Precise characterization of soil moisture dynamics in the vadose zone is fundamental for numerous science and engineering applications, such as soil mechanics, agricultural water management, and hydrological cycle simulation (Vereecken et al., 2015(Vereecken et al., , 2022. Accurate model establishment for soil moisture flow is the key to characterizing soil moisture dynamics.With contact and non-contact soil moisture sensor technology improved (e.g., W. Dorigo et al., 2021;W. A. Dorigo et al., 2011), it is increasingly convenient to access soil moisture observational data. Meanwhile, developments in data-driven algorithms provide opportunities to leverage these rich data, gaining us new insights into modeling, simulating, and understanding soil moisture dynamics (Ghanbarian & Pachepsky, 2022). Most current practices in modeling soil moisture flow have leveraged data-driven approaches. Their core idea is basically to use algorithms to establish mapping to approximate the latent soil moisture dynamics involving traditional machine learning (e.g.,