Tourist Mobility Forecasting with Region-Based Flows and Hierarchical Spatial Tessellation

This paper introduces a novel approach to tourist mobility prediction based on Graph Neural Networks (GNNs) trained with general human mobility (GMD) data, evaluating its performance through multiple spatial scales. By using the Region of Murcia (Spain) as a case study, we demonstrate that enriching GNNs with GMD flows significantly improves prediction accuracy compared to univariate time-series models and CNNLSTM baselines. Specifically, the results reveal that incorporating the total number of visitors and overnight tourists in our model significantly improves the mobility prediction accuracy. In contrast, the benefits for including excursionist flows are limited to short-term forecasts only. Moreover, the improvement in tourist flow prediction is more evident at coarser spatial scales compared to finer municipal areas, suggesting that the utility of GMD is dependent on the spatial granularity of the target region. These findings can be leveraged to inform policy-making and large-scale tourism management.