Application of a multi-wavelength excitation fluorometer and machine-learning algorithm for on-site measurement of long-chain polyunsaturated fatty acids in lakes

Aoki N, Fujibayashi M, Sakamaki T, Kuba T. 2025. Application of a multi-wavelength excitation fluorometer and machine-learning algorithm for on-site measurement of long-chain polyunsaturated fatty acids in lakes. Lake Reserv Manage. 41:112–123. Long-chain polyunsaturated fatty acids, such as eicosapentaenoic acid (EPA) and docosahexaenoic acid (DHA), are important indicators of the nutritional quality of diets for aquatic animals. Seston contributes the primary basal food source; therefore, monitoring the contents (absolute amount per mass or volume) and levels (mass percent of total fatty acids) of these fatty acids in seston is a rational strategy for lake and reservoir management to maintain ecosystem health. However, conventional methods for EPA and DHA analyses in seston are labor-intensive and complex. This study aimed to develop a simple approach for field measurement of EPA and DHA using a multi-wavelength excitation fluorometer (MEX), which was originally designed to estimate the Chl-a concentration derived from each phytoplankton class based on excitation spectra. We proposed that a model could be developed by incorporating machine learning to predict EPA and DHA from excitation spectra. Water samples (151) were collected from the Hokuzan Reservoir in Saga Prefecture. The excitation spectra for each water sample were obtained on-site by MEX, and the EPA and DHA in the seston were analyzed in the laboratory using conventional methods. Based on these data, predictive models were developed using an extra-tree regressor, the one of machine learning algorithms. The model achieved average R² values of 0.78, 0.74, 0.83, and 0.80 for EPA contents, DHA contents, EPA levels, and DHA levels, respectively. Consequently, using MEX enables novel and simple on-site measurements of EPA and DHA, further facilitating spatiotemporally dense data acquisition and potentially improving lake and reservoir ecosystem management.