Prediction of Chlorophyll-a Concentration Distribution Using Generative Adversarial Network (GAN) Algorithm on Processed Remote Sensing Images

Abstract

Variability of chlorophyll-a concentration in the West Coastal Waters of Sumatra is affected by seasonal wind patterns and the Indian Ocean Dipole (IOD) climate anomaly phenomenon. Positive IOD phases increase chlorophyll-a concentrations and negative IOD phases decrease chlorophyll-a concentrations. The “Ocean Color” remote sensing technique for estimating chlorophyll-a concentration as a parameter of the water quality has several weaknesses, such as cloud obstruction, water depth, and wave reflections that obscure the image. Therefore, deep learning methods with Generative Adversarial Network (GAN), especially Pix2PixHD can be an alternative for predicting chlorophyll-a distribution. The model was trained using 348 chlorophyll-a plot images for monthly and 731 chlorophyll-a plot images for daily datasets. The model predicts chlorophyll-a concentrations for 2022 monthly and seven days ahead for 2024 daily. The training results show that the model can predict up to four days in the daily period based on the correlation coefficient value in the linear regression that reaches 0.7. However, the prediction accuracy of the model will decrease over time.