The evaluation of parametric and non-parametric models for total forest biomass estimation using UAS-LiDAR

Forest biomass estimation has drawn substantial attention due to its major impact on climate change and sustainable forest management. In this study, we assessed the performance of distributional and calibrated intensity metrics derived from the Unmanned Aerial System-Light Detection and Ranging (UAS-LiDAR) data to estimate total forest biomass individually and in combination over a ginkgo (Ginkgo biloba L.) planted forest in southeast China. First, the importance of these metrics were investigated and the optimal UAS-LiDAR metrics were selected by the "all-subsets" models. Then, the parametric (multivariate linear model (MLR)) and non-parametric (random forest (RF)) models were investigated for total forest biomass estimation. The results showed that, in general, the combo models (based on distributional and intensity metrics) (CV-R-2=0.92-0.94, rRMSE=7.13%-7.62%) performed better than the separated models (CV-R-2=0.90-0.92, rRMSE=8.05%-9.53%). Secondly, the estimation accuracy obtained from RF (CV-R-2 =0.92-0.94, rRMSE=7.13%-8.05%) was relatively higher than the MLR (CV-R-2=0.88-0.90, rRMSE=7.62%-9.53%). This study demonstrated that compared with the MLR model, RF has stronger potential to enhance the performance of biomass estimation by using UAS-LiDAR-derived metrics, and the implementation of calibrated intensity metrics has shown marked contributions to total forest biomass estimation in planted forest.