A novel single-parameter approach for forecasting algal blooms.
A novel single-parameter approach for forecasting algal blooms.
复制标题
DOI：
10.1016/j.watres.2016.10.076
复制
发表时间：
2017
期刊：
影响因子：
12.8
通讯作者：
Xi Xiao;Junyu He;Haomin Huang;Todd R. Miller;G. Christakos;Elke S. Reichwaldt;A. Ghadouani;Shengpan Lin;Xinhua Xu;Jiyan Shi
中科院分区：
文献类型：
--
作者：
Xi Xiao;Junyu He;Haomin Huang;Todd R. Miller;G. Christakos;Elke S. Reichwaldt;A. Ghadouani;Shengpan Lin;Xinhua Xu;Jiyan Shi
有害的藻类大量繁殖在全球频繁发生，预报可成为控制藻类大量繁殖的一项重要的主动战略。为降低水环境监测成本，提高水华预测精度，提出了一种小波分析与人工神经网络相结合的单参数水华预测方法，并以中国泗岭水库和美国温尼贝戈湖的藻类密度日监测数据为例进行了验证。以中国水库蓝藻细胞密度预测为例，详细说明了建模过程。三个小波神经网络模型占用不同的预测时间间隔，通过使用早期停止训练方法的模型训练进行优化。所有模型均能较好地拟合历史数据和预测蓝藻细胞密度的动态变化，其中最好的模型能提前一天预测蓝藻细胞密度（r= 0.986，平均绝对误差= 0.103 × 104cells mL−1）。其次，这种新方法的潜力进一步证实了精确预测的藻类生物量动态aschl在两个研究地点，证明了其高性能的预测藻类水华，包括蓝藻以及其他开花物种。第三，将小波神经网络模型与现有的藻类预测方法（如人工神经网络、自回归积分滑动平均模型）进行了比较，发现小波神经网络模型的预测精度更高。此外，这种新的单参数方法的应用是成本有效的，因为它只需要一个荧光探针，这仅仅是一个典型的自动监测系统的成本的一小部分（约15%）。因此，新开发的方法为今后预测和管理有害藻华提供了一个有前途和具有成本效益的工具。
Harmful algal blooms frequently occur globally, and forecasting could constitute an essential proactive strategy for bloom control. To decrease the cost of aquatic environmental monitoring and increase the accuracy of bloom forecasting, a novel single-parameter approach combining wavelet analysis with artificial neural networks (WNN) was developed and verified based on daily online monitoring datasets of algal density in the Siling Reservoir, China and Lake Winnebago, U.S.A. Firstly, a detailed modeling process was illustrated using the forecasting of cyanobacterial cell density in the Chinese reservoir as an example. Three WNN models occupying various prediction time intervals were optimized through model training using an early stopped training approach. All models performed well in fitting historical data and predicting the dynamics of cyanobacterial cell density, with the best model predicting cyanobacteria density one-day ahead (r= 0.986 and mean absolute error = 0.103 × 104cells mL−1). Secondly, the potential of this novel approach was further confirmed by the precise predictions of algal biomass dynamics measured aschl ain both study sites, demonstrating its high performance in forecasting algal blooms, including cyanobacteria as well as other blooming species. Thirdly, the WNN model was compared to current algal forecasting methods (i.e. artificial neural networks, autoregressive integrated moving average model), and was found to be more accurate. In addition, the application of this novel single-parameter approach is cost effective as it requires only a buoy-mounted fluorescent probe, which is merely a fraction (∼15%) of the cost of a typical auto-monitoring system. As such, the newly developed approach presents a promising and cost-effective tool for the future prediction and management of harmful algal blooms.