Coastal waters contain suspended sediments, dissolved organic matter, and highly variable aerosols that make satellite observations difficult and can degrade estimates of water quality parameters. Conventional atmospheric correction schemes often assume simplified water and aerosol properties, which break down in turbid or dusty environments and are further compromised by sensor noise and striping. The study argues that a correction method trained directly on real-world conditions is needed to improve the reliability of global coastal ocean color monitoring.

Researchers from Xiamen University report ACA-SIM (atmospheric correction based on satellite-in situ matchup data) as an algorithm built on satellite-AERONET-OC matchups and implemented with a neural network trained on real satellite and field measurements. Their work, published on October 16, 2025 in the Journal of Remote Sensing, targets the long-standing problem of retrieving accurate remote-sensing reflectance (Rrs) from coastal waters affected by aerosols, dust, and smoke, and positions ACA-SIM as an alternative to current NASA methods.

ACA-SIM uses a multilayer neural network trained on more than 8,800 satellite-AERONET-OC matchups that span a wide range of water and aerosol types around the world. Unlike earlier schemes that relied heavily on simulated scenes, the model incorporates real sensor effects such as striping and stray light during training. In validation against independent field datasets, ACA-SIM achieved an average mean absolute percentage difference of about 15 percent in blue spectral bands, compared with about 32 percent for OC-SMART and more than 50 percent for NASA's standard algorithm.

The algorithm maintained stable performance under high solar angles, strong sun glint, and absorbing aerosol conditions, which commonly challenge traditional correction methods. When applied to MODIS-Aqua imagery over the Bohai and Yellow Seas, West Africa dust zones, and Australian bushfire regions, ACA-SIM removed negative Rrs values, reduced striping artifacts, and preserved realistic patterns of water masses in the resulting ocean color maps.

To build the training dataset, the team combined top-of-atmosphere reflectance (?t) from MODIS-Aqua with in situ Rrs measurements from more than 40 AERONET-OC stations worldwide. These matchups covered a broad range of viewing geometries, turbidity levels, and aerosol conditions, including marine and continental aerosols as well as strongly absorbing dust and smoke. A four-layer multilayer perceptron with 512, 256, 128, and 64 neurons was optimized to predict Rrs at 12 MODIS wavelengths between 412 and 869 nanometers.

Validation used both AERONET-OC and ship-based datasets and showed that ACA-SIM consistently outperformed NASA's standard and OC-SMART algorithms in accuracy and robustness. Statistical metrics including the coefficient of determination, mean absolute percentage difference, and bias were used to quantify performance gains over existing correction methods.

"Our goal was to let the algorithm learn from reality rather than simulation," said Prof. Zhongping Lee, corresponding author of the study. "By training ACA-SIM on genuine satellite-field matchups, we allowed it to capture subtle sensor behaviors and atmospheric effects that synthetic datasets cannot reproduce. The outcome is a smarter, more reliable correction system that ensures accurate monitoring of coastal ecosystems under even highly complex atmospheric scenarios."

The team assembled global AERONET-OC records from 2002 to 2024 and matched them with MODIS-Aqua radiance within a plus-or-minus one-hour window to maintain temporal consistency. A quality-control procedure removed contaminated pixels while retaining moderately glinted or hazy scenes so that the model would generalize better to operational conditions. Training used 80 percent of the matchup dataset, with the remaining 20 percent reserved for validation, and early-stopping criteria were applied to limit overfitting.

The success of ACA-SIM underscores the potential of data-driven correction strategies in Earth observation. The authors intend to expand the framework to other satellite sensors, including VIIRS and Sentinel-3, to build a unified cross-platform atmospheric correction system. By generating consistent and accurate ocean color products across instruments, such methods could improve tracking of algal blooms, sediment transport, and carbon fluxes and support management of coastal and marine environments under climate change.

Research Report:Atmospheric Correction of Coastal Waters Based on Satellite-AERONET-OC Matchups

Related Links
Xiamen University
Water News - Science, Technology and Politics