A machine learning-based dissolved organic carbon climatology

Marine dissolved organic carbon (DOC) is a significant carbon reservoir that impacts climate but remains poorly quantified. The absence of a comprehensive DOC climatology impedes model validation, estimation of the current DOC inventory, and comprehension of DOC's role in the carbon cycle and climate. To tackle this issue, we employed boosted regression trees to link a compilation of DOC observations with various environmental climatologies, extrapolating these relationships across the entire ocean to generate annual layer-wise DOC climatologies with associated uncertainties. The prediction performance was satisfactory, with R² values ranging from 0.6 to 0.8 across all layers. In the bathypelagic layer, DOC was primarily predicted by dissolved oxygen, while nutrients were the main predictors in other layers. We estimate the total oceanic DOC inventory to be approximately 690 PgC. Our findings demonstrate that machine learning is a powerful tool for developing climatologies from limited observations.

 

Files description

annual_climatologies.csv
- lon: longitude
- lat: latitude
- surf_doc_avg: average DOC prediction in the surface layer (0 - 10 m)
- surf_doc_sd: standard deviation of DOC prediction in the surface layer (0 - 10 m)
- epi_doc_avg: average DOC prediction in the epipelagic layer (10 - 200 m)
- epi_doc_sd: standard deviation of DOC prediction in the epipelagic layer (10 - 200 m)
- meso_doc_avg: average DOC prediction in the mesopelagic layer (200 - 1000 m)
- meso_doc_sd: standard deviation of DOC prediction in the mesopelagic layer (200 - 1000 m)
- bathy_doc_avg: average DOC prediction in the bathypelagic layer (> 1000 m)
- bathy_doc_sd: standard deviation of DOC prediction in the bathypelagic layer (> 1000 m)

 

seasonal_climatologies.csv
- lon: longitude
- lat: latitude
- season: meteorological season in the northern hemisphere as 1 = DJF, 2 = MAM, 3 = JJA, 4 = SON
- surf_doc_avg: average DOC prediction in the surface layer (0 - 10 m)
- surf_doc_sd: standard deviation of DOC prediction in the surface layer (0 - 10 m)

Disciplines

Chemical oceanography

Location

90N, 90S, 180E, 180W

Data

FileSizeFormatProcessingAccess
Annual climatologies in the 4 layers
5 MoCSVProcessed data
Seasonal climatologies for the surface layer
6 MoCSVProcessed data
How to cite
Panaïotis Thelma, Wilson Jamie, Cael BB (2024). A machine learning-based dissolved organic carbon climatology. SEANOE. https://doi.org/10.17882/101170

Copy this text