinction is especially important for underwater applications, where the spectral composition of light reaching the camera is further altered by the water column. Preferential absorption of longer wavelengths reduces red-channel signal with increasing depth, such that rendering pipelines optimized for visual appearance in air may further rescale or suppress this already diminished information. Under these conditions, rendered imagery can introduce additional, environment-dependent distortion into the relationship between recorded pixel values and coral reflectance. In contrast, RAW imagery preserves the underlying sensor response across channels, retaining information that can be appropriately normalized and interpreted. RAW-based workflows are therefore likely to be particularly important for quantitative underwater imaging, where both water-column spectral filtering and in-camera rendering can otherwise obscure biologically meaningful optical variation.

The implications of this study’s findings extend to how photographic metrics of coral condition are interpreted. Previous studies have demonstrated that substantial declines in symbiont abundance and pigmentation can occur prior to visible bleaching (Fitt et al., 2000), indicating that image-based metrics may detect changes not readily apparent to the human eye. In this context, JPEG-derived “whiteness” or brightness measures may appear sensitive to early physiological change. However, because these metrics are derived from non-linearly rendered imagery, the relationship between pixel values and underlying reflectance—and therefore coral physiology—is inherently distorted and not directly comparable across imaging conditions or studies. In contrast, spectroscopic approaches have shown that linearly scaled reflectance provides a robust basis for relating optical signals to coral pigmentation and physiological state. The RAW-based albedo framework presented here extends this principle to photographic data, providing a physically grounded metric that preserves proportionality with reflected light. As a result, previously reported relationships between image-derived brightness and coral condition should be interpreted with caution and, where possible, revisited using RAW-based workflows to establish consistent and transferable links between optical measurements and coral physiology.

Together, these results define a practical workflow for quantitative photographic assessment of coral reflectance: image acquisition in RAW format, inclusion of a stable, calibrated reflectance target in each scene, and processing that preserves linear sensor response prior to conversion to broadband albedo. Under these conditions, consumer-grade cameras can function as broadband, multi-channel sensors to produce physically interpretable and repeatable albedo estimates, rather than images optimized solely for visual appearance. Adoption of such workflows would improve sensitivity to early pigment change, reduce systematic bias associated with perceptual image encoding, and enhance comparability across studies, sites, and monitoring programs. Ultimately, this work shows that photographic coral assessment becomes quantitatively reliable only when image rendering is bypassed and the optical signal is preserved in unrendered, linear, reference-normalized form.

References

• 1

  AmidC.OlstedtM.GunnarssonJ. S.Le LanH.Tran Thi MinhH.Van den BrinkP.et al. (2018). Additive effects of the herbicide glyphosate and elevated temperature on the branched coral Acropora formosa in Nha Trang, Vietnam. Environ. Sci. pollut. Res.25, 13360–13372. doi: 10.1007/s11356-016-8320-7

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 2

  ApprillA. M.BidigareR. R.GatesR. D. (2007). Visibly healthy corals exhibit variable pigment concentrations and symbiont phenotypes. Coral Reefs26, 387–397. doi: 10.1007/s00338-007-0209-y

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 3

  BrownB. E.DunneR. P.AmbarsariI.Le TissierM. D. A.SatapoominU. (1999). Seasonal fluctuations in environmental factors and variations in symbiotic algae and chlorophyll pigments in four Indo-Pacific coral species. Mar. Ecol. Prog. Ser.191, 53–69. doi: 10.3354/meps191053

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 4

  ChowM.TsangR. H.LamE. K.AngP. (2016). Quantifying the degree of coral bleaching using digital photographic technique. J. Exp. Mar. Biol. Ecol.479, 60–68. doi: 10.1016/j.jembe.2016.03.003

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 5

  FalkowskiP. G.DubinskyZ.MuscatineL.PorterJ. W. (1984). Light and the bioenergetics of a symbiotic coral. Bioscience34, 705–709. doi: 10.2307/1309663

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 6

  FerraraE. F.BauerL.PuntinG.BautzF.CelayirS.DoM. S.et al. (2026). Red, green, blue color indices as proxy for Symbiodiniaceae cell density and chlorophyll content during coral bleaching. Limnology Oceanography: Methods24, e70047. doi: 10.1002/lom3.70047

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 7

  FittW. K.McFarlandF. K.WarnerM. E.ChilcoatG. C. (2000). Seasonal patterns of tissue biomass and densities of symbiotic dinoflagellates in reef corals and relation to coral bleaching. Limnol. Oceanogr.45, 677–685. doi: 10.4319/lo.2000.45.3.0677

  • CrossRef
  • Google Scholar
  • View reference in article

• 8

  GanttS. E.KeisterE. F.ManfroyA. A.MerckD. E.FittW. K.MullerE. M.et al. (2023). Wild and nursery-raised corals: comparative physiology of two framework coral species. Coral Reefs42, 299–310. doi: 10.1007/s00338-022-02333-9

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 9

  GilchristA.KossyfidisC.BonatoF.AgostiniT.CataliottiJ.LiX.et al. (1999). An anchoring theory of lightness perception. psychol. Rev.106, 795. doi: 10.1007/978-3-030-89862-5_263

  • CrossRef
  • Google Scholar
  • View reference in article

• 10

  HedleyJ. D.MumbyP. J. (2002). “ Biological and remote sensing perspectives of pigmentation in coral reef organisms,” in Advances in Marine Biology. Eds. SouthwardA. J.YoungC. M.FuimanL. A. ( Academic Press, San Diego), 279–317.

  • Google Scholar
  • View reference in article

• 11

  HochbergE. J.AtkinsonM. J.AndréfouëtS. (2003). Spectral reflectance of coral reef bottom-types worldwide and implications for coral reef remote sensing. Remote Sens. Environ.85, 159–173. doi: 10.1016/s0034-4257(02)00201-8

  • CrossRef
  • Google Scholar
  • View reference in article

• 12

  HochbergE. J.DucretH.DierssenH. M. (2025). Revised empirical models for estimating coral pigment concentrations from optical reflectance spectra. Appl. Opt.64, 10603–10610. doi: 10.1364/ao.578421

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 13

  Iglesias-PrietoR.TrenchR. K. (1997). Acclimation and adaptation to irradiance in symbiotic dinoflagellates. 2. Response of chlorophyll-protein complexes to different photon-flux densities. Mar. Biol.130, 23–33. doi: 10.1007/s002270050221

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 14

  JiangJ.LiuD.GuJ.SüsstrunkS. (2013). “ What is the space of spectral sensitivity functions for digital color cameras?” In: 2013 IEEE Workshop on Applications of Computer Vision (WACV) ( IEEE), 168–179.

  • Google Scholar
  • View reference in article

• 15

  KrämerW. E.Iglesias-PrietoR.EnríquezS. (2022). Evaluation of the current understanding of the impact of climate change on coral physiology after three decades of experimental research. Commun. Biol.5, 1418. doi: 10.1038/s42003-022-04353-1

  • CrossRef
  • Google Scholar
  • View reference in article

• 16

  McLachlanR.GrottoliA. G. (2021). protocols.io.

  • Google Scholar
  • View reference in article

• 17

  MuscatineL.McCloskeyR. L.MarianE. R. (1981). Estimating the daily contribution of carbon from zooxanthellae to coral animal respiration 1. Limnol. Oceanogr.26, 601–611. doi: 10.4319/lo.1981.26.4.0601

  • CrossRef
  • Google Scholar
  • View reference in article

• 18

  NakamuraT.Van WoesikR.YamasakiH. (2005). Photoinhibition of photosynthesis is reduced by water flow in the reef-building coral Acropora digitifera. Mar. Ecol. Prog. Ser.301, 109–118. doi: 10.3354/meps301109

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 19

  ReichertJ.TirpitzV.AnandR.BachK.KnoppJ.SchubertP.et al. (2021). Interactive effects of microplastic pollution and heat stress on reef-building corals. Environ. pollut.290, 118010. doi: 10.1016/j.envpol.2021.118010

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 20

  RosicN.Delamare-DebouttevilleJ.DoveS. (2024). Heat stress in symbiotic dinoflagellates: Implications on oxidative stress and cellular changes. Sci. Total Environ.944, 173916. doi: 10.1016/j.scitotenv.2024.173916

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 21

  ScheufenT.Iglesias-PrietoR.EnríquezS. (2017). Changes in the number of symbionts and Symbiodinium cell pigmentation modulate differentially coral light absorption and photosynthetic performance. Front. Mar. Sci.4, 309. doi: 10.3389/fmars.2017.00309

  • CrossRef
  • Google Scholar
  • View reference in article

• 22

  SharmaG.TrussellH. J. (1997). Digital color imaging. IEEE Trans. Image Process.6, 901–932. doi: 10.1109/83.597268

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 23

  SiebeckU. E.MarshallN. J.KluterA.Hoegh-GuldbergO. (2006). Monitoring coral bleaching using a colour reference card. Coral Reefs25, 453–460. doi: 10.1007/s00338-006-0123-8

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 24

  StevensM.PárragaC. A.CuthillI. C.PartridgeJ. C.TrosciankoT. S. (2007). Using digital photography to study animal coloration. Biol. J. Linn. Soc90, 211–237. doi: 10.1111/j.1095-8312.2007.00725.x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 25

  SüsstrunkS.BuckleyR.SwenS. (1999). “ Standard RGB color spaces”. In: Color and Imaging Conference ( Society of Imaging Science and Technology) 7, 127–134.

  • Google Scholar
  • View reference in article

• 26

  Torres-PérezJ. L.GuildL. S.ArmstrongR. A.CorredorJ.Zuluaga-MonteroA.PolancoR. (2015). Relative pigment composition and remote sensing reflectance of Caribbean shallow-water corals. PloS One10, e0143709. doi: 10.1371/journal.pone.0143709

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 27

  van WoesikR.ShlesingerT.GrottoliA. G.ToonenR. J.Vega ThurberR.WarnerM. E.et al. (2022). Coral‐bleaching responses to climate change across biological scales. Global Change Biol.28, 4229–4250. doi: 10.1111/gcb.16192

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 28

  VennA. A.WilsonM. A.Trapido‐RosenthalH. G.KeelyB. J.DouglasA. E. (2006). The impact of coral bleaching on the pigment profile of the symbiotic alga, Symbiodinium. Plant Cell Environ.29, 2133–2142. doi: 10.1111/j.1365-3040.2006.001587.x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 29

  WintersG.HolzmanR.BlekhmanA.BeerS.LoyaY. (2009). Photographic assessment of coral chlorophyll contents: Implications for ecophysiological studies and coral monitoring. J. Exp. Mar. Biol. Ecol.380, 25–35. doi: 10.1016/j.jembe.2009.09.004

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 30

  YamanoH.TamuraM. (2004). Detection limits of coral reef bleaching by satellite remote sensing: Simulation and data analysis. Remote Sens. Environ.90, 86–103. doi: 10.1016/j.rse.2003.12.005

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article