[Paper Review] The spatial scan statistic: A new method for spatial aggregation of categorical raster maps

John W. Coulston, Nicola Zaccarelli+3 more|arXiv (Cornell University)|Aug 1, 2014

Data-Driven Disease Surveillance27 references3 citations

TL;DR

This paper introduces the spatial scan statistic as a novel method for spatial aggregation of categorical raster maps, preserving rare classes and spatial heterogeneity better than majority rule or nearest neighbor techniques. By identifying statistically significant spatial clusters, the method maintains ecological accuracy in large-scale assessments where information loss from traditional aggregation is problematic.

ABSTRACT

Multiple-scale and broad-scale assessments often require rescaling the original data to a consistent grain size for analysis. Rescaling categorical raster data by spatial aggregation is common in large area ecological assessments. However, distortion and loss of information are associated with aggregation. Using a majority rule generally results in dominant classes becoming more pronounced and rare classes becoming less pronounced. Using nearest neighbor techniques generally maintains the global proportion of each category in the original map but can lead to disaggregation. In this paper we implement the spatial scan statistic for spatial aggregation of categorical raster maps and describe the behavior of the technique at the local level (aggregation unit) and global level (map). We also contrast the spatial scan statistic technique with the majority rule and nearest neighbor approaches. In general, the scan statistic technique behaved inverse the majority rule approach in that rare classes rather than abundant classes were preserved. We suggest the scan statistic techniques should be used for spatial aggregation of categorical maps when preserving heterogeneity and information from rare classes are important goals of the study or assessment.

Motivation & Objective

• To address information distortion and loss in spatial aggregation of categorical raster maps used in ecological assessments.
• To develop a method that preserves rare classes and spatial heterogeneity during rescaling to consistent grain sizes.
• To provide an alternative to majority rule and nearest neighbor aggregation that reduces bias toward dominant classes.
• To evaluate the performance of the spatial scan statistic in comparison to conventional aggregation techniques.
• To support more accurate, information-rich large-scale ecological monitoring and analysis.

Proposed method

• The spatial scan statistic is applied to identify statistically significant spatial clusters in categorical raster data.
• The method uses a scanning window that moves across the map to detect local clusters of similar categories.
• For each window, a likelihood ratio test evaluates whether the observed category distribution differs significantly from the expected distribution under the null hypothesis of randomness.
• The most likely cluster is selected based on the maximum likelihood ratio, and the process is repeated across all possible window sizes and positions.
• Aggregation is then performed by assigning each cell to the most likely cluster, preserving spatial patterns and rare classes.
• The approach is contrasted with majority rule (which favors dominant classes) and nearest neighbor (which may cause spatial disaggregation).

Experimental results

Research questions

• RQ1 How does the spatial scan statistic compare to majority rule and nearest neighbor methods in preserving class proportions during spatial aggregation?
• RQ2 To what extent does the spatial scan statistic maintain spatial heterogeneity and rare class representation in categorical raster maps?
• RQ3 What are the local and global effects of using the spatial scan statistic on map structure and classification accuracy?
• RQ4 In what scenarios is the spatial scan statistic more appropriate than traditional aggregation techniques for ecological assessments?
• RQ5 How does the spatial scan statistic handle edge effects and irregular spatial patterns in categorical data?

Key findings

• The spatial scan statistic preserved rare classes better than the majority rule, which tends to amplify dominant classes.
• Unlike nearest neighbor aggregation, the spatial scan statistic maintained spatial coherence and reduced artificial disaggregation of categories.
• The method demonstrated superior performance in retaining ecological information and spatial heterogeneity at both local and global map levels.
• The spatial scan statistic produced more accurate and representative maps when the goal was to preserve information from low-frequency categories.
• The technique was particularly effective in identifying and preserving meaningful spatial clusters, even in complex or fragmented landscapes.
• The study concludes that the spatial scan statistic is a robust alternative to conventional aggregation methods when rare class preservation is a priority.