<p>Yaan, situated at the eastern margin of Tibetan Plateau (TP), represents a primary center of heavy precipitation often referred to as China’s "Rainy City". Despite its climatic significance, the synoptic-scale drivers and predictability of regional rainfall events (RRE) in this topographically complex region remain subjects of intense investigation. Focusing on the heavy rainfall events of the past four decades, this study investigates the key circulation factors influencing rainfall in Yaan and develops a physics-informed machine-learning-based model to examine the importance of these factors. The evolutions of heavy RRE around Yaan is intrinsically linked to a migratory precipitation signal originating to the north (precursor) and progressing southward (post) along the eastern edge of the TP. This process is governed by a distinct three-dimensional circulation anomaly, characterized by a warm (cold) anomaly center at 300 (100) hPa and the confront of warm southerly wind and cold northerly wind anomalies in the lower troposphere. As a consequence of the southeastward movement of the three-dimensional circulation pattern, the rainfall moves southward along eastern edge of the TP, reaching Yaan after about 36&#xa0;h. Leveraging these robust physical precursors, a Light Gradient Boosting Machine (LightGBM) model are implemented to predict RRE amounts. While the model accurately captures the magnitude of RRE amount in Yaan, it exhibits a characteristic limitation in replicating the intensity of extreme events. These findings advance the understanding of rainfall-related teleconnections between geographically distant regions and provide helpful guidance for the identification of early signals for rainfall disasters. Furthermore, this study demonstrates a significant methodological pathway for integrating physical meteorological insights into the development of machine learning frameworks to enhance to enhance the precipitation prediction.</p>