Remote sensing is an important method used to estimate forest canopy closure in large scale. The three kinds of remote sensing algorithms for canopy closure retrieval are statistical, physical, and mixed models. Although statistical models are commonly used, they lack physical explanation and are limited in local areas. Physical models have clear understanding on mechanism, which can be used in large areas. However, due to higher complexity, physical models are less applied. The Stochastic Radiative Transfer (SRT) model is applicable in simulating forests with horizontally distributed heterogeneity, which may represent different canopy closures. Exploring the inversion method using the SRT model could improve the efficiency and precision of canopy closure inversion. On the basis of the SRT model, an inversion method has been proposed on canopy closure retrieval of Yunnan pine forests. The fundamental step is to determine the quantitative relationship between the canopy closure and the probability of finding foliage elements in SRT model. To match the Yunnan pine crown shape, an equivalent model was used to correct the cylinder shape assumption. Then, a look-up-table was constructed to inverse the canopy closure to obtain the reflectance from GF-1 and Landsat 8 satellite images. The probability of finding foliage elements and leaf area index were determined in the case of a minimum difference between simulated reflectance and satellite observations, and to calculate the canopy closure on the basis of the stochastic Beer-Lambert-Bouguer law. Thirty plots of field data were used to assess the inversion accuracy. A statistical inversion method based on NDVI was conducted for comparison. Results showed that the inversion can accurately map the canopy closure of Yunnan pine forests in the study area (R2=0.8345, RMSE=0.0688). Reflectance of the bands used for retrieval performed sensitively to canopy closure. The use of composite image from GF-1 and Landsat 8 is feasible. The equivalent shape correction model is reasonable, which reduced RMSE by 0.0466, and the algorithm is flexible in different crown cases. This study can support forward models and inversion methods for large-scale forest canopy closure retrieval. The research could be extended to any tree species by changing the model parameter input, and any crown type by crown shape equivalent correction. © 2020, Science Press. All right reserved.