A machine learning-based monthly streamflow forecast model for Alberta

该项目将为艾伯塔省的两个水文站开发一个基于机器学习(ML)的月度径流预报工具：莱斯布里奇的奥德曼河和迪克森大坝的红鹿河。该工具将提供3-9月的月度预测(每个季节7次预测)，通过纳入前一个月的数据逐月更新，以改进后续的月度预测。机器学习算法在水文学领域有着广阔的应用前景，越来越广泛地应用于径流预报、径流历时曲线估计、降雨径流模型和城市需水量预测。该项目将测试几种最大似然算法的性能，包括人工神经网络(ANN)、支持向量机(SVM)、极限学习机(ELM)和径向基函数网络(RBF)，用于月度径流预报。这些预报对于水库的优化运行、春播和提高水利用效率至关重要，并有助于提供早期的洪涝和干旱预警。此外，气候变化推动的水文循环加剧，以及相应增加的洪涝和干旱事件的频率和规模，使这种预报工具变得越来越重要和有价值。一旦模型的性能被评估并被认为是可接受的，径流预报将由合作组织艾伯塔省环境和公园在线发布，供加拿大灌区、市政当局、工业、企业和艾伯塔省站下游的公众使用。预测的逐月动态更新及其打算直接应用于业务都是这项研究的新特点和重要特点。

This project will develop a machine-learning (ML) based monthly streamflow forecast tool for two hydrological stations in Alberta: the Oldman River at Lethbridge and the Red Deer River at Dickson Dam. The tool will provide monthly forecasts for March-September (7 predictions per season), with month-by-month updating through incorporating data from the previous month to improve subsequent monthly forecasts. Machine-learning algorithms offer great promise in the field of hydrology, and are increasingly widely-applied for streamflow forecasting, estimation of flow duration curves, rainfall-runoff modelling, and urban water demand forecasting. This project will test the performance of several ML algorithms, including artificial neural networks (ANN), support vector machines (SVM), extreme learning machines (ELM), and radial basis function networks (RBF), for monthly streamflow forecasts. Such forecasts are critical for optimal operation of reservoirs, spring seeding, and water use efficiency improvements, and help to provide early flooding and drought warning. Further, climate change-driven intensification of the hydrological cycle, and its corresponding increase in frequency and magnitude of flood and drought events, makes such forecasting tools increasingly important and valuable. Once the performance of the models is evaluated and deemed to be acceptable, the streamflow forecasts will be postable online by the partner organization, Alberta Environment and Parks, for use by Canadian irrigation districts, municipalities, industries, businesses, and the public downstream of the stations in Alberta. Both the dynamic month-by-month updating of the forecasts and their intended direct application to operations are novel and important characteristics of the research.