Scientific reports from around the world present us with the undeniable fact that the global ecosystem is undergoing severe change. As this shift accelerates, it is ever more critical that we are able to quantify the local effects of such changes, and further, their implications, from our daily life to the biological processes that put food on our tables. In this thesis, we study the application of sensor network technology to the observation and estimation of highly local phenomena---specifically at a scale between ten to several hundred square meters. Embedding knowledge about the observed process directly into the sensor nodes' behavior via dedicated resource management or control algorithms allows us to deploy dense networks with low power requirements. Ecological systems are notoriously complex. In our work we must thus be highly experimental; it is our highest goal that we construct an approach to environmental monitoring that is not only realistic, but practical for real-world use. Our approach is centered on a commercially available sensor network product, aided by an off-the-shelf quadrotor with minimal customization. We validate our approach through a series of experiments performed from simulation all the way to reality, in deployments lasting days to several months. We motivate the need for local data via two case studies examining physical phenomena. First, employing novel modalities, we study the eclosion of a common agricultural pest. We present our efforts to acquire data that is more local than commonly employed methods, culminating in a six month deployment in a Swiss apple orchard. Next, we apply a environmental fluid dynamics model to enable the estimation of sensible heat flux using an inexpensive sensor. We integrate the sensor with a wireless sensor network and validate its capabilities in a short-term deployment. Acquiring meaningful data on a local scale requires that we advance the state of the art in multiple aspects. Static sensor networks present a classical tension between resolution, autonomy, and accuracy. We explore the performance of algorithms aimed at providing all three, showing explicitly what is required to implement these approaches for real-world applications in an autonomous deployment under uncontrolled conditions. Eventually, spatial resolution is limited by network density. Such limits may be overcome by the use of mobile sensors. We explore the use of an off-the-shelf quadrotor, equipped with environmental sensors, as an additional element in system of heterogeneous sensing nodes. Through a series of indoor and outdoor experiments, we quantify the contribution of a such a mobile sensor, and various strategies for planning its path.