Satellite Team Rewrites 20 Years of Coastal Water Color Data for China and the US

For two decades, some of the world’s most scientifically valuable coastal waters have been, in effect, half-invisible to the satellites watching them. Murky estuaries, sediment-laden bays, and haze-shrouded shorelines have consistently defeated the standard processing pipelines that turn raw satellite radiance into ocean color measurements. Now a research team led by Shuhui Zhao and Lian Feng, with colleagues at Southern University of Science and Technology, Wuhan University, Shanghai Jiao Tong University, and the Eastern Institute of Technology, has rebuilt that processing chain from the ground up. The result, published in Earth System Science Data, is a twenty-year record of remote sensing reflectance, the fundamental quantity that encodes the color of coastal water as seen from space, covering the coasts of China and the United States from 2003 to 2022, and it dramatically outperforms the standard NASA products in exactly the places where those products have struggled most.

Remote sensing reflectance, usually abbreviated Rrs, is the ratio of the light leaving the water surface to the light arriving from the sun, and it is the bedrock measurement of ocean color science. From Rrs, researchers derive chlorophyll a concentrations, suspended sediment loads, particulate organic carbon, and the inherent optical properties of the water column. The trouble is that the satellite sensor does not measure water-leaving light directly. Roughly ninety percent of the signal recorded at the top of the atmosphere comes from the atmosphere itself, from molecular Rayleigh scattering, aerosol scattering, and gaseous absorption. Stripping that atmospheric contribution away, a procedure known as atmospheric correction, is the single most delicate step in the entire ocean color processing chain, and it is where coastal retrievals have historically fallen apart.

The core of the problem lies in the aerosol models that operational algorithms rely on. The current default, a set of models developed by Ahmad and colleagues in 2010 from AERONET ground observations, was built largely from open-ocean sites. Only three of its calibration stations sat on the coast, all within Chesapeake Bay. These models therefore describe weakly absorbing, marine-dominated aerosols, and they systematically fail when confronted with the strongly absorbing dust, smoke, and urban pollution that hangs over industrialized coastlines. The consequence is a well-documented overcorrection: the algorithm attributes too much of the measured signal to aerosols and too little to the water, producing Rrs values that are biased low, particularly in the blue bands where absorbing aerosols do their worst damage.

A second, less glamorous failure mode compounds the first. The standard cloud mask, designed for the open ocean, flags any pixel whose reflectance at 869 nanometers exceeds a fixed threshold. In turbid coastal water, the water itself can be bright enough at that wavelength to trip the flag, so legitimate observations of sediment-rich estuaries are thrown away as cloud. Entire estuaries, including large portions of the Pearl River Estuary and the Bohai Sea, simply vanish from the standard data record. For anyone trying to track water quality, sediment transport, or long-term environmental change along the world’s most heavily used coastlines, the standard products have been both biased and incomplete.

The new framework attacks both problems simultaneously. For atmospheric correction, the team built on a set of global gridded aerosol models developed by Zhao and colleagues in 2023, which use AERONET observations from 1,475 sites worldwide to generate 3,207 monthly aerosol models distributed across 310 five-degree grid cells. These models capture the strong absorption and high aerosol optical thickness typical of coastal atmospheres. But applying a single model type everywhere would create its own errors, because a single satellite scene can span turbid, aerosol-heavy nearshore water and clear offshore water within a few dozen kilometers. The team’s key innovation is a spatially adaptive selection scheme: pixels are classified as turbid or clear based on their normalized near-infrared water-leaving radiance, which serves as a practical proxy for proximity to land. Turbid nearshore pixels keep their locally informed gridded aerosol model, while clear offshore pixels switch to the conventional open-ocean models, avoiding both the underestimation near shore and the overestimation offshore.

On the masking side, the framework relaxes the cloud threshold by moving it to the 2,130-nanometer shortwave infrared band, where turbid water is genuinely dark and clouds remain bright. That single change recovers vast numbers of previously discarded coastal pixels. Because a relaxed mask risks letting thin cloud slip through, the team added targeted safeguards, including a spectral filter that removes pixels whose blue-band reflectance exceeds their 531-nanometer value, a signature of cloud contamination rather than genuine turbid water, and a mask that excludes saturated signals in extremely turbid conditions. The same quality control chain was applied uniformly to both the new dataset and the standard products, ensuring that every comparison between them was conducted on level ground.

The validation results are striking. Using 6,362 in situ measurements from the SeaBASS archive, matched to satellite overpasses within a three-hour window and screened for spatial homogeneity, the team found that the average regression slope across eight MODIS Aqua bands improved from 0.90 plus or minus 0.06 for the standard product to 1.00 plus or minus 0.08 for the new dataset, a restoration to unity that signals the systematic low bias has been substantially eliminated. The gains are most pronounced in the 488 to 555 nanometer range, the green part of the spectrum most relevant to water color applications. Some challenges persist in the blue bands at 412 and 443 nanometers, where monthly aerosol models struggle with rapidly changing coastal atmospheres, and slight overestimation appears in the red bands, but the overall correction of the pervasive bias is unambiguous.

The improvement in data quantity is just as dramatic. At 443 nanometers, the new dataset delivers a regional average of 56 percent more valid observations along China’s coast and 18 percent more along the US coast. In the Bohai Sea, where semi-enclosed geography and sediment-laden river inflows keep the water persistently turbid, valid retrievals reach up to 140 days per year, and some individual areas gain as many as 94 additional observation days annually compared with the standard product. Single-scene comparisons over the Pearl River Estuary, the Bohai Sea, Chesapeake Bay, and the Mississippi River Estuary show the new dataset recovering fine-scale features, such as the distinct high-reflectance sediment band along the western Pearl River Estuary, that are entirely absent from standard imagery. In sample windows, mean reflectance at 547 nanometers rises from 0.0233 to 0.0270 per steradian in the Pearl River Estuary and from 0.0199 to 0.0242 per steradian in Chesapeake Bay, values that are physically more plausible for these turbid waters.

The team also benchmarked their framework against three advanced alternative atmospheric correction algorithms, OC-SMART, POLYMER, and the NIR-SWIR switching method, across the same four regions. While all methods captured broad spatial patterns, the new dataset consistently retrieved the highest and most physically credible reflectance values in the most turbid zones, where the alternatives often remained close to the biased baseline. The authors attribute this robustness to their modular, physics-based design: rather than depending on pre-trained statistical models whose performance is tied to their training domain, the framework improves a specific radiative transfer component, the aerosol prior, which can be updated regionally without retraining the whole system. That transparency, they argue, offers a clearer path to extending the approach to other sensors and other coastlines.

The twenty-year record already reveals meaningful environmental signals. Along China’s coast, reflectance at 443 nanometers has increased at an average rate of 0.64 percent per year, with peaks of 6.85 percent per year in the central Yellow Sea, while several Bohai Bay embayments show consistent declines that may reflect weakening wind-driven sediment resuspension. Along the US coast, the Gulf Stream region presents a complex fingerprint: rising blue-band reflectance alongside falling green and red values, a pattern the authors link to growing oligotrophy and enhanced upper-ocean stratification associated with regional warming. The dataset, publicly available on Zenodo as monthly composites at 0.01-degree resolution, is positioned as a complement to, not a replacement for, standard products, filling the gap where estuaries, bays, and turbid coastal waters have long been under-served. For coastal managers, climate scientists, and anyone tracking the health of the world’s busiest shorelines, two decades of coastal water color have finally come into focus.

Subject of Research: Improved satellite atmospheric correction for generating a 20-year remote sensing reflectance dataset over Chinese and US coastal waters

Article Title: High-quality remote sensing reflectance products over China and US coast

Article References: Zhao, S., Wu, Y., Lv, J., Zhao, D., Zheng, Y., & Feng, L. (2026). High-quality remote sensing reflectance products over China and US coast. Earth System Science Data, 18(10), 7475-7498. https://doi.org/10.5194/essd-18-7475-2026

Image Credits: AI Generated

DOI: 10.5194/essd-18-7475-2026

Keywords: ocean color, remote sensing reflectance, atmospheric correction, MODIS Aqua, coastal waters, aerosol models, cloud masking, water quality, China coast, US coast, Earth System Science Data, long-term monitoring