Cost distance and potential accessibility as alternative spatial approximators of human influence in LUCC modelling

Most land use/cover change (LUCC) models use socioeconomic predictors which spatially approximate human influence on the landscape. According to our meta-analysis, the simplest spatial approximations – the minimal Euclidean distance and indicators based on administrative areas – are used most often. We argue that these approximators are theoretically unjustified for a significant portion of LUCC research designs, mainly because the Euclidean distance ignore other factors influencing transportation costs and the administrative areas often have insufficient spatial resolution or unrepresentative shape. We introduce improvements which overcome these theoretical challenges: improved slope-based cost distance and exposure to human influence as a variant of potential accessibility. The exposure is modified to be measured in activity-per-area units (e.g., population/km2), which are easier to interpret and compare with other socioeconomic indicators (e.g., population density) than standard potential accessibility units. Logistic regression was used to quantify the predictive power of these alternative approximators within different LUCCs in Slovakia, across three time periods. Their performance (AUC) was compared with standard predictors in polynomial models, and their contribution when combined with biophysical predictors was evaluated using hierarchical partitioning. In line with the study’s theoretical assumptions, cost distance and exposure significantly surpassed the standard approximators of human influence in most comparisons. They achieved a relatively good performance even for some LUCCs (pastures, permanent cultures, forests, agricultural extensification and intensification, afforestation) where standard socioeconomic indicators were weak. These indicators thus showed potential to improve the precision of LUCC models, particularly in mountainous regions.