Multivariate Spatial Modelling of Tree Species Distribution Using Adaptive Sparse Group Lasso: Case Study in Barro Colorado Island Plot

Multivariate spatial point patterns have gained significant popularity as technology enables large-scale data collection. This study focuses on modeling tree species distribution in the Barro Colorado Island (BCI) forest plot, considering 9 tree species and 13 environmental covariates. To understand species co-occurrence with their environment and each other, we apply the inhomogeneous multivariate Poisson point process (IMPP) model. The Random Quadrature (RQ) method is used for efficient parameter estimation, aligning with generalized linear models while minimizing redundant computations. Given the high density of covariates, we apply the adaptive sparse group lasso (ASGL) for variable selection, focusing on the most influential environmental variables per species. The model using ASGL shows superior performance, with an AIC value of 158849.6, indicating a better fit compared to the model without ASGL. The study identifies key environmental factors such as elevation, slope, potassium content, phosphorus content, and solar radiation that significantly influence species distribution. Finally, hierarchical clustering identifies three distinct habitat groups based on their estimated environmental responses. The clustering results reveal three distinct habitat groups. These findings offer valuable insights for afforestation and reforestation efforts.