01-05 December 2025
INCOIS, Hyderabad, India.
| Abstract Submission No. | ABS-05-0284 |
| Title of Abstract | Development of AI/ML driven Rip Current Forecasting System for Indian beaches |
| Authors | Surisetty V V Arun Kumar, Harsh Awasthi*, Parisha Raja, Jimil Patel, Harshit Gautam, Seemanth M, Gireesh Baggu, Chintam Venkateswarlu, Sanket J Shah, Ramesh Madipally, Utkarsh, Shivani M Shah, Sanjib Kumar Deb, C V Naidu, L Sheela Nair, Rashmi Sharma |
| Organisation | Space Applications Centre, ISRO, Ahmedabad |
| Address | 18/A, Yoginiketan Plots, Street No.2, Nirmala Convent Road, Kalawad Road Rajkot, Gujarat, India Pincode: 360001 E-mail: harshawasthi1204@gmail.com |
| Country | India |
| Presentation | Oral |
| Abstract | Rip currents pose a major threat to residents as well as tourists visiting coastal areas, responsible for a significant amount of worldwide drownings each year. Traditional empirical models often fail to capture nonlinear interactions among oceanographic parameters, thus falling short in predicting rip currents accurately. This paper presents the development of a machine learning-driven rip current prediction system, ⿿Safe Beach implemented at the Meteorological and Oceanographic Satellite Data Archival Centre (MOSDAC), Indian Space Research Organization (ISRO). The system utilizes a two-stage approach: detection followed by prediction. Rip currents are first identified via a computer-vision object detection model, YOLO-V11, applied to Timex imagery from public video sources as well as Video Beach Monitoring Stations (VBMS). These detections are stored to create a rip current database, that is used to train a stacked ensemble learning framework for the prediction stage. This ensemble framework integrates various approaches & algorithms, including Random Forest, Gradient Boosting, XGBoost, and Support Vector Machines, with a meta-learner for final prediction. Input parameters include significant wave height, wave period, wave direction, spectral directional spread (from satellite-assimilated WaveWatch III model), and predicted tidal elevation. The system is currently issuing experimental forecasts at 175 Indian beach locations, providing predicted nearshore wave parameters such as breaker height, surf similarity, and breaker angle, along with rip current risk levels. The YOLO-V11 model shows around 90% precision in detection, while the ensemble model demonstrates over 85% prediction accuracy along with reduced false positives compared to traditional methods. Implemented in Python, the system operates autonomously, ingesting daily WaveWatch III outputs and generating five-day rip current forecasts at six-hour intervals. Safe Beach is accessible at https://www.mosdac.gov.in/safebeach. It supports coastal management, lifeguards, and public advisories, demonstrating the potential of AI and ML in operational coastal hazard forecasting and making an impactful contribution to public safety. |
| Are you part of IIOE-2 endorsed project | no |
| Keywords | Rip Currents, Forecasting, Machine Learning, Modelling |
| For Awards | yes |
| Date Of Birth | 12-04-2003 |
| ECSN Registration Number | IIOE2-ECSN-0196 |