Comparison of Two Common Empirical Methods to Model Land-Use Choices in a Multi-Agent System Simulation of Landscape Transition: Implication for a Hybrid Approach

Land-use choice routines embedded in a human-environment system (HES) model must meet more requirements than those in models typically presented in purely economic or psychological studies. This study compares the strengths and shortcomings of two common empirical methods - multi-nominal logistic (MNL) regression and classification tree (CT) analysis - for specifying land- use choices in a multi-agent system simulation framework (Land Use Dynamics Simulator - LUDAS). First, we described design concepts of land-use decision- making mechanism in the LUDAS framework in which household's land-use choice is a component. Next, we compared two common methods for modeling the land- use choice with respect to pre-established criteria: a MNL model was specified to represent assumed rational behavior of human agents, while the CT model used a data-fit hierarchical rule set to represent heuristic process of reflex behavior. The study was conducted based on an intensive household-farm survey in a Central Vietnam's mountainous catchment. Based on the comparative analysis, we recommended explicit strategies for developing structurally realistic models that utilizes the complementarities of the both techniques to better represent bounded rational, yet adaptive, land-use choices in a HES model in the face of uncertainty.