Novel pseudo-logistic spatial regression for the assessment of local/zonal susceptibility to landslides – case study in Central Vietnam (Bình Định Province)

Nuno de Sousa Neves, Matias de Sousa Neves

Abstract. The exploratory geographic modeling process aims to formalize spatial relationships through the combination of metrics or variables that explain spatial phenomena and integrate spatial dimensions. This study adopts a similari-ty-based induction process, evaluating multidimensional distances using derived variables or exploratory geographic modeling metrics.
The methodology integrates exploratory geographic modelling, multidimen-sional distance analysis, and probabilistic classification within a GIS map algebra environment. It incorporates logistic regression principles by embedding similari-ty-based distance structures into a Gaussian-informed probabilistic framework.
Environmental predictors derived from a 12.5 m digital elevation model—slope, Structural Hydric Erosion (EHE), Potential Structural Deposition (DEP), and Potential Structural Dryness (PSD)—are interpreted as expressions of geo-morphological and hydrological dynamics. These variables are treated as interde-pendent signatures of landscape organization shaped by gravity, water flow, and topographic structure.
To address the absence of true absence data, ISODATA-based isocluster par-titioning discretizes the multidimensional environmental space into regions of similarity. Multidimensional scaling (MDS) projects these distances into a two-dimensional Cartesian space, where proximity reflects environmental similarity and separation indicates occurrence versus non-occurrence conditions.
A Maximum Likelihood Classification framework assuming Gaussian distri-butions is adapted into a para-logistic structure using Mahalanobis distance met-rics. The resulting pseudo-logistic model relates occurrence data with structured pseudo-absence locations.