[Skip to Content]

Menu
  • Home
  • My Account
Menu
  • Home
  • East Lakes Regional Division Meeting 2026 Session Gallery
  • Landscape Pedagogy
  • GeoToMATo: Spatial Clustering for Geographic Data Analysis

Please click on the poster thumbnail above to enlarge the image. 

Audio Presentation
Link will open in a new tab.

GeoToMATo: Spatial Clustering for Geographic Data Analysis

Abstract:

Clustering and regionalization are foundational tools in spatial analysis, yet the most widely used methods possess notable limitations. Centroid-based algorithms such as K-Means assume spherical, similarly sized clusters, while density based methods such as DBSCAN, HDBSCAN, and OPTICS offer limited control over how many points remain unclassified and their runtimes scale superlinearly with dataset size. Regionalization algorithms such as Ward and SKATER are designed to produce maximally homogenous groups, and while often optimized for this task under specific objectives and constraints, their emphasis on class homogeneity makes them prone to splitting regions along boundaries of natural within-class variation. This presentation advocates for the use of the Topological Mode Analysis Tool (ToMATo; Chazal et al., 2011), a near-linear-time clustering algorithm from topological data analysis, for spatial point clustering in geography. Unlike HDBSCAN, which has similar theoretical foundations, ToMATo assigns every point to a cluster and then optionally filters noise using a user-defined assignment probability threshold. I then introduce GeoToMATo, an adaptation of ToMATo for geographic regionalization. Rather than optimizing for class homogeneity, GeoToMATo identifies dense modes of attribute similarity in a user-defined spatial graph and determines the extents of their basins of attraction to define cluster boundaries. This yields spatially contiguous regions of attribute similarity that are robust to natural within-class variation, making the method well-suited to tasks such as hyperspectral image segmentation. On the Salinas Valley, CA hyperspectral scene, GeoToMATo is shown to outperform Ward regionalization and SLIC hyperspectral image segmentation in adjusted Rand index against ground-truth labels.

Keywords: Spatial analysis, clustering, regionalization


Authors:

Kenneth Howe, Sinclair Community College; Submitting Author / Primary Presenter

, ;

, ; 

, ; 

, {SubmissionForm.presentersAuthors/affiliation#4};

, ;

, ; 

, ; 

, ;

, ;

 

GeoToMATo: Spatial Clustering for Geographic Data Analysis

Category

Paper Abstract

Description

This abstract is part of the session: Landscape Pedagogy

Custom CSS

double-click to edit, do not edit in source