Solar panels have become cheaper and more efficient in recent years, leading to an upward trend in the number of photovoltaic systems, both private and grid connected. Therefore, accurate predictions of solar power production output have become increasingly important. This paper investigates how tree-based machine learning algorithms can be used to forecast solar energy production output in the geographical setting of Stockholm. An experimental feature selection proved beneficial for Gradient Boosting Regression Trees. Of the algorithms investigated Gradient Boosting Regression Trees performed best for hourly and daily forecasts. This study suggests that snow depth could be used as a parameter to improve performance. The temperature variable did not improve performance.