Use machine learning for timely estimation of sea colours

Information from hyperspectral satellite measurements in the presence of clouds, aerosols and sanglints

We have developed new technologies to estimate marine color properties under sub-ideal search conditions, such as clouds, aerosols, and sanglints. This work demonstrates that atmospheric spectral signals can be separated from surface spectral signals using principal component analysis and hyperspectral measurements from instruments such as the aura ozone monitoring instrument (OMI), tropomi, and pace OCI. It then uses a machine learning approach to learn the relationships between the functions extracted from principal component analysis and acquired ocean properties such as remote sensing reflectance and chlorophyll concentration. This work shows that an approach can be applied to Omi and Tropomi to fill the gaps in traditional marine color algorithms and also obtain sea color properties from UV visible sensors.

On October 15, 2020, daily chlorophyll concentrations from Omi on the left side from the north coast of Brazil, from the central tropomi, and images of the composite chlorophyll searches and tropomi on the right and Modis acqua and Modis Terra show significantly improved spatial coverage compared to those from Modis.

Scientific significance, social relevance, and relationships with future missions:

Satellite-based searching for marine characteristics is important for a further understanding of marine ecosystems. Many current algorithms for marine colour recovery are limited by non-ideal recovery conditions such as clouds, aerosols, and sun glow. We have developed new techniques for “filling the gaps” that clouds, aerosols and sanglints have used hyperspectral instruments to leave in current marine color searches. This technique uses principal component analysis to separate the various spectral features of hyperspectral measurements. Given these different spectral features, machine learning algorithms are used to learn the relationship between these spectral features and the amount of capture, such as remote sensing reflectance and chlorophyll concentration. This work applies an approach to hyperspectral ozone mapping equipment (OMI) that has been in flight for over 17 years and helps to fill gaps in historical marine color datasets such as Modis and Seawifs. Additionally, the technique can be applied to the highly spatially resolved tropospheric monitoring instrument (Tropomi) of hyperspectral instruments, and can provide near-realistic time-based sea-color datasets using “gap-filled” marine color information over the next few years.

In the future, this new technique could be applied to many future hyperspectral missions, such as Earth competition mission tempos and gems, and possibly enhance the daytime coverage of sea-color information. Additionally, measurements are made at much higher spectral resolution than heritage marine color instruments such as Modis and Seawifs, making them applicable to future ocean color missions such as Pace OCI and Glimr.

Data Source:

Omi (ozone monitoring equipment), Tropomi (troposphere monitoring equipment), Aqua Modis (medium resolution imaging spectroradiometer), and Terra Modis

References:

Fasnacht, Z., Joiner, J., Haffner, D., Qin, W., Vasilkov, A., Castellanos, A.. , and Krotkov, N. It was accepted by the frontier through remote sensing, 2022.

04.2022