Wind energy represents a leading renewable resource for the production of clean and sustainable electricity for on- and off-grid networks. Nevertheless, tower and foundation costs, which typically represent about 30 percent of the total installation cost of existing wind turbines, limit the operating altitude (hub height) of conventional turbines to no more than 220m. Consequently, conventional systems are not able to utilize significantly stronger winds that are present at higher altitudes. Airborne wind energy (AWE) systems eliminate both the tower and foundation by using tethers and a lifting body to reach higher altitudes where stronger wind exists. In the target installation sites, it is desirable to maximize the percentage of total energy generated from the wind, recognizing that the AWE system will need to be supplemented with conventional sources. This leads to two critical control challenges: (i) Optimizing the operating altitude of an AWE system to maximize the energy generated from the wind and (ii) developing a supervisory controller for an integrated AWE-battery-generator system, recognizing that the optimal control of the overall system requires strategic coordination of the three elements. This dissertation describes and validates, using real wind data, a statistical modeling and hierarchical control approach to addressing the aforementioned challenges. Specifically, physics-driven models are used to characterize the AWE system itself, whereas statistical models are used to characterize the stochastically varying wind profile and electricity demand. A hallmark of the proposed modeling approach lies in the development surrogate regret metrics from these statistical models (termed the energy deficit metric for a stand-alone AWE system and generator excess metric for the integrated system), which provide estimates of the difference between the optimal output of the system and the existing output of the system. Ultimately, these surrogate regret metrics are used to manage a balance between exploration and exploitation of the spatiotemporally varying environment, through several novel candidate hierarchical control structures. Each of the hierarchical control structures fuses coarse, global control at an upper level with fine, local control at a lower level, using a combination of model predictive control and extremum seeking tools.