Forest is one of the most widely distributed terrestrial ecosystems on earth. Global-scale biomass estimation has become a research hotspot. It is important to accurately estimate the spatial distribution of forest above-ground biomass (AGB) because its carbon budget forms part of the global carbon cycle and ecosystem productivity. Remotely sensed data have been widely used to quantitatively obtain the biophysical characteristics of vegetation canopy structure. The use of optical and microwave remote sensing in combination with field measurements can provide an effective method to improve the estimation of forest biomass over large regions. In order to improve the accuracy of estimating forest above-ground biomass from remotely sensed data, the methods for obtaining AGB data using a physically-based canopy reflectance model inversion approach and two other empirical statistical regression methods were introduced in this paper. A geometric-optical canopy reflectance model was run in multiple-forward mode (MFM) using multispectral HJ1B imagery to derive forest biomass at the Helan Mountain Nature Reserve region in the northwest of China. Structural parameters of the forest inventory were carried out in 50 separate 30 m by 30 m randomly distributed plots, and the data was used for either model development or validation. The two other empirical-statistical models were also established to estimate the biomass in the area. A multiple stepwise regression model was developed to estimate the forest above-ground biomass by integrating the field measurements of 30 sample plots with ALOS/PALSAR Synthetic Aperture Radar (SAR) backscatter remotely sensed data. The pre-processing of the HJ1B scenes included radiometric calibration, atmospheric correction, and georeferencing. Radiometric data were converted from radiance to reflectance. Additionally, spectral mixture analysis (SMA) was applied to decompose a mixture of spectral components of HJ1B into vegetation, soil, and shade fractions. The vegetation fraction image was fused with PALSAR data using the discrete wavelet transform (DWT) method. As a comparison, a regression model was also created by integrating field measurements with the fused image. Error levels for the three models and the field-measured data were analyzed. MFM predictions of AGB from HJ1B imagery were compared with the results from the SMA and PALSAR multiple stepwise regression models. Simultaneously, the estimation biomass using the three methods was evaluated for 20 field validation sites. The result shows that a good fit can be found between the AGB estimated by geometric-optical canopy reflectance model and the field-measured biomass with a R² (coefficient of determination) and RMSE (root mean-square error) of 0.61 and 8.33 t/hm², respectively. MFM provided the lowest error for all validation plots and its estimated accuracy is a little better than that of the SMA model (R²=0.60, RMSE=9.417 t/hm²). PALSAR multiple stepwise regression model has the worst estimation accuracy (R²=0.39, RMSE=14.89 t/hm²) and had a higher error. Consequently, it can conclude that geometric-optical canopy reflectance model and spectral mixture analysis (SMA) approach were considerably more suitable for estimating the forest biomass in mountainous terrain. Moreover, it demonstrates a good potential for monitoring the indicators of forest ecosystem by combined with the optical and polarimetric SAR remote sensing synergistic research.