Robust estimation of pasture biomass using dual-polarisation TerrASAR-X imagery

The efficient management of pasture is a key driver of profitability for New Zealands pastoral dairy and meat industries. In this paper, we report results from a study that tests the viability of using short-wavelength imaging radar to estimate pasture biomass. We use images from the TerraSAR-X satellite, comparing the HH & HV and VV&HV dual-polarisation combinations. Several linear models have been tested for field-measured pasture biomass. The best model is a linear regression using backscatter from the HH&HV combination, resulting in a residual standard error of 317 kg/ha. This standard error is 61% less than the standard error for the VV&VH combination (511 kg/ha), which clearly suggests adopting the HH&HV model. A model for prediction of the pasture biomass rank percentile is somewhat less convincing than the model for the pasture biomass itself. Finally, a repeated measures analysis for the HH&HV pair, suggests invariance of the regression for pasture biomass over time, except for a possible outlier date 01 September 2008.