Show simple item record

dc.contributor.authorImchen, I.
dc.contributor.authorNeetu, S.
dc.contributor.authorRajmohan, S.
dc.contributor.authorAdithyan, V.
dc.contributor.authorSohan, P. M.
dc.date.accessioned2025-07-04
dc.date.accessioned2025-07-04T06:45:26Z
dc.date.available2025-06-14
dc.date.available2025-07-04T06:45:26Z
dc.date.issued2025
dc.identifier.citationOcean Dynamics,vol.75(6);2025;pp.61
dc.identifier.issn1616-7228
dc.identifier.otherhttps://link.springer.com/article/10.1007/s10236-025-01708-0#citeas
dc.identifier.urihttps://doi.org/10.1007/s10236-025-01708-0
dc.identifier.urihttp://drs.nio.res.in/drs/handle/2264/9483
dc.descriptionIF2023=2.2
dc.description.abstractOne of the most significant variables in the ocean is the mixed layer depth (MLD). It acts as an interface between the ocean and the atmosphere influencing the climate as well as the marine ecosystem. Argo floats are most often used to sample this layer but its spatial and temporal coverage is still not sufficient to understand the MLD variability in the north Indian Ocean. Relative scarcity of in-situ sub-surface data require alternate methods to compute these sub-surface variables. In this context, we explore the prospects of machine learning to derive MLD from easily available reanalysis and remotely sensed surface variables for the two basins in the north Indian Ocean; Arabian sea (AS) and Bay of Bengal (BoB). Using Argo data, we demonstrate the use of multiple machine learning models such as artificial neural network (ANN), support vector machine (SVM), extreme gradient boosting (XGBoost), random forest (RF) and convolutional neural networks (CNN) for the MLD estimation and compare their skills. Though ANN and CNN have been used before for predicting MLD in basins other than north Indian Ocean, SVM, XGBoost and RF model skills are not yet demonstrated. Our results show that XGBoost outperforms all the other models followed by RF, ANN, SVM and then CNN in both the basins. XGBoost also takes the least computational time to train the model. Results also show that deeper MLDs (greater than 50�m) have larger prediction errors compared to shallower MLDs (less than 50�m). Among physical parameters, sea surface salinity (SSS) stands out as the most significant predictor for MLD in the BoB, whereas it has little impact in the AS. This highlights the models' capability to identify the importance of input parameters tied to different oceanic processes. Other important predictors for both the regions are latitude, month, sea level anomaly (SLA) and sea surface temperature (SST). Our results demonstrate that tree-based ensembles (XGBoost, RF) used on tabular data provide very promising results and should be used for the prediction of various sub-surface oceanic variables.
dc.languageen
dc.publisherSpringer
dc.relation.isreferencedbySCI
dc.rightsOnly the author version of the copright reserved papers uploaded. Contact drs@nio.org for any complaints regarding copyright violations
dc.titleEstimation of ocean mixed-layer depth using machine learning techniques
dc.typeJournal Article
dc.affiliation.authorCSIR-National Institute of Oceanography, Dona Paula, Goa, India


Files in this item

FilesSizeFormatView

There are no files associated with this item.

This item appears in the following Collection(s)

Show simple item record