Cyanobacterial harmful algal blooms and the toxins they produce are a global water-quality problem. Monitoring and prediction tools are needed to quickly predict cyanotoxin action-level exceedances in recreational and drinking waters used by the public. To address this need, data were collected at eight locations in Ohio, USA, to identify factors significantly related to observed concentrations of microcystins (a freshwater cyanotoxin) that could be used in two types of site-specific regression models. Real-time models include easily or continuously-measured factors that do not require that a sample be collected; comprehensive models use a combination of discrete sample-based measurements and real-time factors. The study sites included two recreational sites and six water treatment plant sites. Real-time models commonly included variables such as phycocyanin, pH, specific conductance, and streamflow or gage height. Many real-time factors were averages over time periods antecedent to the time the microcystin sample was collected, including water-quality data compiled from continuous monitors. Comprehensive models were useful at some sites with lagged variables for cyanobacterial toxin genes, dissolved nutrients, and (or) nitrogen to phosphorus ratios. Because models can be used for management decisions, important measures of model performance were sensitivity, specificity, and accuracy of estimates above or below the microcystin concentration threshold standard or action level. Sensitivity is how well the predictive tool correctly predicts exceedance of a threshold, an important measure for water-resource managers. Sensitivities > 90% at four Lake Erie water treatment plants indicated that models with continuous monitor data were especially promising. The planned next steps are to collect more data to build larger site-specific datasets and validate models before they can be used for management decisions.

中文翻译：

---

预测俄亥俄州某些湖泊中蓝藻水华引起的微囊藻毒素浓度作用水平超标。

蓝藻有害藻华及其产生的毒素是全球水质问题。需要使用监视和预测工具来快速预测公众使用的娱乐和饮用水中的氰毒素活动水平超标。为了满足这一需求，在美国俄亥俄州的八个地点收集了数据，以鉴定与观察到的微囊藻毒素（一种淡水氰毒素）浓度显着相关的因素，这些因素可用于两种类型的站点特定回归模型。实时模型包括容易或连续测量的因素，这些因素不需要收集样本；全面的模型结合了基于离散样本的测量和实时因素的组合。研究地点包括两个娱乐地点和六个水处理厂地点。实时模型通常包括变量，例如藻蓝蛋白，pH，比电导以及流量或表压高度。许多实时因素是采集微囊藻毒素样品之前一段时间内的平均值，包括从连续监测仪收集的水质数据。在某些位置，对于蓝细菌毒素基因，溶解的营养物和（或）氮磷比而言，具有滞后变量的综合模型非常有用。由于模型可以用于管理决策，因此模型性能的重要衡量指标是敏感性，特异性和微囊藻毒素浓度阈值标准或作用水平以上或以下的估计准确性。敏感性是预测工具正确预测超出阈值的能力，阈值是水资源管理者的重要衡量指标。敏感性> 四家伊利湖水处理厂的90％表示，具有连续监测数据的模型特别有希望。计划的下一步是收集更多数据以构建更大的特定于站点的数据集并验证模型，然后将其用于管理决策。