Abstract: Objectives: The inversion of forest height parameters based on the RVoG (Random Volume Over Ground) model and single baseline PolInSAR (Polarimetric Interferometric Synthetic Aperture Radar) observation data faces significant issues related to a lack of observational information. By incorporating multi-baseline observation data, it is possible to supplement the observational information of model parameters and enhance the accuracy of forest height parameter inversion. However, due to the differences in multi-baseline observation environments, the inversion model can become over-parameterized, leading to ill-posed problems in forest height parameter inversion, which severely restricts the inversion accuracy. Therefore, addressing the illposed problem of multi-baseline model parameter inversion is crucial for improving the accuracy and stability of forest height parameter inversion. Methods: A multi-baseline inversion model is constructed based on the theory of least squares estimation to achieve the fusion of multi-baseline observational information for forest height parameter inversion. Subsequently, a regularization method is employed to estimate the model parameters, thereby mitigating the effects of ill-posed problem and enhancing the accuracy of forest height parameter estimation. On this basis, a regularization parameter is selected based on the variance and bias of the forest height parameter estimates as they vary with regularization parameters, effectively improving the stability of forest height parameter estimation. Results: Experiments on forest height inversion using multi-baseline PolInSAR data indicate that the conventional forest height inversion has a root mean square error (RMSE) of 9.47 meters and a coefficient of determination (R²) of 0.72. In comparison, the RMSE for forest height inversion using the multi-baseline regularization method is 7.36 meters, with an R² of 0.80. Thus, the new method demonstrates a 22% reduction in RMSE and a relative improvement in the coefficient of determination, effectively enhancing the accuracy and stability of forest height inversion using PolInSAR. Conclusions: The theory of least squares estimation can effectively integrate multi-baseline observational information for forest height parameter inversion;however, the influence of ill-posed problem is significant. By combining this with regularization method, the impact of ill-posedness can be reduced, thereby improving the accuracy and stability of forest height parameter estimation. This constitutes an effective method for PolInSAR multibaseline integrated forest height inversion.