Eutrophication caused by human activities worldwide can lead to the formation of harmful algal blooms (HABs) in freshwater lakes, reservoirs, and marine environments. HABs not only deplete dissolved oxygen in water bodies, causing aquatic organisms to suffocate but also release toxic substances, severely damaging fishery resources and aquatic ecosystems. The biological mass concentration and community structure of phytoplankton are critical indicators for assessing aquatic ecological conditions. Real-time monitoring of these parameters is essential for early warning of HABs and environmental protection. In vivo fluorescence spectroscopy, a rapid analytical method based on the fingerprint fluorescence characteristics of phytoplankton, is a promising technology with the potential for online automation and in situ measurements. However, due to the instability of in vivo fluorescence spectra and the limitations of traditional spectral library construction methods, existing spectral analysis methods face challenges such as slow analysis speed, poor quantitative stability, and low classification accuracy. To address these issues, we propose a high-precision analysis method for the overlapping spectra of phytoplankton based on an interpolated reference spectral library (HPRA_method). This method demonstrates excellent accuracy and stability in experiments involving five phyla of phytoplankton, providing a new technical approach for monitoring and protecting aquatic ecosystems.

We focus on five phyla of phytoplankton. Laboratory-standard protocols are used to cultivate the experimental algae, and samples from the mid-growth phase are selected for three-dimensional fluorescence spectroscopy measurements and the standard chlorophyll-a mass concentration. First, the three-dimensional fluorescence spectral characteristics of different phytoplankton phyla are analyzed, and an interpolated reference spectral library is constructed based on the variation range of mass concentration-normalized spectra at different growth stages. Second, a hierarchical partitioning analysis method is proposed. In the first layer, the full spectral range is used to analyze Cyanophyta and Cryptophytes, while in the second layer, specific spectral regions (excitation: 370?650 nm; emission: 650?720 nm) are used to analyze Chlorophyta, Bacillariophyta, and Pyrrophyta. Subsequently, a greedy algorithm-based spectral library search algorithm is developed by integrating the interpolated reference spectral library with the least squares method. Furthermore, a dual-validation algorithm is proposed to enhance the reliability of chlorophyll-a mass concentration analysis by evaluating the similarity between spectral data and reference chlorophyll-a mass concentration. Finally, the HPRA_method is applied to analyze the spectral test set, and the results are compared with those obtained using the standard methods and traditional analysis methods (INLS_method).

The experimental results demonstrate that the HPRA_method outperforms the INLS_method in both total chlorophyll-a mass concentration measurement and classification accuracy. For single and mixed algal samples, the fitting curve k-values of the HPRA_method are 0.952 and 0.990, with correlation coefficients (R2) of 0.964 and 0.919, respectively. The confidence bandwidths are 4.8% and 10.6%, and the prediction bandwidths are 51.7% and 53.2%, respectively. In classification analysis, the mean absolute relative errors (MARE) of chlorophyll-a mass concentration for Cyanophyta, Chlorophyta, Bacillariophyta, Pyrrophyta, and Cryptophytes are 7.4%, 15.1%, 17.8%, 13.1%, and 8.9%, respectively. No non-target classifications are detected for Cyanophyta, Chlorophyta, and Cryptophytes, whereas Bacillariophyta and Pyrrophyta exhibit non-target classifications at average frequencies of 0.07 and 0.18, respectively. Furthermore, the minimum iteration counts and average running time for HPRA_method are 3×102 and 3.2 s per sample, respectively. These values represent only 0.1% and 0.12% of those required by INLS_method, indicating significantly higher analytical efficiency.

We propose an HPRA_method for in vivo phytoplankton fluorescence spectra based on an interpolated reference spectral library. Experimental results show that this method can accurately measure and classify the total chlorophyll-a mass concentration of five phytoplankton phyla, with significantly better classification accuracy and stability compared to traditional fluorescence spectroscopy methods. Particularly, HPRA_method exhibits higher precision and stability, especially in classifying algae with highly similar spectral features (e.g., Chlorophyta, Bacillariophyta, and Pyrrophyta) and detecting low-mass-concentration non-target algae. Additionally, the method offers significantly faster analysis speed and efficiency, providing a new technical approach for the rapid assessment of phytoplankton biological mass concentration and community structure. This research lays an important foundation for future aquatic ecological monitoring and environmental protection efforts.