基于ARIMA模型的长潭水库叶绿素a浓度预测

Prediction of chlorophyll a concentration in Changtan Reservoir based on ARIMA model

长潭水库面临水体富营养化风险。叶绿素a(Chl-a)作为水体富营养化的重要指示指标,预测其变化趋势对库区生态健康管理具有重要意义。为此,运用2012—2019年长潭水库坝口站的在线监测数据,分别基于日均值和月均值构建了Chl-a预测的ARIMA模型。结果表明:(1)基于日均值的Chl-a短期预测模型不需要进行差分转换,自回归阶数和移动平均阶数分别取8、0,确定系数(R2)和校正R2均达到0.952,预测时长在9 d内,均方根误差(RMSE)<1倍预测序列标准差,平均绝对百分误差(MAPE)<100%。(2)基于月均值的Chl-a中长期预测模型在季节性调整的基础上加入水温和氮磷比作为协变量,也不需要进行差分转换,自回归阶数和移动平均阶数分别取0、1,季节性自回归阶数和季节性移动平均阶数分别取1、0,季节周期为12,R2和校正R2分别达到0.664、0.647,预测时长在5 a内,RMSE<1倍预测序列标准差,MAPE<100%。

Changtan Reservoir faces the potential risk of eutrophication.Chlorophyll a(Chl-a)is an important indicator of eutrophication,so it is of great significance to predict its change trend for ecosystem health management of Changtan Reservoir.Therefore,ARIMA models were constructed to predict Chl-a based on the daily and monthly average Chl-a from the online monitoring data of Bakou Station during 2012-2019.Results showed that:(1)short-term prediction model based on daily average Chl-a didn’t need differential conversion.Autoregression order and moving average order should be 8 and 0,respectively.Thus,the R 2 and adjusted R 2 both reached 0.952.When prediction time was within 9 d,the root mean square error(RMSE)was<1 time of prediction series standard deviation,and the mean absolute percentage error(MAPE)was<100%.(2)Medium/long-term prediction model based on monthly average Chl-a after adding water temperature and N/P ratio and seasonal adjustment didn’t need differential conversion,either.Autoregression order and moving average order should be 0 and 1,respectively.Seasonal autoregression order and seasonal moving average order should be 1 and 0,respectively.Seasonal period was 12.Thus,the R 2 and adjusted R 2 reached 0.664 and 0.647,respectively.When prediction time was within 5 a,the RMSE was<1 time of prediction series standard deviation,and the MAPE was<100%.