Discovering motifs in time series data has been widely explored by recent researches due to the challenge regarding the variety of data characteristics from real-world datasets. It is being motivated by the potential to identify relevant implicit information.Many time series data mining techniques were adapted to solve the problem regarding motif identification in time series. In this context, some motif discovery approaches were proposed. It can be divided as exact approach that uses brute-force algorithm and approximated that are based on Random Projection algorithm. Other characteristics such motif-length and dimensionality were object of researches and works. However, the environment around spatial-time series data still an open gap according to the literature in the area. This work proposed an new approach to discover motif in spatial-time series based on combining series to transform the spatial-time series data into single time series to then apply a Random Projection algorithm. For the spatial-time series motif identified a clustering function is applied to group correlated motifs and then ranked based in a proposed function. The approach is validated in a seismic dataset that is data intensive area which produces and process real big datasets composed by sets of spatial time series. It presented impressive results identified with accuracy the seismic horizons, an important characteristics in seismic analysis.