Integrating ecological and machine learning approaches for river health assessment using microalgae in tidal-influenced tropical rivers

Abstract

Microalgae are sensitive indicators of aquatic ecosystem health, yet their roles in short tropical rivers under tidal influence remain underexplored. This study aimed to evaluate microalgal distribution, environmental drivers, and machine learning applications for river health assessment in three tidal-influenced rivers in Penang, Malaysia: Sungai Batu Ferringhi, Sungai Tukun, and Sungai Keluang. Ten stations across upstream, midstream, and downstream segments were sampled monthly from August 2023 to July 2024 during spring and neap tide conditions. Overall, 17 water quality parameters were measured, and eight key drivers influencing water quality were identified using principal component analysis. A total of 454 species were recorded, with Bacillariophyta dominating (> 75%). Downstream sites with higher salinity, temperature, and nutrients supported estuarine diatoms (Skeletonema costatum and Cylindrotheca spp.), whereas upstream areas with lower ionic strength and higher dissolved oxygen supported salinity and pollution-sensitive taxa such as Eunotia sp. and Achnanthes sp. Canonical correspondence analysis showed that tidal mixing, salinity, and nutrient gradients explained 88–92% and 95–96% of assemblage variation in the phytoplankton and epilithic algae datasets, respectively. Hierarchical clustering delineated three ecological zones. Six machine learning algorithms were evaluated for phytoplankton and epilithic algae prediction, including artificial neural network (ANN), support vector machine (SVM), decision tree (DT), k-nearest neighbors (KNN), random forest (RF), and extreme gradient boosting (XGBoost). For phytoplankton prediction, ANN, SVM, DT, and KNN achieved comparable accuracies (60%). In the epilithic algae dataset, ANN and SVM performed poorly (33%), whereas RF and XGBoost achieved higher overall accuracies (67%). The results demonstrate that integrating ecological and machine-learning approaches enhances river health assessment and highlights the strong predictive potential of microalgal bioindicators in tropical tidal river systems.