Analysing the effect of different aggregation approaches on remotely sensed data

The effect of spatial aggregation varies depending on the aggregation logic. This study examined and compared the effect of both categorical and numerical aggregation. Categorical aggregation focused on the majority rule-based (MRB), random rule-based (RRB), and point-centred distance-weighted moving window (PDW). Both RRB and PDW have a stochastic component. Numerical aggregation focused on mean aggregation and central pixel resampling (CPR). The change in class proportions and landscape metrics with respect to a fine-resolution base image were assessed. RRB, PDW, and CPR preserved class proportion with decreasing spatial resolution. MRB increased the proportion of the dominant class and decreased all other class proportions, whereas mean aggregation increased the proportion of non-dominant class. All approaches led to a less clumped pattern for each class, except MRB, which made the dominant class more clumped. For all classes, RRB, PDW, and CPR led to a lower distortion in shape complexity than MRB and mean aggregation. RRB responded similarly for all realizations, but variability in PDW could be minimized by choosing a specific parameter value. The study showed that RRB, PDW, and CPR can be used in, for example, studies on ecological resource management where consistency of the class proportions at coarser resolutions is required, and that PDW is the best option. MRB can be used in regional-level as well as national-level agriculture or forest planning, where the delineation of the dominant class is required.