Background The thermal diffusivity and vertical seepage velocity of porous media are critical to characterizing the mechanisms underlying subsurface heat transfer. These two parameters are typically determined through inversion using temperature data since they are difficult to measure directly. Fiber-optic distributed temperature sensing (FO-DTS) can yield high-resolution, continuous temperature profiles, providing data for parameter identification. However, conventional numerical inversion suffers from limitations such as mesh dependency and sensitivity to initial values when used to process continuous monitoring data.

Methods This study proposed a joint geothermal parameter inversion method that integrates FO-DTS data with a physics-informed neural network (PINN). Specifically, the steady-state convection-diffusion equation was embedded as a soft regularization term into the loss function of a PINN to provide a physical constraint. In this manner, the parameter inversion problem was transformed into a physics-guided optimization process, enabling the simultaneous estimation of vertical seepage velocity and thermal diffusivity. Results and Conclusions The results indicate that the temperature field derived from PINN-based inversion exhibited a root mean square error (RMSE) of 0.20 ℃ compared to observed temperature data. The thermal diffusivity and vertical seepage velocity derived from the inversion were roughly consistent with the results from the upwind finite difference-based inversion. Furthermore, the sensitivity analysis of key hyperparameters confirms the high robustness of the PINN-based inversion framework. The results of this study provide a novel approach characterized by both physical consistency and data adaptability for identifying parameters in geothermal-seepage coupling systems.