Moving from expert opinion to empirical evidence: Data-driven parametrization of urban connectivity models using movement-proxy data

Abstract Context The landscape connectivity of cities is increasingly recognized as crucial for biodiversity conservation and ecosystem services. Yet, modelling ecological connectivity in cities remains challenging because landscape resistance is often based on expert judgment rather than empirical evidence, leading to varying modelling results and limited use for planning. Objectives We developed and tested a data-driven framework for empirically parametrizing resistance and movement-distance parameters in functional connectivity models from movement-proxy data—information on the presence/absence of animal movement from direct observation or camera traps. At each step, we ensured that the connectivity model reflected the behavioural and spatial properties of the observations. We applied the framework for the common blackbird ( Turdus merula ) in Munich, Germany. Methods We used observations of flying blackbirds as movement-proxy data in a logistic regression framework, testing alternative combinations of resistance and movement distances. Model selection identified the parameter sets best supported by the data. The resulting parameters were validated using repeated out-of-sample validation and compared against an expert-based connectivity model. Results Connectivity derived from empirically estimated parameters increased the probability of observing flying blackbirds. Across repeated validations, the empirical model achieved a mean AUC of 0.76 and R 2 of 0.17. It performed moderately better than the expert-based model. Depending on their height, buildings exhibited varying resistance to flying blackbirds. Results indicate that expert assessments may oversimplify urban barriers. Conclusions The approach provides a transparent, reproducible framework for using movement-proxy data to derive maps of landscape resistance. It offers a step toward more data-driven urban connectivity modelling.