Energy-economic and coupled human-natural system models are often used to explore potential energy futures and their implications for climate. There are many uncertain assumptions in the human system models that drive those futures, and in previous work we used a traditional Monte Carlo approach to explore socio-economic uncertainties in a multi-sector, multi-region energy-economic-emissions model of the global economy and generate probabilistic ensembles. The amount of data and information generated from these large ensembles is immense and it can be difficult to sort through and extract relevant insights. The goal of this work is to apply a variety of scenario discovery techniques to the probabilistic ensembles in order to extract insights related to energy futures, with a particular focus on the penetration of renewable energy. We apply Classification and Regression Trees (CART) with Random Forest Classifier (RFC) and Time Series Clustering (TSC) to explore key input drivers of the share of renewable generation, how those drivers can vary over time and across regions, different types of pathways for renewables, and relationships among model outputs. We find that the key drivers of renewables can vary significantly based on the policy scenario, region and time period. In particular, the time series clustering revealed interesting dynamics that are missed by looking at individual years. Through this work we demonstrate the value of scenario discovery techniques in drawing insights from large ensembles of energy-projecting models by facilitating the identification of drivers, relationships among variables and areas of the uncertainty space that are particularly interesting or relevant.