Understanding the responses of stream ecosystems to environmental disturbances is essential for maintaining and restoring healthy ecosystems. In this study, we analyzed the associations between benthic macroinvertebrate communities and environmental factors using machine learning approaches to identify key stressors potentially influencing stream ecosystem health. Various machine learning models were evaluated, with random forest (RF) and gradient boosting machine (GBM) identified as the optimal models for predicting tolerant species (TS) and Ephemeroptera, Plecoptera, and Trichoptera (EPT) species densities. SHAP analysis revealed that watershed variables, such as elevation, flow velocity, and slope, significantly influenced EPT and TS populations. EPT population density increased with elevation and flow velocity but decreased significantly with higher levels of biochemical oxygen demand (BOD), total nitrogen (TN), and agricultural land-use proportions, with negative effects becoming evident beyond threshold levels. Conversely, TS population density showed a positive response to elevated BOD, TN, and agricultural land-use proportions, stabilizing at the threshold levels of BOD and TN, but continuing to increase with greater agricultural land use. Through machine learning, this study provides critical insights into how environmental variables are associated with the distribution of benthic macroinvertebrate communities. By identifying threshold levels of key stressors, this approach offers actionable guidance for managing agricultural runoff, enhancing riparian buffers, and implementing sustainable land-use practices. These findings contribute to the development of integrated watershed management strategies that promote the long-term sustainability of stream ecosystems.
Read full abstract