Nowadays, the massive amount of spatio-temporal data available exceeds the human capability to absorb them (i.e., to achieve insights). A possible approach to address this issue is through less detailed representations of phenomena so that the data complexity can be decreased making easier for the users to achieve meaningful insights. In this paper, we discuss the state of the art of modeling spatio-temporal phenomena at different levels of detail (LoDs). We found that granularities play an important role to hold spatio-temporal data at different LoDs. A novel granularity framework is proposed, allowing the definition of a granularity over any domain (including spatial and temporal granularities) as well as it allows transposing knowledge from the original domains to granularities (i.e., known relationships and its properties on the domain). Finally, a granularities-based model is proposed, based on the proposed granularity framework, for dealing and relate different LoDs of spatio-temporal data.