Comparative Analysis of Ocean Color Forecasting Models: A Case Study on Indian Southern Peninsula Using Sarima and Deep Learning Approaches

Abstract

In this research paper, the main goal is to compare ocean color forecasting models, including classical methods like SARIMA and various deep learning models such as multilayer perceptrons, Long short-term memory networks (LSTM), recurrent neural networks, convolution-based networks (CNN-LSTM), dilated causal convolutions (DC-CNN), Gated Neural Network (GRU), and SeriesNet. The study utilizes 18 years of monthly mean data from the Indian Southern Peninsula. The collected data is processed to remove outliers using residual analysis and Facebook Prophet Library, and further denoised using Fast Fourier Transform. Model Accuracy is evaluated using metrics like Root mean square error (RMSE), mean absolute percentage error, symmetric mean absolute error, mean squared logarithmic error, and R2 score. The study aims to identify the most suitable model for accurate ocean color forecasting, with SARIMA emerging as the best-fit model, achieving an RMSE score of 0.0086.