Comparing Intercell Distance and Cell Wall Midpoint Criteria for Discrete Global Grid Systems

MJ Gregory, AJ Kimerling, D White and K Sahr

Email: matt.gregory@orst.edu


Many diverse applications study processes and patterns at a global scale. To aid in this research, discrete global grid systems (DGGSs) enable environmental modeling, monitoring and sampling across the earth at a variety of spatial scales. A DGGS can be evaluated on a set of topological and geometric criteria, two of which, intercell distance and the "cell wall midpoint or Heikes-Randall criterion", form the basis of this study. These two properties have been noted to be important for dynamic modeling applications.

This study focuses on results obtained from measurements of six different global partitioning methods. Each method was further subdivided into different design choices, which included frequency of cell subdivision (2- or 3-frequency edge partition), predominant tessellating shape (triangle, hexagon, diamond or quadrilateral), and base modeling solid (octahedron, icosahedron, or sphere). Intercell distance and cell wall midpoint measures were statistically normalized to be comparable among the methods studied. The results were further standardized to a common mean intercell distance (89.02 km) to determine performance rankings for the methods. Finally, the distortion of each method was presented graphically to understand the underlying spatial pattern.

For intercell distances, the Fuller-Gray method had the best performance, while two quadrilateral grids (Tobler-Chen and Equal Angle) performed substantially worse. For the cell wall midpoint criterion, the Equal Angle grid had the lowest overall distortion with the Snyder and Fuller-Gray methods also performing relatively well. The Tobler-Chen grid generally had the poorest performance for this property, especially at the higher recursion levels studied. All methods based on the icosahedron outperformed those based on the octahedron for both criteria studied. Aggregation of triangles into hexagons and diamonds seemed to have little impact on intercell distance measurements, although dual hexagon aggregation had markedly different statistics and spatial patterns for the cell wall midpoint property. Frequency of subdivision seemed to account for very little variation. Maps of spatial variation for both criteria show patterns of localized distortion which are unique to each method studied