When searching for sensor data, sensor instances, or Sensor Web Enablement (SWE) services the description of the observed phenomenon plays an important role. Obviously, every user searching for sensor data needs to specify in which kind of sensor data he is interested. In current SWE applications, the information about the observed phenomenon is provided as a unique link encoded as a Uniform Resource Name (URN). However, relying on those URNs to perform string based search for sensor observables has serious drawbacks when it comes to realizing advanced sensor discovery tools as the meaning of the observables is ignored. This work presents an approach that makes use of semantic annotations of SWE resources. The presented solution relies on a dictionary for sensor observables, the Sensor Observable Registry (SOR). This dictionary comprises URNs identifying observables, definitions of these observables in natural language, and pointers to formal phenomenon definitions contained in ontologies. This makes it possible to rely on existing reasoning mechanisms for determining equivalent or related observables (e.g., specializations or generalizations) to the one specified by a user. Finally, an approach is presented, how the SOR can be used for enhancing the sensor discovery process by linking it to sensor catalogues and registries.