Integrating airborne LiDAR and space-borne radar via multivariate kriging to estimate above-ground biomass

Abstract

Understanding and investigating synergies between LiDAR (light detection and ranging) and SAR (synthetic aperture radar) provide new and innovative opportunities to characterize above-ground biomass. We demonstrate a spatial modeling framework that integrates above-ground biomass transects, derived from plot-based field data and small-footprint discrete return LiDAR, with complete wall-to-wall spaceborne L-band and C-band SAR to predict biomass over a larger area. Transect intervals of 2000m, 1000m, and 500m were tested. Co-kriging, regression kriging, and regression co-kriging were used to extend the LiDAR-derived biomass transects. LiDAR-derived above-ground biomass and L-band backscatter (HV polarization) were moderately correlated, with a maximum semivariance distance between the LiDAR-derived biomass and SAR data of 374m. Regression kriging at a sample interval of 500m showed the smallest root mean squared error (RMSE) and mean absolute error (MAE) at 203.9Mgha−1 and 131.6Mgha−1, respectively. The mean error (ME) showed an average bias of −14.0Mgha−1. Predictions using regression co-kriging at a sample interval of 2000m resulted in the highest RMSE and MAE values at 238.2Mgha−1 and 164.6Mgha−1, respectively. ME also was highest, averaging −37.4Mgha−1. Regardless of the spatial modeling technique employed, lower errors in predicted above-ground biomass were associated with smaller transect intervals. Moderate correlations between the LiDAR-derived above-ground biomass and the radar data impacted the predictive accuracy of the spatial models; however, overall variation in above-ground biomass in the study area was well represented. This study demonstrated that a sampling framework integrating LiDAR data with space-borne radar data using a spatial modeling approach can provide spatially-explicit above-ground biomass estimates for large areas. Such a sampling framework can be used in combination with ground plot and land cover data to assess carbon stocks under conditions where more common optical remote sensing approaches are difficult to implement.