Integrating AQUATOX, Ecological Big Data, and Machine Learning for Short-Term Chemical impact Assessment on River Ecosystems

Abstract

This thesis develops a methodology for assessing the impact of chemical accidents on aquatic ecosystems using ecological modeling. The study consists of three main parts: First, it generates a quantifiable species-specific ecological parameter database for domestic aquatic ecosystems. This involves estimating representative biomass for 687invertebrate species and 157 fish species in Korean rivers using length-weight relationships and literature data. The study proposes and evaluates a method for applying these parameters to ecological models of domestic river ecosystems. The most accurate method (R2 = 0.5529 for invertebrates, R2 = 0.9349 for fish) used genus-level averages of family-level averages of length- weight constants. Second, the research parameterizes food web interactions between aquatic species in Korean rivers. It employs a Deep Neural Network model, constrained by ecological domain knowledge, to estimate food preferences among species groups. The model is trained on big data from species presence surveys across various regions. The optimal model (R2 = 0.8953, RMSE = 0.2369) was achieved when the maximum trophic level for species consuming producers was set at 3.2, and predation was allowed only when the predator's trophic level exceeded that of the prey. Third, the study applies the AQUATOX model to assess the impact of a phenol spill on the Iksan Stream ecosystem. The model is constructed using water quality, flow, and ecological data from various monitoring points along the stream. A control simulation is established to represent the stream's ecosystem without the phenol spill, followed by a perturbed simulation incorporating observed phenol concentrations (maximum 2.9 mg/L). The model's performance is evaluated using Chlorophyll a concentrations, with a no improvement in the perturbed scenario (zero increase in Nash-Sutcliffe Efficiency). The results demonstrate complex ecosystem responses to the phenol spill, highlighting the importance of considering food web interactions and long-term effects in chemical spill impact assessments. This research contributes to the development of more comprehensive environmental risk assessments and management strategies for aquatic ecosystems in Korea.

Yeom, Jaehoon (염재훈). Integrating AQUATOX, Ecological Big Data, and Machine Learning for Short-Term Chemical impact Assessment on River Ecosystems (AQUATOX 모델, 생태 빅데이터, 머신러닝을 활용한 하천 생태계의 단기 화학물질 노출 영 향 평가). School of Environment and Energy Engineering. 2025. 235p. Prof. Sang Don Kim