Application Study of Artificial Intelligence and Big Data to Environment Systems: Water and Atmosphere Systems

Abstract

Data-driven models are one of the widely used technologies to support decision-making in the field of the environment. In particular, although artificial intelligence (AI) models have been advanced significantly since their introduction to environmental systems, the application is considerably lower compared to other fields. AI algorithms require a lot of data to recognize patterns of systems. Therefore, the performance of the models mainly depends on the quantity and quality of the data. However, the quality of the data in the environmental field, which is highly dependent on on-site monitoring, is not enough. It means that not only the selection of an algorithm but also the construction of big data that can be used for training plays an important role in performance. This study conducted applied research on the construction of big data and AI models appropriate for analyzing environmental systems such as freshwater ecology, watersheds, and atmosphere. In the freshwater ecosystem, a method to objectively predict the occurrence of non-native fish species was proposed using a machine learning algorithm. In order to build big data that can be applied to the model, the global database provided by two different fields, water environment and ecology, was integrated. In the watershed system, an AI algorithm was used to improve the water quality prediction accuracy of the deterministic model (soil and water assessment tool, SWAT). The output of the deterministic model was employed as additional data to increase the quality of the data for AI. The prediction accuracy was evaluated by comparing it with a deterministic model. In the atmosphere system, we developed a deep learning model that estimates particulate matter (PM) using CCTV and related observation data. The convolution neural network (CNN) specialized for image processing was used. The AI models and new significant big data construction approaches, commonly proposed in this study, can be used as a reference protocol for analyzing new systems.