In response to the urgent global call for renewable energy alternatives, as dictated by the Paris Agreement, there is an increasing need to accurately map and monitor the growth of solar farms around the world. A critical component of achieving net zero emissions by 2050 is a clear understanding and accurate tracking of photovoltaic solar energy capacities globally. However, the undocumented nature of many of these installations presents a serious obstacle. To address this challenge, this thesis develops Solis-seg, a Deep Neural Network designed to detect and segment solar farms in satellite imagery. The Solis-seg model pushes the boundaries of current capabilities in photovoltaic detection, attaining a mean Intersection over Union (IoU) score of 96.26% on a dataset covering approximately 30,000 solar farms in Europe. As demonstrated in comparative experiments as reported and discussed in this work, this performance surpasses any previous result on a continental-spanning dataset reported in the literature. As part of this work, we apply Neural Architecture Search (NAS) to the problem of segmenting solar farms in satellite imagery, thus unveiling significant insights and potential avenues for future exploration. In doing so, it assesses the practicality of NAS in an important sustainability context. Furthermore, this thesis offers a critical reassessment of the widely endorsed method of utilizing transfer learning from classification tasks for semantic segmentation. Therefore, this research is a meaningful contribution to the field of satellite imagery analysis, encouraging experimentation with advanced techniques in the rapidly developing domain of machine learning for earth observation. By underscoring the escalating importance of renewable energy resources and offering an efficient and scalable solution to track global progress towards sustainable energy goals, this thesis aligns with the broader goal of facilitating the global energy transition towards sustainable sources.