Sensor networks use a group of collaborating sensor nodes to collect information about real world phenomena. Sensor nodes use low-power short-range radio links to communicate with each other. Communication between sensor nodes shows significant variation over time and space. This can lead to unreliable and unpredictable network performance. These dynamic and lossy characteristics of wireless links pose major challenges for building reliable sensor networks and raise new issues that data delivery protocols must address. This thesis addresses the problems of designing protocols to overcome time-varying environmental conditions that lead to unpredictable network performance. The goal is to provide reliable data delivery in sensor networks and to minimise energy use. The major contributions of this thesis are: measuring the performance of wireless links in field trials on a time scale of weeks; systematic analysis of strengths and weaknesses of existing data delivery protocols; and the design, implementation and testing of a novel autonomic communication framework. We have measured link quality over time in experiments in unattended outdoor environments. Most previous work focused on spatial properties and experiments were not extensive, only lasting for a few hours. Besides common phenomena found in other work, such as the variation of network performance over time and the existence of asymmetric links, we find that links are independent over long time scales, and performance patterns of links are different. We also analyse the performance of data delivery protocols that use different techniques to improve reliability in sensor networks. Through systematic analysis of strengths and weaknesses of existing data delivery strategies, we find that networks using a single technique can only perform well for a limited range of link conditions. Different strategies are required in different operating conditions. Based on these experimental and theoretical studies, a novel autonomic communication framework (ACF) for wireless sensor networks is proposed. Nodes in this ACF are able to change their behaviour to adapt to time-varying environments so that optimal network performance can be achieved. Our framework provides a holistic solution for reliable data delivery to overcome time-varying wireless links. Our implementation and experimental evaluations demonstrate that this holistic framework is effective for reliable and energy-efficient data delivery in realistic sensor network settings.