<p>Tourist flow dynamics in scenic areas exhibit strong spatial–temporal heterogeneity and are influenced by multiple environmental and socio-economic factors. Accurate prediction of visitor flows is essential for sustainable tourism management and spatial planning. To address the limitations of traditional forecasting approaches in handling nonlinear and multi-source geospatial data, this study proposes a geographic information system (GIS)-supported tourist flow prediction framework based on multi-source data fusion and deep learning. The proposed model integrates GIS data with environmental variables, holiday information, and historical visitor statistics to construct a spatial–temporal dataset for scenic area management. A hybrid deep learning architecture combining Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (Bi-LSTM), and an Attention Mechanism (AM) is developed to capture complex spatial and temporal dependencies in tourist flow patterns. Empirical experiments were conducted using data collected from three representative scenic attractions between 2021 and 2024. The results demonstrate that the proposed model significantly outperforms baseline prediction methods, achieving an RMSE of 0.872, MAPE of 11.42%, and MAE of 0.065. Furthermore, the model maintains strong predictive stability under abnormal conditions such as public health emergencies, reducing short-term prediction errors to 6.05%. By integrating GIS-based spatial information with multi-source tourism data, the framework provides an effective tool for visitor flow forecasting, spatial management, and sustainable tourism planning. The findings contribute to the development of intelligent tourism systems and support data-driven decision-making for destination management and environmental governance.</p>