- A new machine learning technique identifies links and patterns in the ocean that make sense.
- It reveals a total of 5 dynamic consistent regions that make up nearly 93.7% of the world’s ocean, along with their most dominant physical parameter.
Before the emergence of advanced observational and modeling methods, the physical/dynamical state of the ocean was determined by using large-scale quasi-laminar descriptions like Stommel-Arons flows, abyssal recipes, or Sverdrup balance.
Recent advances in modeling capability and instrumentation have shown that ocean physics can be characterized by intricate spatial and temporal variability. Every region in the sea has a unique state that depends on several factors such as local meteorology, proximity to western and eastern boundaries, and more.
To detect what physics are most dominant in a given location, one needs to examine an overwhelming number of data points for several parameters, including salinity, velocity, temperature, and how things alter with depth.
Since it’s impossible for any human to decipher such enormous amounts of data, MIT researchers have developed a new machine learning method to identify links and patterns in the ocean that make sense.
What Did The Algorithm Solve?
The research team used the ‘Estimating the Circulation and Climate of the Ocean‘ (ECCO) to obtain data about what’s happening in the global ocean. ECCO provides ocean variability, coastal physics, biological cycles, and geodesy, based on billions of parameters recorded in the last 2 decades.
They then applied K-means clustering — a method of vector quantization — to detect robust patterns within data and determine the dominant physics in the sea. The outcomes revealed a total of 5 clusters, representing 5 dynamic consistent regions that makeup nearly 93.7% of the world’s ocean.
The biggest cluster, for instance, accounts for about 43% of the global sea: its most dominant parameter is the wind stress on the sea surface which is balanced by torques in the bottom. This parameter is mostly recorded in subpolar and subtropical gyres in the Northern Hemisphere, large portions of the Arctic ocean, and a thin ribbon in the Southern Ocean.
Oceans clustered by similar parameters | Credit: Maike Sonnewald
Similarly, the other 4 clusters show the dominant physical parameter and where exactly it can be found in the global ocean. The remaining 6.3% of the ocean areas were quite challenging to pin down.
In the next study, researchers will use the same machine learning technique with higher resolution data to track the remaining 6.3%. They will focus on factors sensitive to climates, such as gyre circulation and overturning.
For now, this tool can help oceanographers and scientists ease their analysis, compare regions to ones that behave similarly, and focus their research in the right places.