Enhancing Biodiversity Conservation and Biological Assessment of Freshwater Ecosystems with Machine Learning

다국어 초록 (Multilingual Abstract)

Freshwater ecosystems, which are biodiversity hotspots, are facing a crisis owing to rapid biodiversity loss caused by anthropogenic disturbances. Changes in freshwater ecosystems bring about transformations in their physical, chemical, and biological aspects, necessitating appropriate management. To conserve biodiversity, both structural and functional aspects of community must be considered to implement appropriate conservation and management policies.
The structure of ecological communities typically pertains to their composition, with a primary emphasis on biodiversity. Indicator species and biological indices have been used to develop conservation strategies and ecosystem assessments. Stoneflies (Plecoptera) are representative indicators of freshwater ecosystems and serve as indicators of clear streams. Many stonefly species face extinction threats and include a high proportion of endemic species. Therefore, understanding the distribution and ecology of stoneflies is crucial.
In terms of the community aspect, a River InVertebrate Prediction And Classification System (RIVPACS)-type model that utilizes community composition for ecosystem assessment predicts the presence or absence of each family and assigns different scores based on the ecological characteristics of the taxa. The health status of the corresponding community is evaluated by comparing the observed and expected scores. In contrast, the functional aspect deals with practical interactions within the ecosystem. A prominent example of the functional aspect of community is the consideration of the functional traits of species and energy flow (food web) within the ecosystem. Functional traits that reflect the life history of species and can capture aspects of ecosystem health may not be well-captured by taxonomic aspects alone. However, previous studies of food webs have faced obstacles because of their complexity and hierarchical nature of food web. Additionally, the development of models that capture these characteristics requires significant effort. In this regard, an approach using a metaweb (an aggregated set of all possible trophic interactions) that considers only the presence or absence of species and feeding relationships can provide a solution for constructing food webs. Employing this approach is expected to facilitate the creation of food webs and enable research on their attributes.
The hypothesis of this study was that abiotic environmental conditions influence the structure and function of ecological communities and that community structure responds to changes in habitat environments. In particular, the aim of this study was to investigate whether the environment influences not only the structural aspects but also the functional aspects of ecosystems. Therefore, this thesis elucidates the connections between various aspects of ecological communities in freshwater ecosystems and the abiotic environment, with the ultimate goal of conducting research that forms a basis for the effective conservation and management of freshwater ecosystems.
Chapter 1 introduces the importance of freshwater ecosystem conservation and management in respond to the threats facing freshwater biodiversity. It discusses key concepts that form the foundation of this research, including community structure and function, food webs, biological indices, and machine learning methods. Furthermore, an overview of the overall flow and framework of the research is provided from Chapter 2 to Chapter 5.
Chapter 2 characterizes the distribution patterns of an important indicator species, stoneflies, in South Korea and identifies the key factors influencing their conservation. The distribution patterns of Plecoptera assemblages at the study sites were analyzed using a self-organizing map (SOM) and hierarchical cluster analysis (HCA) to classify the study sites into seven clusters. The characteristics of Plecoptera assemblages and environmental conditions of each cluster were compared, along with the environmental characteristics of the habitats where key species were found. In total, 32 Plecoptera taxa were recorded, including three endemic species. Among these taxa, four species showed relatively wide and abundant distributions compared with the other species. The clusters based on assemblage patterns exhibited gradients according to various environmental variables at different scales, and the dominance patterns of key species varied among the clusters. Generally, the dominant species were found in cool mountainous streams and negatively affected by turbidity. This study provides valuable insights into conservation and management strategies for Plecoptera.
Chapter 3 focuses on evaluating variations in community composition in response to environmental variables and explores the relationship between taxonomic and functional diversity within the context of reservoir ecosystems. Four categories of functional traits were used to assess functional diversity: generation per year, adult lifespan, adult size, and functional feeding groups. HCA was employed to classify reservoirs based on their community composition, revealing significant differences in physicochemical and land-use variables among the clusters. The use of SOM facilitated the identification of patterns in functional traits, whereas network association analysis revealed relationships among these traits. These findings support the concept of species survival strategies, such as r- and K-selection. Additionally, a positive relationship was observed between functional richness and taxonomic diversity, whereas functional evenness was not significantly associated with taxonomic diversity. These results suggest that different forms of diversity complement each other when identifying biodiversity and establishing new criteria for evaluating the health of freshwater ecosystems.
Chapter 4 investigates the influence of the environment on the food web structure, assesses the extent to which it operates, and employs the metaweb approach. A comprehensive metaweb containing trophic interaction information for all species was constructed based on an extensive literature review. Local food webs were then created based on the metaweb and genus lists, and metrics reflecting the properties of the food webs were calculated. The study sites were classified using SOM based on food web patterns, and were subsequently clustered using HCA. Differences in food web properties and environmental variables between clusters were compared. The metaweb successfully formed local food webs and their relationships were identified. To identify the impact of environmental variables, random forest (RF) models were created for each food web metric using environmental variables. The importance of the environmental variables was evaluated using the mean decrease in impurity (MDI) and Shapley additive explanations (SHAP) values. SHAP-dependent plots were used to elucidate the influence of each variable on food web metrics. The most frequently significant variables varied for the different food web metrics but riffles were the variable most frequently considered crucial. Other variables associated with human disturbances caused changes in the food web structure in proportion to their magnitude. By constructing a metaweb suitable for freshwater ecosystems in South Korea and investigating the relationship between food web characteristics and environmental variables, this study facilitates research on the ecological dynamics of food webs. Moreover, this study provides a foundation for using food-web characteristics for biomonitoring by linking food webs with abiotic environments.
Chapter 5 focuses on the development of a stream ecosystem health assessment method for South Korea, utilizing the RIVPACS-type approach. The RF model was used to predict the presence or absence of benthic macroinvertebrate families, and modified Biological Monitoring Working Party (BMWPK) scores were assigned to each taxon. The average BMWPK score (ASPT) for each site was calculated based on assigned scores. The observed and expected ASPT ratios were used to evaluate ecosystem health. The evaluation results were similar to those obtained using conventional species-based assessment methods. Considering that this study produced results similar to those of previous species-based studies, even at the family level, it provides a more cost-effective evaluation method and suggests the possibility of applying this method in South Korea.
In Chapter 6, I reflect on the content covered in Chapters 2−5. I examine the importance of spatial analysis, excluding the effect of time, and discuss the key environmental variables that influence communities and the reasons behind them. Additionally, I explore the relationship between my study and the environment as a framework for shaping communities. Furthermore, I will discuss the significance of this study and its potential applications.
This study investigated how different aspects of an ecosystem, in terms of both structure and function, change when considering various environmental variables. Although different approaches have been employed to explore ecological communities, common patterns between communities and the environment have been identified across all studies. Additionally, this study revealed that human impacts are expected to be the most significant disturbance, indicating that greater effort should be directed toward conservation and management. The results of this study can be used to establish more effective strategies for freshwater ecosystem conservation and evaluate the health status of freshwater ecosystems by utilizing appropriate structural and functional indicators.

목차 (Table of Contents)

• List of Tables v
• List of Figures vi
• Abstract xii
• Chapter 1: General Introduction 1
• 1.1. Freshwater biodiversity loss 1
• 1.2. Taxonomic and functional diversity 1
• 1.3. Food web 3
• 1.4. Biological indicator and indices 4
• 1.5. Ecological models and data mining 5
• 1.6. Objectives of the study 7
• 1.7. Structure of thesis 7
• Chapter 2: Unraveling the Distribution Patterns and Habitat Conditions of Plecoptera in South Korea 10
• 2.1. Introduction 10
• 2.2. Materials & Methods 12
• 2.2.1. Ecological data 12
• 2.2.2. Data analysis 16
• 2.3. Results 18
• 2.4. Discussion 28
• 2.4.1. Abundance and occurrence patterns 28
• 2.4.2. Influence of environmental variables on Plecoptera assemblages 28
• 2.4.3. Conservation of Plecoptera biodiversity 30
• 2.5. Conclusion 32
• Chapter 3: Diversity of Benthic Macroinvertebrate Assemblages in Reservoirs of South Korea: A Taxonomic and Functional Perspective 39
• 3.1. Introduction 39
• 3.2. Materials & Methods 41
• 3.2.1. Ecological data 41
• 3.2.2. Data analysis 42
• 3.2.2.1. Taxonomic diversity and functional diversity 42
• 3.2.2.2. Community characteristics 42
• 3.2.2.3. Functional characteristics of communities 45
• 3.3. Results 47
• 3.3.1. Patterns of taxonomic diversity 47
• 3.3.2. Functional traits and diversity 52
• 3.3.3. Relationship between diversity indices 53
• 3.4. Discussion 58
• 3.4.1. Taxonomic community structure 58
• 3.4.2. Relationships between functional traits of macroinvertebrates 59
• 3.4.3. Taxonomic diversity and functional diversity 61
• 3.4.4. Limitations of the study 62
• 3.5. Conclusion 63
• Chapter 4: Exploring Food Web Structures and Environment Influence: A Metaweb Approach 69
• 4.1. Introduction 69
• 4.2. Materials & Methods 74
• 4.2.1. Ecological data 74
• 4.2.2. Metaweb construction 77
• 4.2.3. Food web metrices 78
• 4.2.4. Data analysis 80
• 4.2.4.1. Patterning based on food web metrices 80
• 4.2.4.2. Variable importance and effects on food web metrics 83
• 4.3. Results 85
• 4.3.1. Choosing food web metrices 85
• 4.3.2. Relationship between food web metrics, health scores and environment 90
• 4.3.3 Patterning food webs based on their structure 109
• 4.3.4 Variable importance and effects 119
• 4.4. Discussion 132
• 4.4.1. Relationship between food web metrices 132
• 4.4.2. Relationship between food web metrices and environmental variables 133
• 4.4.3. Metaweb approach 135
• 4.5. Conclusion 137
• Chapter 5. Applicability of the RIVPACS-type Stream Ecosystem Health Assessment Methods using Benthic Macroinvertebrates in South Korea 138
• 5.1. Introduction 138
• 5.2. Materials & Methods 140
• 5.2.1. Ecological data 140
• 5.2.2. Data analysis 142
• 5.2.2.1. Classification according to the stream order 142
• 5.2.2.2. Selection of reference sites 144
• 5.2.2.3. Selection of environmental variables 144
• 5.2.2.4. Training of prediction model 145
• 5.2.2.5. Prediction of health index using predictive model 145
• 5.2.2.6. Model evaluation 146
• 5.3. Results 147
• 5.3.1. Community composition 147
• 5.3.2. Selection of environmental variables 147
• 5.3.3. Community prediction model 150
• 5.3.3.1. Model results by taxa 150
• 5.3.3.2. Health assessment results 150
• 5.4. Discussion 154
• 5.4.1. Model error 154
• 5.4.2. Bias in evaluation 154
• 5.4.3. Determining the number of taxonomic groups for prediction 155
• Chapter 6. General Discussion 159
• 6.1. Spatial variation without seasonality 159
• 6.2. Factors influencing community 160
• 6.3. Impact of environment: shaping ecosystem structure and function 162
• 6.4. Multifaceted assessment of the future 162
• 6.5. Implications 164
• 6.6. Concluding remarks 164
• Appendix 166
• References 169