01-05 December 2025
INCOIS, Hyderabad, India.
| Abstract Submission No. | ABS-04-0286 |
| Title of Abstract | Ocean observations imputation using Machine Learning Algorithms |
| Authors | Srinivasu Kotta*, Hasibur Rahman, Siva Srinivas Kolukula, Ponnam Lakshmi Tharun, Venkat Seshu Reddem, Venkata Jampana |
| Organisation | INCOIS |
| Address | INCOIS Hyderabad, Telangana, India Pincode: 500090 E-mail: srinivas.kotta116@gmail.com |
| Country | India |
| Presentation | Oral |
| Abstract | Ocean observations are essential for understanding the marine environment, from short-term processes to long-term climate change studies. They are fundamental for monitoring phenomena such as sea-level rise, ocean warming, ecosystem variability, etc. While surface observations have increased since the advent of remote sensing, subsurface oceanic processes can only be studied reliably using in-situ measurements and model simulations. Ocean reanalysis products are a valuable source for studying subsurface features. High-quality independent observational data are required for the validation of ocean reanalysis products. However, gaps in observational data hinder the reliability of reanalysis products. In this study, we address these challenges by applying machine learning (ML) algorithms to fill data gaps in observed oceanic currents. We use daily surface and sub-surface current observations from 12 OMNI buoys across the North Indian Ocean during 2011-2024. Since remote sensing techniques cannot directly measure total surface currents, we utilize analysis products such as OSCAR and the Copernicus Marine Service (CMS) Glob Current dataset to supplement observations as training data. We first evaluate OSCAR and CMS surface current speeds against OMNI buoy measurements and then use CMS current as predictors. We train to reconstruct missing OMNI buoy current speed data from 2011 to 2022. Further, we use 2023-2024 buoy observations for the validation. Independent evaluation statistics show that the ML algorithm performs very well in estimating the daily current variability. This approach demonstrates the potential of ML-based imputation for enhancing the completeness and reliability of long-term oceanographic datasets, thereby improving studies of climate variability and ocean processes. |
| Are you part of IIOE-2 endorsed project | no |
| Keywords | Observations, Gap filling, Machine Learning, Climate change |
| For Awards | yes |
| Date Of Birth | 21-05-1991 |
| ECSN Registration Number | IIOE2-ECSN-0168 |