Publications

Kolluru, S; Tiwari, SP (2022). Modeling ocean surface chlorophyll-a concentration from ocean color remote sensing reflectance in global waters using machine learning. SCIENCE OF THE TOTAL ENVIRONMENT, 844, 157191.

Abstract
The spatial and temporal variations of Chlorophyll-a (Chl-a) in clear and coastal waters are critical for assessing the health of the marine environment. Machine learning models have been proven to model complex relationships and provide better accuracy estimates of the derived parameters compared to traditional empirical models. The present study proposes a novel approach to derive Chl-a by using multi-layer perceptron Neural Network (MLPNN) with Resilient backpropagation method based on the four ocean color bands existent in most of the ocean color sensors. The NNs are trained on NASA's bio-optical Marine Algorithm Dataset (NOMAD) and tested on three different datasets (i) SeaWiFS and, (ii) MODIS Aqua matchup dataset, and (iii) simulated dataset for the Red Sea. These three datasets cover significant variations range in Chl-a levels under both oligotrophic and eutrophic conditions. The influence of different variations in inputs used in NN training is assessed and hyperparameter tuning of the NN is performed to obtain best NN configuration to derive Chl-a. Accuracy assessment of the present study with other global algorithms are performed by comparing the modeled and observed values of the Chl-a. The performance matrices computed from the developed model were promising. Therefore, this study provides a potential approach for the retrieval of improved Chl-a estimates in the global clear and coastal waters as compared to the traditional blue-green band ratio algorithms. Furthermore, the developed algorithm and existing algorithms are applied to SeaWiFS, MODIS, VIIRS, and Hawkeye satellite ocean color data to demonstrate how it may be utilized to accurately depict the spatial distribu-tion of ocean color features in global waters, phytoplankton blooms and some of the physical processes in the Arabian Sea and the Red Sea. The findings of this work have potential to advance the ocean color remote sensing and biogeo-chemical cycles and processes in coastal and open ocean waters.

DOI:
10.1016/j.scitotenv.2022.157191

ISSN:
1879-1026