Simple scan tests for spatial clustering on a square lattice

Many data sets collected for study areas consist of data values collected on a square lattice or for a set of pixels. Remotely sensed data provide perhaps the most common example. Frequently, we wish to know whether there are subregions of the study area that exhibit spatial clustering. In this article, we suggest how an approach for finding spatial clusters may be combined with the common practice of using 3-by-3 and 5-by-5 smoothing filters or kernels to construct two simple and easy-to-implement scan-type tests. A simulation experiment shows that the power of these tests to find clusters compares favorably with an alternative test that is more complicated. The tests use simulated data, changes in a remotely sensed image for a study region in Texas, and data about wheat yields.