Mixed Layer Depth Estimation From Multisource Remote Sensing Data Using Clustering-Machine Learning Method

Abstract

The oceanic mixed layer is essential for air–sea interactions, influencing energy exchanges, climate dynamics, and marine ecosystems through its depth, and seasonal variability. Currently, the mixed layer depth (MLD) is estimated using in-situ observations or model data, both of which are costly and resource-intensive. This study develops a clustering estimation model utilising multisource ocean data to enable faster and more accurate MLD estimation. The model accounts for the temperature and salinity characteristics of different oceanic regions. The K-means clustering method was employed to partition the Pacific Ocean, and the lightGBM model was applied to estimate the MLD in individual subregions. Alongside commonly used sea surface parameters, wind stress curl and precipitation were included as inputs. Feature analysis was conducted separately for the models in each partition. The estimated MLD was compared with that of the in-situ data, showing consistency with observed trends and effectively capturing the spatiotemporal characteristics of MLD across seasons and geographic locations. The estimation error (RMSE) was less than 11.2 m. To assess practical applicability, comparative experiments using remote sensing data were performed, highlighting the model's feasibility and utility. By integrating clustering analysis with advanced estimation models, this study provides a novel approach for accurately reproducing the Pacific Ocean's MLD, which is useful for better analyzing the changes in ocean heat flux and vertical dynamics of seawater.