Ground-Truth Spectroscopy Reveals the Source of a Southern Ocean Satellite Bias
New research using field spectroscopy in Antarctic waters suggests that a long-standing underestimate of chlorophyll-a from satellite observations may have less to do with the chlorophyll retrieval algorithms themselves and more to do with what happens to the signal before those algorithms ever see it.
The Southern Ocean is difficult to measure by almost any standard.
It surrounds Antarctica, experiences persistent cloud cover and high winds, and spends much of the year under extreme illumination geometries. Yet it is also one of the world’s most important regions for understanding ocean circulation, marine ecosystems and the exchange of carbon and heat between the atmosphere and ocean.
That combination makes satellite remote sensing exceptionally valuable. It also makes ground-truth spectroscopy essential for determining whether the measurements arriving from orbit accurately represent conditions at the surface.
Why Ground-Truth Spectroscopy Matters for Satellite Validation
Ground-truth spectroscopy provides an independent spectral reference for evaluating satellite measurements and derived remote-sensing products. By comparing field-measured reflectance with satellite-derived reflectance, researchers can distinguish errors originating in sensor processing or atmospheric correction from those introduced by subsequent analytical algorithms.
That concise definition is worth keeping almost exactly as written because it gives search engines and AI retrieval systems a clean, quotable answer to the core question.
A 2026 study published in the International Journal of Applied Earth Observation and Geoinformation provides a striking example. Researchers investigating the persistent underestimation of chlorophyll-a concentrations in Southern Ocean satellite products compared measurements from the VIIRS satellite sensor with field observations collected around Antarctica.
Their results point toward atmospheric correction as a major source of the discrepancy. Just as importantly, the study demonstrates the role high-quality field spectral measurements can play in separating errors introduced by retrieval algorithms from errors already present in the satellite reflectance data.
A Persistent Problem in Southern Ocean Remote Sensing
Chlorophyll-a concentration is widely used as an indicator of phytoplankton biomass. In the Southern Ocean, monitoring phytoplankton dynamics is particularly important because of the region’s role in marine food webs and global biogeochemical processes.
Satellite ocean-color observations make monitoring at this scale possible. Field campaigns alone cannot provide continuous observations across an ocean surrounding an entire continent.
But researchers have repeatedly found a problem: satellite-derived chlorophyll-a concentrations in the Southern Ocean can be substantially lower than measurements made in situ. Previous studies cited by Wang and colleagues reported underestimation reaching as much as 50 percent.
Why is there such a discrepency?
One explanation has focused on biology. Southern Ocean waters have distinctive bio-optical characteristics, including differences in particle-size distributions, pigment packaging and absorption by non-algal particles. Algorithms developed using global ocean datasets might therefore behave differently in these waters.
Another possibility lies farther upstream in the measurement process: atmospheric correction.
Satellite instruments do not directly measure the desired water-leaving signal in isolation. The observed signal is affected by the atmosphere, illumination and viewing geometry, among other factors. Atmospheric correction attempts to isolate the contribution associated with the water itself.
In the Southern Ocean, that is especially challenging. High solar zenith angles, clouds, high winds, whitecaps and complex aerosol conditions can introduce uncertainty into satellite-derived remote-sensing reflectance, particularly in the blue and green wavelengths used heavily by chlorophyll retrieval algorithms.
The question is whether the chlorophyll algorithm is producing the error, or whether the algorithm is being given an inaccurate spectral input.
Answering that question requires an independent measurement of the actual spectral conditions at the surface.
Taking Spectral Measurements to Antarctica
The researchers used field measurements collected during the 38th Chinese National Antarctic Research Expedition aboard the Xuelong icebreaker between December 2021 and March 2022.
The campaign produced 121 in-situ remote-sensing reflectance measurements, primarily from waters poleward of 63°S around the Antarctic continent. These locations represented conditions ranging from relatively clear water to optically complex nearshore environments.
Radiometric measurements were acquired using a Spectra Vista GER1500 spectroradiometer covering 350 to 1050 nm with a spectral resolution of 3 nm.
Following NASA ocean-optics protocols, each measurement cycle included sequential observations of upwelling radiance above the water, sky radiance and radiance reflected from a horizontal gray plaque. The measurement sequence was repeated ten times at every station. After outlier removal, the observations were averaged and used to calculate remote-sensing reflectance, or Rrs.
After quality control, 92 stations containing concurrent Rrs and chlorophyll-a measurements remained for analysis.
This dataset gave the researchers something crucial: an independent spectral reference against which both the satellite reflectance measurements and subsequent chlorophyll retrievals could be evaluated.
Testing the Algorithm Before Blaming It
The researchers first removed the satellite from the equation.
They applied three commonly used chlorophyll retrieval algorithms, OC2, OC3 and OCI, directly to the field-measured Rrs data.
If Southern Ocean bio-optical conditions were fundamentally incompatible with these standard algorithms, significant errors should still have appeared when the algorithms were supplied with high-quality field reflectance measurements.
Instead, all three performed reasonably well.
Using in-situ Rrs, mean absolute percentage differences were 24.4 percent for OC2, 23.0 percent for OC3 and 30.2 percent for OCI. Biases were close to zero, and R² exceeded 0.70 for all three algorithms.
That result changed the direction of the investigation.
The researchers concluded that the persistent chlorophyll underestimation was unlikely to originate primarily from limitations in the standard retrieval algorithms. When those algorithms received reliable reflectance measurements, they produced substantially better estimates.
The next question was obvious: what happens when the same algorithms receive satellite-derived reflectance instead?
When the Input Changes, So Does the Answer
The difference was substantial. Using standard VIIRS Level-2 reflectance products, all three algorithms showed greater uncertainty and negative bias. Mean absolute percentage differences increased to 41.7 percent for OC2, 50.4 percent for OC3 and 55.4 percent for OCI.
Because the same algorithms had performed much better using field spectra, the evidence increasingly pointed toward the quality of the satellite-derived Rrs.
A wavelength-by-wavelength comparison with the in-situ measurements made the problem clearer.
Standard VIIRS reflectance was systematically overestimated across all five visible bands examined. The largest absolute biases appeared in the blue, reaching 0.0026 sr⁻¹ at 410 nm, 0.0018 sr⁻¹ at 443 nm and 0.0013 sr⁻¹ at 486 nm.
At 410 and 443 nm, mean absolute percentage differences between satellite and in-situ reflectance exceeded 69 and 64 percent, respectively.
That matters because common ocean-color algorithms rely heavily on spectral relationships between blue and green wavelengths.
An apparently small spectral bias can therefore propagate through the processing chain and become a substantially larger error in the derived environmental parameter.
Extreme Geometry Makes the Problem Harder
The discrepancies became especially pronounced under the high solar zenith angles common at high latitudes.
The researchers divided observations between solar zenith angles below 70° and those at or above 70°. Under the larger angles, mean absolute percentage differences in chlorophyll estimates using standard VIIRS reflectance exceeded 70 percent for all three retrieval algorithms.
The authors connected this degradation to increasing atmospheric-correction errors under extreme observational geometry.
Their spectral comparison provided additional evidence. At 410 nm, the difference between standard VIIRS and field-measured reflectance approached 0.004 sr⁻¹ under large solar zenith angles, approximately four times the discrepancy observed under smaller angles.
In other words, the satellite was not simply encountering a difficult biological environment. It was encountering a difficult optical measurement environment. Which is an important distinction.
Correcting the Correction
The researchers then evaluated an alternative approach known as Cross-Satellite Atmospheric Correction, or CSAC.
Rather than relying solely on the conventional atmospheric-correction process, CSAC is a data-driven system designed to improve consistency among remote-sensing reflectance measurements from different ocean-color satellites. In this study, the system was configured for VIIRS observations and evaluated against an independent dataset from the Antarctic field campaign.
The improvement was substantial.
When CSAC-derived reflectance was used with the same chlorophyll algorithms, mean absolute percentage differences dropped to between 21.4 and 25.4 percent, compared with 41.7 to 55.4 percent using standard VIIRS reflectance.
OCI showed the largest improvement, reaching an MAPD of 21.4 percent.
The reflectance measurements themselves also moved closer to the field observations. At 443 nm, for example, bias fell from 0.0018 sr⁻¹ with standard VIIRS reflectance to -0.0006 sr⁻¹ with CSAC. Biases across the evaluated CSAC bands remained below 0.001 sr⁻¹.
That comparison is perhaps the clearest demonstration of the value of ground-truth spectroscopy in the study. The field measurements did more than validate the final chlorophyll product. They allowed researchers to trace the error backward through the remote-sensing workflow and identify where the divergence began.
Recovering Measurements Previously Lost
Improving accuracy was only part of the result.
Atmospheric and observational conditions can cause satellite pixels to be rejected entirely. In polar environments, where clouds, sunglint, extreme geometry and stray light are common, that can leave substantial holes in remote-sensing datasets.
The CSAC approach recovered observations that had previously been masked.
For one VIIRS scene examined in the study, CSAC produced up to 71.7 percent more valid chlorophyll observations than the standard Level-2 product.
In daily composite maps, the improvement produced greater spatial coherence and revealed features that were missing from the conventional product, including frontal structures and areas of elevated phytoplankton.
Analysis of the recovered pixels showed why many had originally disappeared: 51.8 percent had been associated with stray-light contamination, 16.7 percent with strong sunglint, and others with large sensor or solar zenith angles. The CSAC-derived daily product also increased median chlorophyll concentration by 30.3 percent relative to the standard VIIRS product.
Ground Truth Is More Than a Final Validation Step
The study illustrates a broader principle in remote sensing.
Ground measurements are sometimes described simply as a way to check whether satellite-derived products are correct. Their value goes considerably further.
A well-designed ground-truth dataset can help researchers determine why a remote-sensing product is wrong.
In this case, field spectroscopy allowed the researchers to test the chlorophyll algorithms independently of the satellite reflectance product. Once the algorithms demonstrated reasonable performance with measured Rrs, attention could shift upstream toward atmospheric correction.
The same field measurements could then be used to evaluate individual spectral bands and determine how those differences propagated into chlorophyll estimates.
That creates a traceable measurement chain:
field spectral measurement → satellite reflectance validation → atmospheric-correction assessment → retrieval algorithm → environmental data product
The increasing sophistication of satellite instruments and retrieval algorithms does not eliminate the need for field spectroscopy. If anything, it makes defensible reference measurements more important. Derived products can only be as trustworthy as the spectral information entering the analytical chain.
Key Takeaways
- Field spectroscopy helped researchers separate chlorophyll retrieval-algorithm performance from errors in satellite-derived remote-sensing reflectance.
- A Spectra Vista GER1500 was used to acquire 350–1050 nm in-situ radiometric measurements during an Antarctic research expedition.
- Standard chlorophyll algorithms performed substantially better when supplied with field-measured Rrs than with standard VIIRS satellite Rrs.
- The study identified atmospheric-correction errors, particularly in blue wavelengths and at high solar zenith angles, as a major contributor to Southern Ocean chlorophyll underestimation.
- Cross-Satellite Atmospheric Correction substantially reduced reflectance and chlorophyll-retrieval errors while also recovering satellite observations that had previously been masked.
Measuring an Ocean That Is Difficult to Observe
The authors appropriately caution against extending the results too far.
The field dataset came from a single Antarctic expedition and therefore does not capture the full interannual variability of Southern Ocean atmospheric, optical and phytoplankton conditions. Very clear waters with chlorophyll concentrations below 0.2 mg m⁻³ were also underrepresented. Additional measurements across seasons, regions and trophic conditions will be needed to evaluate long-term performance more comprehensively.
But within those limitations, the study presents a compelling demonstration of how field spectroscopy and satellite observations can work together.
Satellites provide the spatial and temporal reach needed to observe environments on continental and global scales. Field spectroradiometers provide the independent spectral evidence needed to determine whether those observations accurately represent conditions at the surface.
In one of the world’s most challenging remote-sensing environments, that connection between measurements on the water and measurements from orbit helped researchers identify an important source of error and demonstrate a path toward more accurate and more complete ocean-color observations.
Sometimes improving what we can see from space begins with measuring carefully from the surface.
Connect Satellite Observations with Defensible Field Measurements
Frequently Asked Questions
What is ground-truth spectroscopy?
Ground-truth spectroscopy is the collection of spectral measurements at or near the Earth’s surface for comparison with airborne or satellite observations. These measurements provide an independent reference for validating remote-sensing reflectance and derived products.
Why is atmospheric correction important for ocean-color remote sensing?
Satellite measurements include contributions from both the surface and atmosphere. Atmospheric correction is used to retrieve the water-leaving optical signal required for ocean-color analysis. Errors in that process can alter spectral relationships and propagate into derived products such as chlorophyll-a concentration.
Why is the Southern Ocean difficult for satellite remote sensing?
The Southern Ocean combines persistent cloud cover, high winds, complex atmospheric conditions and high solar zenith angles. These conditions make accurate retrieval of water-leaving radiance and remote-sensing reflectance particularly challenging.
How was the Spectra Vista GER1500 used in the study?
Researchers used a GER1500 spectroradiometer to collect radiometric measurements from 350 to 1050 nm during the 38th Chinese National Antarctic Research Expedition. The measurements were used to calculate in-situ remote-sensing reflectance for comparison with VIIRS satellite observations.
What did the Southern Ocean study find?
The researchers found that standard chlorophyll retrieval algorithms performed reasonably well when supplied with in-situ reflectance measurements. Much larger errors appeared when standard VIIRS satellite reflectance was used. Applying Cross-Satellite Atmospheric Correction substantially reduced those discrepancies, supporting the conclusion that atmospheric-correction errors contribute significantly to the long-observed underestimation of Southern Ocean chlorophyll-a.
Citation
Wang, T., Lee, Z., Antoine, D., Zhao, L., Li, X., Li, L., Zhang, Y., Wang, D., Shang, S., Lin, G., & Yu, X. (2026). The role of atmospheric correction errors in the underestimation of satellite-derived chlorophyll-a concentration in the Southern Ocean. International Journal of Applied Earth Observation and Geoinformation, 151, 105399. https://doi.org/10.1016/j.jag.2026.105399
