Modelos de autorregresión espacial para la evaluación de la susceptibilidad por movimientos en masa

Spatial autoregression models for the evaluation of landslide susceptibility

Edier Aristizábal1,*

1 Departamento de Geociencias y Medio Ambiente, Facultad de Ciencias, Universidad Nacional de Colombia. Calle 59A # 63-20, Medellín, Antioquia, Colombia.

* Autor para correspondencia: (E. Aristizabal) This email address is being protected from spambots. You need JavaScript enabled to view it. 

How to cite this article:

Aristizábal, E. (2026). Modelos de autorregresión espacial para la evaluación de la susceptibilidad por movimientos en masa. Boletín de la Sociedad Geológica Mexicana, 78(1), A031025. https://doi.org/10.18268/BSGM2026v78n1A031025 

Manuscript received: July 31, 2025. Corrected manuscript received: September 17, 2025. Manuscript accepted: September 29, 2025.

ABSTRACT

Landslides are critical geomorphological processes that substantially reshape the landscape through the downslope movement of soil and rock, often triggered by factors such as rainfall, earthquakes, or anthropogenic interventions. These processes pose significant hazards to infrastructure, human safety, and socioeconomic stability. Conventional statistical models frequently fail to adequately capture the spatial nature of landslide susceptibility, often leading to biased or misleading outcomes due to the assumption of independence among observations, which ignores inherent spatial heterogeneity and spatial dependence. This study addresses these limitations by employing spatial autoregressive models, which explicitly account for spatial dependence through the integration of neighborhood matrices. The dataset comprises catchments from the Colombian Andes, incorporating morphometric predictors at both local and regional scales, including slope, hypsometry, basin area, and annual precipitation. We built a neighborhood matrix based on distance criteria, recognizing that geoenvironmental factors influencing landslides often extend beyond direct boundaries, requiring a broader understanding of spatial interactions. Our findings demonstrate that incorporating spatial dependence significantly enhances both the predictive accuracy and interpretative power of the models when compared to conventional approaches. Analysis using Moran’s Index revealed that basin slope and precipitation exhibit strong spatial dependence, forming clusters of similar values, which underscores the necessity of accounting for spatial effects. The Spatial Durbin Error Model (SDEM) outperformed alternative models by providing higher adjusted R2 values and optimizing the balance between model complexity and fit, as measured by the Akaike Information Criterion (AIC). By explicitly integrating spatial neighborhoods in landslide susceptibility assessment, this study provides a robust and reliable assessment of landslide susceptibility, which is crucial for understanding and managing these hazards in regions like the Colombian Andes.

Keywords: landslide, susceptibility, spatial dependence, Colombia.