Deep Learning-based Methodology to Predict Abnormally High Water Temperatures using Satellite Big Data

Abstract

Over the last few years, abnormally high water temperature (AHWT) phenomena have occurred more often around the Korean Peninsula. These phenomena damage extensively to the maritime economy by causing a mass stranding of farmed fish. Also, AHWT causes illnesses by exacerbating the propagation of Vibrio pathogens. To mitigate damages caused by AHWT, it should be responded as quickly as possible or forecast in advance. In this paper, therefore, I propose a deep learning-based methodology to predict AHWT occurrences using the satellite data. Thus, to achieve my goal, it is necessary to set up high-performance computing and storage systems for efficiently processing the large-scale satellite dataset. The AHWT phenomenon is dependent not just on the air temperatures but on the change of various oceanic conditions (e.g., ocean currents, sea surface winds, sea salinities, etc.). Consequently, I need to organize and analyze the ocean satellite big data in order to determine the most significant input data in training deep learning models. Then the probability of AHWT occurrence is estimated from the trained model. I expect that this study will contribute to mitigating the damages from AHWTphenomena and preventing the destruction of aquaculture industry environments.