Spatial cluster analysis is the grouping of similar spatial objects into classes (clusters) in such a way that the objects within the cluster are highly similar compared to the objects outside of the cluster. Spatial clustering forms an important part of spatial data mining (Han et al., 2001; Miller et al., 2009). A wealth of spatial clustering tools are currently available with immense application potential.
In earth observation studies, spatial cluster techniques are often applied to identify zones with similar land covers by using earth observation data as input. An example of such a technique is the K-means classifier (Han et al., 2001; Miller et al., 2009). This unsupervised classification technique makes several clusters (e.g. land use classes) of which each pixel is assigned to the cluster with the nearest mean (Han et al., 2001). The amount of clusters can be freely defined by the user just as the input metrics to perform the classification. A drawback of the K-means classifier is the need to specify the amount of output clusters. Density Based Spatial Clustering (DBSC) overcomes this issue since it automatically defines the optimal amount of clusters (Miller et al., 2009). In this type of clustering technique, dense regions of objects (proximate objects) are clustered and separated from regions with low density (noise) (Han et al., 2001; Liu et al., 2012). Finally, another frequently applied spatial clustering technique is the hierarchical agglomerative clustering. This technique makes use of a dendrogram to decompose the data into clusters. The agglomerative approach is a bottom-up approach in which all objects are first grouped in a distinct cluster and while moving upward in the tree, pairs of clusters are merged based on some metrics (e.g. spatial proximity) (Han et al., 2001).
Spatial cluster techniques have many advantages when dealing with big datasets which is often the case when working with earth observation data. Its simplicity to use and the fast increase of cloud computing power makes from it powerful techniques to extract spatial patterns out of the data. It allows to translate raw earth observation data into a more user-friendly data product by showing the spatial patterns of the data.