Fluorescence Spectra Predict Microcystin-LR and Disinfection Byproduct Formation Potential in Lake Water

Disinfection byproducts (DBPs) and algal toxins can be expensive to monitor and represent significant potential risks to human health. DBPs, including haloacetic acids and trihalomethanes, are possible or probable human carcinogens. Microcystin-LRproduced by cyanobacteriais linked with various adverse health effects. Here we show that fluorescence spectra predict both microcystin-LR occurrence and DBP formation potential (DBPfp) in lake water. We compared models with either fluorescence spectra or a suite of water quality predictors as inputs. A regularized logistic regression model with fluorescence spectral inputs correctly classified 94% of test data with respect to microcystin-LR occurrence, with a 96% probability of correctly ranking a detect/nondetect pair. Regularized linear regression predicted DBPfp based on fluorescence inputs with a combined R2 of 0.83 on test data. A gradient-boosted classifier with seven water quality inputs was comparable in detecting microcystin-LR (91% correct), as was UV254 in predicting DBPfp (combined test R2 = 0.84), but no single parameter matched fluorescence spectra over both predictive tasks. Results highlight the potential for multiparameter monitoring via fluorescence spectroscopy, extending previous work on predicting DBPs alone. As a high-frequency monitoring tool, this approach could supplement mass spectrometric methods that may only be applicable at low frequency due to resource limitations.

消毒副产物（Disinfection byproducts, DBPs）与藻毒素的监测成本高昂，且对人体健康存在显著潜在风险。其中，卤乙酸、三卤甲烷等消毒副产物属于可疑或确认人类致癌物。由蓝藻产生的微囊藻毒素-LR（Microcystin-LR）与多种不良健康效应相关。本研究证实，荧光光谱可同时预测湖水中微囊藻毒素-LR的存在情况与消毒副产物生成势（DBP formation potential, DBPfp）。
我们对比了分别以荧光光谱或多组水质指标作为输入的预测模型。采用荧光光谱作为输入的正则化逻辑回归模型，针对微囊藻毒素-LR的存在情况在测试集上的分类准确率达94%，且对“检出/未检出”样本对的正确排序概率为96%。基于荧光光谱输入的正则化线性回归模型，在测试集上预测消毒副产物生成势的综合决定系数（R²）为0.83。
采用7项水质指标作为输入的梯度提升分类器在微囊藻毒素-LR检测任务中表现相当（分类准确率91%），UV254在预测消毒副产物生成势时的综合测试集决定系数为0.84，但没有任何单一参数能在两项预测任务中都媲美荧光光谱的表现。
本研究结果凸显了通过荧光光谱开展多参数监测的潜力，拓展了此前仅针对消毒副产物预测的相关研究。作为一种高频监测手段，该方法可弥补因资源限制仅能低频开展的质谱分析法监测的不足。