CO₂ geological storage is a promising strategy for mitigating global carbon emissions, with reservoir porosity and permeability being key parameters for assessing storage potential. However, characterizing the spatial distribution of these properties at the field scale presents a major challenge due to data scarcity and heterogeneity. Conventional methods often struggle to effectively integrate discrete well-log and core measurements with continuous seismic data, while common machine learning approaches tend to neglect crucial geological controls like diagenetic processes. This limitation often leads to models with poor generalization beyond the training data.To overcome these challenges, we propose a novel multi-source, multi-scale data fusion approach for predicting porosity and permeability in tight sandstone reservoirs. Our method integrates high-precision core data with comprehensive well-log and seismic data, leveraging multiple machine learning algorithms to enhance the predictive power and generalization of the model. Taking the Upper Paleozoic Shiqianfeng Formation in the Ordos Basin as a case study, we demonstrate that this fusion approach significantly improves the accuracy of reservoir property prediction at the field scale.The results provide a more reliable foundation for evaluating storage potential and forecasting injection capacity in commercial-scale CO₂ geological storage projects. This study addresses the common issue of limited well and core data in seismic work areas, offering a robust and intelligent solution for reservoir characterization.