R-VCNN: A Next-Gen Hybrid Ocean Color Retrieval Algorithm for Accurate Chlorophyll-A in the Indian Ocean

Deep learning (DL) methodologies have emerged as powerful tools for modeling complex, high-dimensional, and inherently noisy datasets, rendering them particularly well-suited for retrieving Chlorophyll-a (Chl-a) concentrations from ocean color satellite observations. This study introduces R-VCNN, a novel hybrid framework called Recurrent Vanilla Convolutional Neural Networks (R-VCNN), which synergistically integrates Recurrent Neural Networks (RNNs) and Vanilla Convolutional Neural Networks (VCNNs) augmented by an empirical blend algorithm. Tailored for the Multi-Spectral Instrument (MSI) onboard the Sentinel-2 platform, R-VCNN leverages spatial feature extraction alongside temporal sequence learning to substantially enhance Chl-a estimation accuracy. The model is meticulously trained and validated on a comprehensive set of geographically diverse in-situ observations, ensuring robust generalization across a wide range of aquatic environments. Comprehensive comparative analyses demonstrate that R-VCNN achieves an accuracy of 96.42%, surpassing all benchmark approaches, including standalone CNNs (94.90%), RNNs (94.54%), B1D-CNN, MDN, and classical empirical algorithms such as OC3 and OCI, which exhibit accuracies below 85%. The framework also consistently records the lowest RMSE and MAE values, affirming its superior capability for precise Chl-a retrieval. Furthermore, targeted ablation studies underscore the importance of each architectural component: removing the LSTM module or the feature fusion strategy leads to RMSE increases of up to 47% and noticeable drops in accuracy, validating the critical contributions of these elements. Additionally, during a documented harmful algal bloom (HAB) event, R-VCNN effectively identified spatial anomalies in Chl-a concentrations, highlighting its potential for real-time ocean color monitoring and ecological risk assessment. Overall, this study underscores the transformative potential of advanced hybrid DL architectures like R-VCNN in enhancing the accuracy and reliability of satellite-based oceanographic parameter retrieval, offering scalable, generalizable solutions for marine ecosystem monitoring, environmental management, and climate change mitigation.