Deep Learning methods have become increasingly popular for time-series forecasting in recent years. One common way of improving time-series forecasts is to use ensembles. By combining forecasts of different models, for example, calculating the mean forecast, it is possible to get an ensemble that performs better than every single member. This paper further investigates methods of combining multivariate multi-step forecasts from an ensemble of model forecasts using deep learning, specifically supervised and reinforcement learning. We test our two approaches on 5 water levels at small to medium-sized rivers, datasets that have strong seasonality and rare events. Both approaches outperform single models. The supervised approach outperforms naive ensembles and is trained significantly faster than the reinforcement learner.