Heat Map Segmentation

Abstract

Many geospatial datasets can be represented as a heat map, such as rainfall, population density, terrain elevation, and others. These heat maps tend to form clusters of high density areas among a background of low density areas. This gem presents an automatic way to detect such clusters, and segment the heat map into areas. Experiments are conducted for two datasets which correlate to population density and show that the segmentation aligns with metropolitan areas and is stable to the choice of dataset. The segmentation described in this gem can potentially aid geospatial algorithms by supplying a smart divide-and-conquer strategy, such that the algorithm does not need to run for the entire Earth, but rather there can be a fine-grained model for each area.