Causal machine learning methods for understanding land use and land cover change

Abstract

Context: Understanding the roles of different drivers in land use and land cover change (LULCC) is a critical research challenge. However, as LULCC is the result of complex, socio-ecological processes and is highly context dependent, achieving such understanding is difficult. This is particularly true for causal modelling approaches that are critical for effective policy formulation. Causal machine learning (ML) methods could help address this challenge, but are as yet poorly understood or applied by the LULCC community. Objectives: To provide an accessible introduction to the state of the art for causal ML methods, their limitations, and their potential applications understanding LULCC. Methods: We conducted two workshops where we identified the most promising ML methods for increasing understanding of LULCC dynamics. Results: We provide a brief overview of the challenges to causal modelling of LULCC, including a simple example, and the most relevant causal ML approaches for addressing these challenges, as well as their limitations. Conclusions: Causal ML methods hold considerable promise for improving causal modelling of LULCC. However, the complexity of LULCC dynamics mean that such methods must be combined with domain understanding and qualitative insights for effective policy design.