loading page

Random forest regression on multi-platform in-situ ocean observations: Investigating high-frequency nutrient dynamics in the Southern Ocean
  • +4
  • Sangmin Song,
  • Paige D Lavin,
  • Alison R Gray,
  • Zachary Nachod,
  • Dhruv Balwada,
  • Lilian A Dove,
  • Andrew F Thompson
Sangmin Song
School of Oceanography, University of Washington

Corresponding Author:[email protected]

Author Profile
Paige D Lavin
NOAA/NESDIS Center for Satellite Applications and Research, Cooperative Institute for Satellite Earth System Studies/Earth System Science Interdisciplinary Center (CISESS/ESSIC), University of Maryland
Alison R Gray
School of Oceanography, University of Washington
Zachary Nachod
University of Hawai'i at Mānoa
Dhruv Balwada
Lamont-Doherty Earth Observatory, Columbia University
Lilian A Dove
Brown University
Andrew F Thompson
Environmental Science and Engineering, California Institute of Technology

Abstract

Nutrient cycling in the ocean is mediated by physical mixing processes that span diverse spatial and temporal scales. New biogeochemical profiling floats (BGC-Argo) have begun to observe nutrient distributions globally, but their 10-day cycling period limits the types of processes they can capture. Small-scale dynamics, occurring on \(O\)(1) day and \(O\)(1) km, remain particularly difficult to observe in-situ. Here, we show that random forest regression (RFR) can recover high-frequency information by leveraging the sampling strategies of multiple ocean profilers. Our RFR is trained, validated, and tested on BGC-Argo and shipboard data to within ~3% accuracy, then applied to observations from two rapid-sampling Seagliders deployed during the Southern Ocean Glider Observations of the Submesoscale (SOGOS) experiment in 2019. This approach generates novel nitrate distributions at 50 times the horizontal resolution of the original float data. Using the high-resolution RFR outputs, we identify signatures of nutrient injection into the mixed layer that coincide with enhanced stirring in a turbulent region downstream of the Southwest Indian Ridge. Relating these intermittent transport events to biological time series suggests that small-scale stirring mediates additional nutrient drawdown and primary production in this region. In our exploration of high-frequency nitrate variability in the Southern Ocean, RFR extends the capabilities of the BCG-Argo array and allows for deeper understanding of biogeochemical cycling at a more comprehensive set of scales. As a flexible approach that can be generalized to suit other multi-platform observing systems, RFR presents new opportunities to maximize value from existing datasets.
29 May 2024Submitted to ESS Open Archive
30 May 2024Published in ESS Open Archive