| Abstract Submission No. | ABS-4.1-0177 |
| Title of Abstract | Machine Learning Approaches for Predicting Algal Blooms: A case study from the Indian Coast |
| Authors | Ranith R*, Nandini Menon N, Sanjiba Kumar Baliarsingh, Alakes Samanta, Susmita Raulo |
| Organisation | Nansen Environmental Research Centre (India) |
| Address | Nansen Environmental Research Centre (India) Kochi, Kerala, India Pincode: 682506 E-mail: rajranith@nerci.in |
| Country | India |
| Presentation | Oral |
| Abstract | Harmful algal blooms (HABs) are increasingly frequent along the Indian coast, creating significant ecological, economic, and public health challenges. This study integrates multi-sensor satellite datasets, in-situ observations, reanalysis model data and machine learning models to analyze spatio-temporal trends of HABs and improve predictive capabilities. Ocean colour data from OCCI, were processed to generate time series datasets of chlorophyll-a, Kd490, and turbidity from 2003 to 2024, enabling long-term trend assessments using the Mann-Kendall test and Sen¿s slope estimator. Similarly, sea surface temperature, and wind speed for the region was obtained from Copernicus ¿ CMEMS, and ERA-5 datasets respectively, and processed to derive time series for 2003 to 2024. Using percentile analysis, chlorophyll-based threshold for algal bloom was identified for the coastal and offshore waters. The analysis of the threshold-filtered chlorophyll data revealed a statistically significant increase in bloom frequency along the southwest coast during post-monsoon months, particularly in Kerala and GoM, Mumbai, Gujarat, Odisha, Andhra Pradesh and west Bengal coast, where nutrient enrichment and coastal upwelling intensify bloom formation. Machine learning models, including Random Forest and XGBoost, were trained on historical satellite and environmental data and validated against in-situ field measurements, achieving an average predictive accuracy of around 87%. Chlorophyll-a, SST anomalies, wind and proxies for nutrient input emerged as the most influential variables in predicting bloom events. This integrated approach demonstrates that satellite-based monitoring combined with robust modeling frameworks can provide reliable, near-real-time forecasting of HAB occurrences. The findings highlight the importance of developing operational early warning systems to support fisheries, coastal ecosystem management, and public health advisories. Scaling these predictive tools through cloud-based platforms and integrating them with decision-support systems can significantly enhance the capacity of coastal stakeholders to mitigate the adverse impacts of HAB events under a rapidly changing climate. |
| Keywords | algal blooms, Noctiluca, Northern Indian ocean, machine learning, Random forest, support vector machine |