Collaborative embedded systems (CES) typically operate in highly dynamic contexts that cannot be completely predicted during design time. These systems are subject to a wide range of uncertainties occurring at runtime, which can be distinguished in aleatory or epistemic. While aleatory uncertainty refers to stochasticity that is present in natural or physical processes and systems, epistemic uncertainty refers to the knowledge that is available to the system, for example, in the form of an ontology, being insufficient for the functionalities that require certain knowledge. Even though both of these two kinds of uncertainties are relevant for CES, epistemic uncertainties are especially important, since forming collaborative system groups requires a structured exchange of information. In the autonomous driving domain for instance, the information exchange between different CES of different vehicles may be related to own or environmental behavior, goals or functionalities. By today, the systematic identification of epistemic uncertainties sourced in the information exchange is insufficiently explored, as only some specialized classifications for uncertainties in the area of self-adaptive systems exist. This paper contributes an epistemic uncertainty classification scheme for runtime information exchange (EURECA) in collaborative system groups. By using this classification scheme, it is possible to identify the relevant epistemic sources of uncertainties for a CES during requirements engineering.