Combining BioGeoChemical-Argo (BGC-Argo) floats and satellite observations for water column estimations of the particulate backscattering coefficient
<p>As the second-largest carbon reservoir on Earth, the ocean regulates the carbon balance through dissolved and particulate organic carbon (POC) forms. Monitoring carbon cycle processes is key to understanding the climate system. Although most organic carbon in the ocean exists in dissolved f...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Copernicus Publications
2025-08-01
|
| Series: | Ocean Science |
| Online Access: | https://os.copernicus.org/articles/21/1677/2025/os-21-1677-2025.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | <p>As the second-largest carbon reservoir on Earth, the ocean regulates the carbon balance through dissolved and particulate organic carbon (POC) forms. Monitoring carbon cycle processes is key to understanding the climate system. Although most organic carbon in the ocean exists in dissolved form, POC, despite its smaller share, plays a vital role by connecting surface biomass production with the deep ocean and sedimentation processes. POC estimation is achieved by measuring proxies like the particulate backscattering coefficient (<span class="inline-formula"><i>b</i><sub>bp</sub></span>) estimated from satellite observations and in situ sensors, such as the BioGeoChemical-Argo (BGC-Argo) floats. Previous studies have integrated data from BGC-Argo floats and satellite sensors, demonstrating the potential of machine learning models to estimate vertical bio-optical properties within the water column. The approach presented here enhances the estimation within the top 250 m of the water column compared with previous works. The estimations are performed in two distinct regions, the North Atlantic and the Subtropical Gyres, and across several layers within two maximum depth limits of 50 and 250 m. Data from BGC-Argo profiles and the Ocean and Land Colour Instrument (OLCI) sensor are used together to build a training dataset for a random forest model, which is applied with different sets of variables. Additional considerations regarding our datasets include short time criteria for matchups (<span class="inline-formula">±24</span> h) and high spatial resolution. The random forest model shows promising results, especially within the first 50 m in the Subtropical Gyres.</p> |
|---|---|
| ISSN: | 1812-0784 1812-0792 |