Development of Multiple Linear Regression Models for Predicting Chronic Iron Toxicity to Aquatic Organisms

Area of Science:

• Environmental toxicology and chemistry
• Aquatic toxicology
• Ecotoxicology

Background:

• Iron (Fe) toxicity to aquatic organisms requires accurate prediction for establishing water quality guidelines (WQGs).
• Toxicity is influenced by water chemistry parameters like dissolved organic carbon (DOC), hardness, and pH.
• Existing models may not fully capture species-specific responses to iron toxicity under varying environmental conditions.

Purpose of the Study:

• To develop multiple linear regression (MLR) models for predicting iron toxicity to aquatic organisms.
• To identify and quantify the effects of toxicity-modifying factors (TMFs) such as DOC and pH on iron toxicity.
• To derive site-specific water quality guidelines (WQGs) for iron using developed models and established methodologies.

Main Methods:

• Developed multiple linear regression (MLR) models using toxicity data for Ceriodaphnia dubia, Pimephales promelas, and Raphidocelis subcapitata.
• Evaluated the influence of dissolved organic carbon (DOC) and pH as toxicity-modifying factors (TMFs).
• Integrated MLR models with species sensitivity distributions (SSD) and applied European Union and US Environmental Protection Agency (USEPA) approaches for WQG derivation.

Main Results:

• MLR models demonstrated good performance (adjusted R² of 0.68-0.89) in predicting iron toxicity.
• DOC and pH were identified as significant TMFs for P. promelas and R. subcapitata; DOC was a TMF for C. dubia.
• Derived WQGs ranged from 114 to 765 μg l⁻¹ Fe (EU approach) and 199 to 910 μg l⁻¹ Fe (USEPA approach) across evaluated TMF conditions.

Conclusions:

• Species-specific MLR models are effective for predicting iron toxicity and deriving WQGs.
• An Excel-based tool was developed to facilitate the calculation of iron WQGs under various TMF conditions.
• Further research on the pH sensitivity of aquatic insects to iron toxicity is recommended due to model uncertainties.