Multi-objective automatic data assimilation of a hydrological model based on classification of initial hydrologic states

点击翻译按钮获取中文摘要

The Power Operations division of Rio Tinto Aluminium (RTA) has the mandate to manage two large water**resources systems, namely the Lac-Saint-Jean system in Québec and the Nechako system in British-Columbia.**To manage the water resources systems efficiently and safely, multiple scenarios of inflow predictions are**produced and are fed into water resources management optimization models. The quality of the forecasted**inflows is crucial as it allows adequately mitigating flooding risks as well as maximizing hydropower**generation for a given volume of water.**Hydrologic models are used to produce inflow predictions and these are first trained on past data and then used**to produce inflow forecasts by driving the model with forecasted inputs like precipitation and temperature. The**initial states of the model are key parameters required to obtain an accurate forecast. To provide the best**possible initial states before starting the forecast simulations, an expert analyst compares the model states with**current observations and corrects these states manually if necessary. This procedure yields reliable short-term**forecasts. The skill of long-term forecasts, however, is poor due to the unclear propagation of the manual**changes through the complex system over a long horizon. This research is intended to automatize the process**of initial state updating using advanced sensitivity analysis and classification algorithms. The automatic**procedure will search for the optimal corrections to achieve both reliable short-term and long-term performance**of the forecasts based on the initial conditions of the catchment. This will help RTA to optimize their**operations using sustainable hydroelectricity.