极端干旱下河口咸潮上溯对径流过程的响应(Ⅱ):深度学习预测

Response to runoff variations of saltwater intrusion in an estuary (Ⅱ):prediction by deep learning

咸潮上溯是影响沿海地区水安全的重要环境问题。咸潮过程的有效模拟和精准预测是沿海河口地区咸潮防控与水安全保障面临的关键技术问题。针对传统数值模拟和数理统计预测方法在计算效率以及复杂条件下适用性不足的问题,采用深度学习方法建立了广东省东江河口咸潮上溯的预测模型,探讨了咸潮上溯对潮汐以及径流过程变化的响应特征。结果表明:综合物理过程统计指标和神经网络训练的深度学习方法,可以有效解析潮汐、径流以及河口含氯度之间复杂、动态的响应关系,东江河口万江站含氯度预测的决定系数达到0.8以上。极端干旱下东江下游博罗站的径流过程指标对于河口含氯度预测的准确度具有显著影响,是东江河口咸潮预测建模的关键变量。径流对河口咸潮的影响具有时间延迟,博罗站径流对河口含氯度影响的滞后时间为24~48 h。此外,感潮河段的流量往复幅度也对河口含氯度变化具有一定影响,可以作为咸潮预测的新指标。

Saltwater intrusion is an important environmental problem that threatens water security in estuaries under climate change. The effective simulation and accurate prediction of saltwater movement in the estuarine area play import role in preventing saltwater intrusion. However, current methods including traditional numerical simulation and statistical models have limited capability to predict saltwater intrusion efficiently due to the high computational costs and inflexible assumptions of those methods. Given this, an innovative deep learning method was established to predict saltwater intrusion in the Dongjiang Estuary in Guangdong Province, China where the saltwater intrusion problem has become more serious caused by increasing droughts under climate change. The relationship between the changes of runoff to the sea and the estuarine saltwater movement was analyzed to identify the efficient indicators that can be used to predict saltwater intrusion. The results show that the deep learning method combing statistical indicators of the hydrological process can effectively project the complex, dynamic and nonlinear relationships among tides, runoffs to the sea and the changes of chlorine in the estuary. The coefficient of determination in the saltwater intrusion prediction reached above 0.8. It is also found that the average hourly flow rate during the previous 48 hours in the Boluo Station in the lower reaches of the Dongjing River has a significant impact on the accuracy of the chloride prediction. The estuarine chloride decreased significantly when there was a high flow rate of freshwater released to the sea during the drought. Moreover, the effect of runoff on the saltwater intrusion has a time delay and the lag time of the runoff from Boluo Station on the chloride decreases in the estuary is 24~48 h. In addition, the reciprocating amplitude of the flow in the tide-sensitive reaches also has a certain influence on the change of estuarine chloride content, which can be used as a new indicator for saltwater intrusion prediction.