Inversion of forest aboveground biomass from combined images of GF-1 and GF-3

  Objective  The objective of this study was to explore the feasibility of GF-1, GF-3, and combination of GF-1 and GF-3 for total forest aboveground biomass (AGB) and its component inversion.

  Method  In this study, the vegetation indices, texture characterizations extracted from GF-1, backscattering coefficients, texture characterizations and polarimetric decomposition features extracted from GF-3 were used independently and in combination to estimate total AGB and component AGB of a pure forest of Yunnan pine (Pinus yunnanensis), located in Xiaoshao Forest Region in Yiliang County, Yunnan Province of southwestern China. A fast iterative features selection method for k-NN method (KNN-FIFS) was applied in forest total AGB and component AGB inversion and the leave one out cross validation (LOOCV) method was used to evaluate the model and the inversion results and the results were mapped and analyzed.

  Result  The joint GF-1 and GF-3 data had the highest accuracy for inversion of forest AGB and each component AGB, and the R² of each exceeded 0.710, and the values of RMSEr were between 22% and 27%, among which the inversion accuracy of foliage was the best, with the model’s R² of 0.714, RMSE of 10.270 t/ha, and RMSEr of 24.58%; except for foliage component AGB, forest AGB and other components of AGB had better accuracy than the inversion results with GF-3 features when only the features extracted from GF-1 were used for the inversion.

  Conclusion  Combining GF-1 optical data and GF-3 fully polarized SAR data can achieve a certain degree of complementarity to improve the inversion accuracy of forest AGB and fractional AGB. In addition, the KNN-FIFS method is robust in the inversion of AGB and fractional AGB of pure Yunnan pine forests at low biomass levels, and the important parameters preferred by KNN-FIFS mostly are texture features extracted form SAR and optics data.