Exploring the relationship between the optical properties of water and the quality and quantity of dissolved organic carbon in aquatic ecosystems: strong correlations do not always mean strong predictive power

Abstract

Optical absorbance and fluorescence spectroscopies have the potential to play an important role in monitoring aquatic ecosystems. This paper explores the relationship between the optical absorption and fluorescence characteristics of water samples taken from 18 sites (spanning a range of aquatic environments including lowland rivers, small and large dams, and floodplain wetlands) with their dissolved organic carbon (DOC) concentration and 28 days bioavailability Both optical absorbance in the ultraviolet region and fluorescence above excitation wavelengths of about 240 nm and emission wavelengths above about 350 were correlated with DOC concentration (r² > 0.8). The initial (pre-incubation) optical absorbance in the UV region (r² ≈ 0.7–0.8) and fluorescence attributed to both ‘humic like’ (r² = 0.84) and ‘tryptophan-like’ (r² = 0.87) fluorophores correlated with DOC bioavailability. These correlations were used to develop empirical linear models relating the initial optical properties of water with DOC quantity and quality. The robustness of these models was then tested against a second suite of water samples from 12 different sites, collected independently of those used in the model development. Although based on strong correlations, the empirical models were not particularly good at predicting the bioavailability of DOC in the model validation samples. We suggest that one of the reasons for the low predictive power of the models is that the strong correlations observed between DOC bioavailability and optical absorbance in the UV region or fluorescence and are co-incidental rather than causal. Changes in UV-absorbance or fluorescence during the incubation experiments are not consistent with changes in DOC concentration. One of the best predictors of DOC bioavailability is the initial concentration of DOC. We argue the strong correlation between DOC bioavailability and initial fluorescence intensity or UV absorption simply reflects the strong correlation between initial DOC concentration and initial fluorescence intensity or UV absorption. We argue that unless there is an underlying causal relationship between two components (the component of interest and a surrogate measure for that component) care should be taken in extrapolating correlative models beyond the data set used to create them.