Abstract

Accurately estimating the shrubland aboveground biomass (AGB) of the desert steppe and understanding its dynamic changes are vitally important for studying the regional carbon cycle and sustainable use of grassland resources. Synthetic aperture radar (SAR) data are seldom applied to estimate shrubland AGB, particularly in arid and semiarid desert steppe landscapes. The main objective of this study was to model shrubland AGB in arid and semiarid desert steppe regions with HJ1B and RADARSAT-2 C-band data. The main species was Caragana microphylla, one of the most widespread shrubs in desert grasslands. Three regression models based on stepwise multiple linear regression (SMLR), partial least squares regression (PLSR) and random forest regression (RFR) were developed between the measured AGB and radar backscatter coefficient, vegetation indices and texture information. The three methods were compared, and their performance was evaluated for the estimation of shrubland AGB from HJ1B and RADARSAT-2 data. The results show that SAR can produce higher accuracy with R² = 0.34 – 0.42 than PLSR and RFR in arid and semiarid desert steppe regions. The RF results were slightly worse than those of SMLR and PLSR for sparse vegetation. The use of C-band SAR data to monitor dynamic changes in shrubland AGB in northern China will be of great value.