Abstract

Above-ground biomass change accumulated during four growth seasons in a hemi-boreal forest was predicted using airborne L- and P-band synthetic aperture radar (SAR) backscatter. The radar data were collected in the BioSAR 2007 and BioSAR 2010 campaigns over the Remningstorp test site in southern Sweden. Regression models for biomass change were developed from biomass maps created using airborne LiDAR data and field measurements. To facilitate training and prediction on image pairs acquired at different dates, a backscatter offset correction method for L-band data was developed and evaluated. The correction, based on the HV/VV backscatter ratio, facilitated predictions across image pairs almost identical to those obtained using data from the same image pair for both training and prediction. For P-band, previous positive results using an offset correction based on the HH/VV ratio were validated. The best L-band model achieved a root mean square error (RMSE) of 21 t/ha, and the best P-band model achieved an RMSE of 19 t/ha. Those accuracies are similar to that of the LiDAR-based biomass change of 18 t/ha. The limitation of using LiDAR-based data for training was considered. The findings demonstrate potential for improved biomass change predictions from L-band backscatter despite varying environmental conditions and calibration uncertainties.

抽象的

使用机载L波段和P波段合成孔径雷达（SAR）反向散射预测了半北方森林在四个生长季节中累积的地上生物量变化。雷达数据是在瑞典南部Remningstorp测试地点的BioSAR 2007和BioSAR 2010活动中收集的。生物量变化的回归模型是根据使用机载LiDAR数据和现场测量创建的生物量图开发的。为了便于对在不同日期获取的图像对进行训练和预测，开发并评估了L波段数据的反向散射偏移校正方法。基于HV / VV后向散射比的校正，有助于跨图像对的预测，几乎与使用相同图像对的数据进行训练和预测所获得的预测相同。对于P波段，使用基于HH / VV比率的偏移校正对先前的阳性结果进行了验证。最佳的L波段模型的均方根误差（RMSE）为21吨/公顷，最佳的P波段模型的均方根误差为19吨/公顷。这些精度与基于LiDAR的18吨/公顷生物质变化的精度相似。考虑了使用基于LiDAR的数据进行训练的局限性。这些发现表明，尽管环境条件和校准不确定性各不相同，但通过L波段反向散射可以改善生物量变化的预测潜力。考虑了使用基于LiDAR的数据进行训练的局限性。这些发现表明，尽管环境条件和校准不确定性各不相同，但通过L波段反向散射可以改善生物量变化的预测潜力。考虑了使用基于LiDAR的数据进行训练的局限性。这些发现表明，尽管环境条件和校准不确定性各不相同，但通过L波段反向散射可以改善生物量变化的预测潜力。