Abstract: Aiming at three key indicators of aboveground biomass (AGB), vegetation optical depth (VOD) and canopy height model (CHM) in biomass estimation, a regression composite model BorutaShap-CNN-LSTM-GRU (BCLG) based on Cyclone Global Navigation Satellite System (CYGNSS) data was proposed. The model combines the advantages of convolutional neural network (CNN), long short-term memory (LSTM) and gated recurrent unit (GRU), and optimizes the performance of the model through BorutaShap feature selection technology to improve the inversion accuracy. By comparing the inversion results of the validation set of AGB and VOD at 6 months and CHM at 12 months, compared with the full-parameter CLG model without BorutaShap, the BCLG model increased the \mathitR^2 of AGB, VOD and CHM from 0.78, 0.96 and −0.08 to 0.83, 0.97 and 0.21, respectively, and the RMSE decreased from 25.08 t/ha, 0.05 m and 8.34 m to 22.29 t/ha, 0.04 m and 7.15 m. Compared with the single CNN, LSTM and GRU models after BorutaShap, the BCLG composite model showed significant advantages in the inversion accuracy of AGB, VOD and CHM, which proves that the proposed BCLG model has good performance in biomass index inversion based on CYGNSS, and provides a new technical means for remote sensing biomass monitoring and evaluation.