Dissolved organic carbon and coloured dissolved organic matter in inland waters

CDOM shifts the water-leaving reflectance spectrum in measurable ways, letting satellites estimate dissolved organic carbon export from peatlands and forests. Retrieval accuracy depends heavily on separating CDOM from sediment and phytoplankton signals.

Sensors

Sentinel-2 MSI: 10 m (visible) and 20 m (red-edge, SWIR) spatial resolution, 5-day revisit at mid-latitudes with two satellites. Bands B2 (490 nm) and B3 (560 nm) are the primary CDOM retrieval channels; the B2/B3 ratio and band-difference algorithms are well-documented in peer-reviewed literature. Atmospheric correction over dark inland water remains the principal source of error.
Landsat 8/9 OLI: 30 m resolution, 16-day single-satellite revisit (8-day combined). The coastal/aerosol band (Band 1, 435–451 nm) gives OLI a spectral handle closer to the CDOM absorption peak than MSI, though signal-to-noise over dark water is limited. Archive extends to 1984 (TM/ETM+), enabling multi-decade DOC trend analysis where empirical models are carefully recalibrated.
PACE OCI: Launched February 2024. Hyperspectral (340–890 nm at 5 nm resolution) with daily global coverage at 1 km. Purpose-built ocean colour radiometry with a UV-to-NIR range that captures the CDOM absorption slope directly. Spatial resolution limits application to larger lakes and river plumes, but spectral richness allows semi-analytical inversion that multispectral sensors cannot support.
DESIS (ISS hyperspectral): 30 m resolution, 400–1000 nm at roughly 2.55 nm spectral sampling. Tasked acquisitions from the ISS, so revisit is irregular (days to weeks depending on orbital precession). Useful for site-specific calibration and validation of coarser CDOM products; not suitable for operational monitoring at scale.

What CDOM absorption actually measures, and what it does not

Coloured dissolved organic matter is the light-absorbing fraction of dissolved organic carbon. It absorbs exponentially toward shorter wavelengths, with an absorption coefficient at 440 nm (aCDOM(440)) commonly used as the standard reporting metric. The spectral slope S (typically 0.010–0.020 nm⁻¹ across 275–295 nm) carries information about molecular weight and photodegradation state. Crucially, CDOM and DOC are correlated but not synonymous. In peat-draining catchments the relationship is tight; in systems receiving autochthonous production or wastewater, it breaks down. Any satellite-derived DOC figure is therefore a CDOM-to-DOC conversion, not a direct carbon measurement, and that conversion must be validated locally.

Satellites measure water-leaving radiance, which is a small residual after the atmosphere has absorbed and scattered most of the signal. Over dark inland water, the atmospheric path radiance can exceed the water-leaving signal by an order of magnitude. This is not a minor caveat; it is the central challenge. Atmospheric correction algorithms designed for open ocean (such as the standard NASA approach using near-infrared dark-pixel assumptions) fail over turbid or CDOM-rich water. Inland-water corrections using SWIR bands or iterative schemes (ACOLITE, iCOR, C2RCC) reduce the error but do not eliminate it.

Retrieval methods: from band ratios to semi-analytical inversion

Empirical approaches dominate operational use. The most widely applied are simple band-ratio or band-difference algorithms using blue and green reflectance. For Sentinel-2, ratios such as Rrs(490)/Rrs(560) correlate with aCDOM(440) across many lake types, with published root-mean-square errors in the range of 0.2–0.5 m⁻¹ depending on the training dataset and water type. These algorithms are fast and transferable within a region, but they are not universal: a model trained on Finnish boreal lakes will perform poorly on turbid glacial lakes or subtropical reservoirs without recalibration.

Semi-analytical models (QAA, GSM, GIOP) decompose the total absorption and backscattering signal into contributions from phytoplankton, CDOM, and non-algal particles. They require no local training data in principle, but they carry their own assumptions about spectral shapes and particle backscattering coefficients. PACE OCI's hyperspectral coverage makes semi-analytical inversion far more tractable than it is with four or five multispectral bands, because the full shape of the absorption spectrum is observable rather than inferred. For Sentinel-2 and Landsat, the limited number of visible bands means that CDOM, non-algal particles, and phytoplankton absorption are partially confounded, particularly when all three co-vary, which they do in many productive or turbid systems.

Machine-learning retrievals trained on in-situ match-up databases (such as those assembled under the GLORIA dataset) have shown improved accuracy over empirical band ratios in cross-validation, but they inherit whatever biases exist in the training data. A model with few peatland samples will underperform in peatland catchments. Honest uncertainty quantification, per-pixel, is not yet standard in most published workflows.

When sediment and CDOM arrive together

The retrieval problem becomes significantly harder when suspended particulate matter (SPM) is elevated alongside CDOM. Both reduce blue reflectance; SPM also raises red and NIR reflectance, which is the standard diagnostic for separating them. In practice, many CDOM-rich rivers (Scottish Highland rivers, Amazonian blackwater tributaries, boreal streams after rainfall events) carry enough fine organic particles to blur the spectral boundary between dissolved and particulate fractions.

There is no clean satellite-based solution to this co-variance problem. Researchers have used the SPM-sensitive red band as a correction term in CDOM algorithms, and others have applied matrix inversion across more bands. Neither approach eliminates the ambiguity; it reduces it. Reporting aCDOM(440) without an accompanying SPM estimate and an uncertainty flag is, in most dynamic river systems, overconfident. This is a limit of the physics, not the sensor.

Peatland and forest DOC export: what the satellite signal is actually tracking

The applied interest in CDOM remote sensing is largely about carbon accounting. Peatlands store roughly 30% of global soil carbon, and DOC leaching from degraded or drained peat is a measurable carbon flux. In river systems draining peat-dominated catchments in the UK, Scandinavia, and Canada, satellite-derived aCDOM(440) time series can track seasonal and inter-annual DOC export patterns at scales that sparse gauge networks cannot cover.

Long Landsat archives make trend detection possible. Studies using Landsat data over Scottish and Scandinavian lakes have documented browning trends consistent with increased DOC loading, linked to reduced acid deposition, climate warming, and changes in hydrology. The satellite cannot directly attribute the cause; it records the optical consequence. Quantifying the carbon flux in tonnes per year still requires combining the satellite-derived concentration estimate with river discharge data, which introduces a second source of uncertainty.

For forest catchments, the signal is subtler. DOC concentrations are lower, and the spectral contrast between CDOM-rich and CDOM-poor water is smaller. Sentinel-2's 10 m resolution is genuinely useful here for resolving small headwater lakes and stream confluences that would be subpixel noise in a 30 m or 1 km product.

Honest limits of what current sensors can deliver

Cloud cover is a persistent problem across the boreal and temperate zones where most peatland DOC research is concentrated. In Scotland or Finland, cloud-free Sentinel-2 acquisitions over a given lake may number fewer than 20 per year. Compositing over time smooths seasonal signals. This is not a solvable problem with current passive optical sensors; it is a structural constraint.

Minimum detectable aCDOM(440) differences are roughly 0.1–0.3 m⁻¹ for Sentinel-2 over clear to moderately absorbing water, based on published sensitivity analyses, though this degrades near the 10 m pixel edges of small water bodies due to adjacency effects from surrounding land. Lakes smaller than about 3–5 pixels across (30–50 m diameter for Sentinel-2) produce unreliable retrievals because mixed land-water pixels dominate.

PACE OCI changes the spectral picture substantially but not the spatial one. At 1 km, it is irrelevant for most inland water bodies except the Laurentian Great Lakes, Lake Baikal, Lake Tanganyika, and similarly large systems. The promise of PACE for inland water lies in its role as a validation and process-understanding tool, not as an operational mapping platform for small lakes and rivers.

Putting it to work operationally

A practical CDOM monitoring programme typically combines three components: a locally validated empirical or semi-analytical algorithm, an atmospheric correction scheme appropriate for dark water (ACOLITE and C2RCC are the most widely used open-source options), and a time-series framework that flags cloud-contaminated pixels rather than silently including them.

Satellize runs CDOM retrieval pipelines on open Sentinel-2 and Landsat archives, applying per-scene atmospheric correction quality flags and delivering aCDOM(440) rasters with per-pixel uncertainty estimates. The approach follows the same open-method philosophy used in its Tonga crop-estimation programme: documented algorithms, traceable inputs, and outputs that a client's own scientists can interrogate. For catchment managers, water utilities, or national environmental agencies wanting to track DOC trends across a lake district or river network, the next concrete step is a scoping call to define the target water bodies, agree on the validation dataset, and establish what cloud-free frequency is actually achievable in the region of interest.

Typical figures

Spatial resolution (Sentinel-2 MSI)	10 m (B2, B3, B4); 20 m (B5–B7, B8A, B11, B12)
Spatial resolution (Landsat 8/9 OLI)	30 m all visible/NIR/SWIR bands; 15 m panchromatic (not used for CDOM)
Spatial resolution (PACE OCI)	1 km; daily global coverage
Revisit (Sentinel-2 A+B combined)	5 days at equator; 2–3 days above 50° N
Revisit (Landsat 8+9 combined)	8 days at equator
Key spectral bands for CDOM	Blue (440–490 nm) and green (560 nm); coastal/aerosol band (435 nm, Landsat OLI Band 1); full UV–NIR spectrum at 5 nm (PACE OCI)
Minimum lake size for reliable retrieval	Approximately 30–50 m diameter (Sentinel-2); 90–150 m (Landsat); 3–5 km (PACE OCI)
Typical retrieval uncertainty (aCDOM 440 nm)	0.2–0.5 m⁻¹ RMSE for empirical algorithms; varies by water type and atmospheric correction quality
Archive depth	Sentinel-2: 2015–present; Landsat: 1984–present (TM/ETM+/OLI); PACE: February 2024–present
Primary atmospheric correction tools	ACOLITE, C2RCC, iCOR (all open-source; validated for inland water)

Analytics Satellize can run

aCDOM(440) concentration raster	Empirical band-ratio or semi-analytical QAA inversion on atmospherically corrected Sentinel-2 or Landsat reflectance	Cloud-masked GeoTIFF per scene with per-pixel uncertainty layer; delivered to client GIS or cloud storage
Seasonal DOC export index	Time-series aggregation of aCDOM(440) retrievals combined with published local CDOM-to-DOC regression coefficients; trend decomposition (seasonal + inter-annual)	Annual summary report with time-series plots per lake or river reach; tabular data export
Lake browning trend detection	Long-term Landsat archive analysis using Mann-Kendall trend test on annual median aCDOM(440); change-point detection	Per-lake trend map (GIS polygon layer) with significance flags and magnitude estimates
CDOM-SPM co-variance flag	Red and NIR band thresholding to identify scenes where suspended sediment confounds CDOM retrieval; per-pixel quality mask	Quality-flag layer accompanying every aCDOM product; flagged scenes excluded from trend calculations by default
Peatland catchment DOC flux estimate	Satellite-derived CDOM concentration combined with client-supplied or satellite-estimated discharge proxy; mass-balance integration over defined catchment outlet	Monthly DOC flux time series (kg C per day) with uncertainty bounds; CSV and PDF report
PACE OCI spectral validation layer	Match-up analysis between PACE OCI hyperspectral retrievals and Sentinel-2 empirical products over large lakes; bias and RMSE statistics	Validation report documenting algorithm performance and recommended recalibration coefficients for the target region

Sources

Who does the work

We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.