<p>Transfer learning is widely recognized for its effectiveness in leveraging external information to enhance the learning performance and predictive accuracy of target domain models. However, research on transfer learning within the context of the semiparametric varying-coefficient spatial autoregressive model is currently absent. In this study, we address this gap by introducing a transfer learning approach tailored to this model. Our method aims to improve estimation and prediction accuracy by effectively transferring knowledge from source data to the target model. We propose different algorithms for the cases where the transferable sources are known and unknown, respectively. Through extensive simulation experiments and real-world applications, we validate the efficacy of our proposed approach.</p>