三种机器学习模型在太湖藻华面积预测中的应用

基于2014—2018年太湖气象水文水质数据与卫星遥感数据，分别采用支持向量机(SVM)、长短记忆神经网络(LSTM)、极端梯度提升树(XGBoost)模型模拟全太湖、贡湖、南部沿岸区、中西北湖区的蓝藻水华(简称藻华)面积。结果表明:(a)XGBoost全太湖与分区藻华面积回归模型模拟效果较好，其次是SVM、LSTM回归模型；不同时间尺度下SVM、XGBoost回归模型对全太湖藻华面积模拟结果偏小，但有效模拟了藻华的发展趋势。(b)XGBoost分类模型在全太湖、中西北湖区模拟准确率较高，优于SVM、LSTM分类模型；在贡湖、南部沿岸区，3种分类模型准确率均较高。(c)以当天、提前1 d的气象水文水质因子作为全太湖与分区藻华面积模型输入，XGBoost回归与分类模型模拟精度较高、稳健性较好，预测应用情景较好。

Area prediction of cyanobacterial blooms based on three machine learning methods in Taihu Lake

Based on atmospheric-hydrological data and satellite remote sensing data from 2014 to 2018, the support vector machine(SVM), long short-term memory model(LSTM), extreme gradient boosting(XGBoost)model were applied to predict the cyanobacterial bloom area in fields including the whole region, Gonghu Bay, southern coastal region and central north-western region of Taihu Lake. The results demonstrated that the XGBoost regression model had better accuracy than SVM and LSTM regression model in the whole region and subdivided regions. Compared with the observed cyanobacterial bloom area, simulated areas of SVM regression model and XGBoost regression model were lower in the Taihu Lake under different time scales, while the development tendency of cyanobacterial blooms was effectively simulated. In addition, the XGBoost classification model had better accuracy than SVM and LSTM classification model for the whole region and the central north-western region of Taihu Lake. Three classification models had high accuracy in the Gonghu Bay and the southern coastal region of Taihu Lake. Finally, taking the atmospheric-hydrological data and water quality data of the same day and one day advanced as model inputs, the XGBoost regression model has high accuracy and robustness in cyanobacterial bloom area simulation, which had a promising application prospect for the cyanobacterial bloom prediction.