A SENSITIVITY ANALYSIS OF RIVER ENVIRONMENT FACTORS THROUGH DEEP LEARNING

Abstract

The water environment of the most important watersheds of Japan generally have not improved
in a considerable manner in the last two decades although central and local governments have made
considerable management and improvement efforts, such as increasing sewerage system coverage rates
nationwide and installing advanced wastewater treatment systems. It is believed that the marginal effects of
these direct efforts have been diminishing. This study seeks to discover the most effective water environment
improvement measures in a wider range other than those direct measures. An artificial intelligence (AI) model
has been constructed with Deep Learning technology by applying the watershed information from 104
watersheds as teacher data to train the AI model. The well-trained AI model is used to identify the effectiveness
of all the direct and indirect water-environment-related factors, ranging from geological/geographical factors,
hydrological/hydraulic factors to socio-economic factors. This study concludes by pointing out that Deep
Learning through big data can reveal and simulate the complicated relationships between river management
goals and diverse water environment factors. It is hoped that this study will contribute to establishing a more
reliable river environment planning and management methodology.