Extreme Learning Machine model for assessment of stream health using the Qualitative Habitat Evaluation Index

Abstract

The Extreme Learning Machine (ELM) approach was used to predict stream health with a Qualitative Habitat Evaluation Index (QHEI), and watershed metrics. A dataset of 112 sites in Ontario, Canada with their Hilsenhoff Biotic Index (HBI) and richness values was used in the development of two ELM models. Each model used 70 and 30% of the dataset for training and testing respectively. The models show a great fit with Root Mean Square Error (RMSE)=0.12 and 0.33 for HBI and richness test models, respectively. Then, features elimination based on ELM coefficients and coefficient of variation showed a slight increase in the models' RMSE to reach 0.09 and 0.33 correspondingly. Accordingly, this high predictability of the models in this research provide better insights into which factors influence HBI or richness, and suggests that ELM has a better architecture than other machine learning models and ANN to learn complex non-linear relationships. Also, sensitivity analysis expressed channel slope as the most affecting stream-health parameter for stream health. HIGHLIGHTS Evaluating stream health using the QHBI is an uncertain process.; AI models capture this uncertainty and predict the affecting parameters on stream health.; ELM-based models prove they can predict the most affecting QHEI parameters in stream health evaluation.;

Keywords

• extreme learning machine (elm)
• sensitivity analysis
• stream restoration
• watershed metrics