The heat content of the Arctic Ocean is crucial globally, affecting climate, weather, sea levels, and ecosystems. It serves as an indicator of broader climate change effects worldwide, connecting ecosystems, economies, and societies globally.
Highlighting the importance of such studies, Palanisamy Shanmugam, professor at the Department of Ocean Engineering, Indian Institute of Technology, Madras, said, “The Arctic OHC (ocean heat content) is a vital measure to represent the global climate system. This research and the OHC estimates produced from it will lead to better understanding of global climate change such as sea level rise and temporal trends of polar sea ice extent decline.”
Researchers from IIT Madras have created an artificial neural network (ANN) model to estimate OHC in ice-covered Arctic regions. They have linked satellite-based sea ice data to in-situ CTD (conductivity, temperature, depth) profiles to estimate OHC up to 700 metres deep. This model accurately predicts OHC changes and tracks spatio-temporal variations, offering insights into historical trends and regional patterns. Led by professor Palanisamy Shanmugam and his student Kondeti Vijay Prakash, the team published its work in the peer-reviewed journal IEEE Access .
Scarce data
The challenge for reliable OHC estimates is the scarcity of data. Estimating OHC in the inaccessible top ocean layer involves approximations, leading to uncertainties. Improving accuracy involves creating in-situ databases for different layers and refining modelling studies.
This is where the availability of satellite-derived sea-ice parameters becomes crucial for model development. Enhancing these parameters, especially snow concentration, can improve the efficiency and accuracy of the OHC model for the Arctic region in the future. Prakash said that the research team overcame the challenge of scarce in-situ data by building an optimum ANN architecture to model the spatial, temporal, and depth variabilities of Arctic OHC with greater accuracy than previously possible. He added, “As a result, the research provides an elaborate and comprehensive framework of ice-covered Arctic heat content estimation in a near real-time and wide coverage of satellite observation data.”
The study uses satellite data products like sea ice concentration, sea ice thickness, surface temperature, ambient air temperatures, and snow depth. Daily sea ice thickness and surface temperature products from the APP-x product suite were used in the study. Surface and 2m air temperatures from satellite observations over the Arctic region were utilised. Snow depth data were collected from the TOPAZ4 reanalysis products.
In combination with the satellite data products, the researchers used data from instruments like the WHOI-ITP, which measures temperature and other properties of the ocean under the ice.
A promising model
The researchers developed an ANN model to estimate changes in OHC based on various sea ice thermodynamic parameters. The model is based on theoretical considerations about various factors affecting heat transfer in the region, including heat advection by Atlantic and Pacific waters, heat exchange at different boundaries (ocean-atmosphere, ocean-continent, ocean-seabed) and sea ice state (thickness, extent, properties).
ANN is a machine learning technique that learns patterns from data and establishes relationships between inputs and outputs. They experimented with different configurations of the ANN architecture, including the number of hidden layers, number of neurons, activation functions, and scaling techniques. The researchers divided their data into different sets to develop and test their model. They wanted their data to cover different regions and times to account for the various factors that influence the ocean’s heat content. They used about 13,932 samples for model development and 5,995 samples for independent validation.
The ANN model takes these inputs, processes them through multiple layers, and produces an estimate of OHC change. The performance of the model is assessed using various statistical metrics. Finally, a comparison is made between the model-derived OHC values and the OHC values obtained from the Multi Observation Global Ocean ARMOR3D L4 analysis system. The model accurately estimates OHC changes at different depths across a 0.25° spatial scale, considering various sources of uncertainty and minimizing data noise. The model also provides a promising tool for estimating spatial and temporal OHC changes in the ice-covered Arctic and has the potential to be further refined for deeper layers.