multiScaleR: a generalizable approach for multiscale ecological modeling and scale of effect estimation

Analyses in landscape ecology often seek to understand how the landscape surrounding field survey locations relates to ecological responses measured at those sites. A central challenge in these studies is defining the spatial scale at which landscape variables matter. While the limitations of standard approaches to estimating this scale are well known, practical alternatives remain limited and often difficult to apply. Using simulation and the newly developed R package `mulitScaleR`, this paper describes the performance of scale optimization in relation to data type, sample size, effect size, sample independence, raster surface correlation, spatial autocorrelation, and habitat aggregation. I demonstrate how `multiScaleR` is a significant and accessible advancement for estimating scales of effect. The package builds upon existing methods that apply kernel weighting functions to landscape variables but is more general and versatile than existing methods. Functions have been optimized for computational speed and efficiency, including parallelization, use of sparse matrices, and C + + , facilitating efficient analyses of large data sets. Maximum likelihood-based regression frameworks commonly used in landscape ecology, including models from `unmarked`, `spaMM`, and `glmmTMB` can be seamlessly integrated with `multiScaleR`. The package provides a complete workflow for fitting models, conducting model selection, and spatially projecting models. Two critical insights emerge from simulations and analyses with `multiScaleR`: (1) scales of effect can be estimated with high accuracy and precision alongside regression parameters, but (2) achieving reliable estimates requires large sample sizes. `multiScaleR` is a purpose-built R package to estimate scales of effect of landscape variables in regression analyses. The accessibility and flexibility of this package make it a powerful new resource in the toolbox of spatial ecologists.