A two-layer multivariate decomposition framework for spatiotemporal tourism demand forecasting based on Internet search and city synergy factors

Accurate forecasting of tourism demand plays a critical role in optimizing the allocation and management of tourism resources. However, identifying tourist intentions and trends remains challenging due to the complexity and multidimensional nature of influencing factors. To address these challenges, this study integrates Internet search indices with tourist arrival data to trace the spatiotemporal dynamics of tourism demand in a target city and its surrounding regions. A novel two-layer multivariate decomposition ensemble framework is proposed to reduce data complexity and achieve comprehensive analysis and feature extraction. The framework employs a two-step process: multivariate empirical mode decomposition is used to extract intrinsic mode functions (IMFs) that represent distinct features, followed by multivariate variational mode decomposition applied to high-complexity IMFs to uncover deeper interrelationships. These IMFs are then predicted using a multivariate gated recurrent unit, and the final forecasting result is obtained by aggregating the predictions. Empirical evaluations on two case studies, Nanjing and Haikou, demonstrate that the proposed framework achieves superior predictive accuracy compared to baseline models. The results highlight the effectiveness of integrating spatiotemporal data with advanced decomposition methods to uncover latent patterns in tourism demand. Furthermore, we provide actionable recommendations for tourism policymakers and stakeholders, offering insights into resource allocation, congestion management, and sustainable tourism development.