The short answer is that ensembles are less sensitive to outliers in your training data, which helps them avoid overfitting and generalize better when applied to new data.
A single decision tree is easy to explore and can be useful for understanding patterns in your data. However, single tree models will often overfit making them poorer predictive models. An ensemble of models built on samples of the data can become a powerful predictor by averaging away the errors of each individual model.