Modeling realized gains in Douglas-fir (Pseudotsuga menziesii) using laser scanning data from unmanned aircraft systems (UAS)

Abstract

Tree breeding programs form an integral part of sustainable forest management by providing genetically improved stock for reforestation. These programs rely on accurate phenotyping of forest trials, which become increasingly difficult to assess as trees grow larger and canopy closure occurs. Airborne laser scanning (ALS) provides three-dimensional point cloud information on forest structure which can be used to characterize phenotypes of forest trees. We analyzed 22-year-old realized gain trials of coastal Douglas-fir (Pseudotsuga menziesii [Mirb.] Franco var. menziesii) at three sites in coastal British Columbia, Canada, using dense point clouds produced from ALS acquired by unmanned aircraft system (UAS). We assessed the accuracy of ALS data against ground estimates of stand maximum height (r2 = 0.90, p < 0.001) and leaf area (r2 = 0.82, p < 0.001). We characterized phenotypes in blocks of differing levels of predicted genetic gain by generating a suite of quantitative point cloud metrics to describe four categories of stand attributes: Height, Density, Heterogeneity, and Volume. By normalizing all metrics to percent change from the means of unimproved control blocks, we analyzed point cloud metrics in terms of realized gains comparable across sites. Variable importance scores derived from conditional Random Forests indicated that descriptors of canopy height were the most important predictors of genetic gain for volume-per-ha. We selected a simple bivariate regression model using gains in mean canopy height and effective leaf area index to predict realized genetic gain for total stand volume (R2 = 0.82 – 0.94, p < 0.01, RMSE = 9.12 – 10.8%). Based on the consistent performance of this model across sites, we suggest that characterizing genetic trials in terms of increases in tree height and leaf area is a robust approach to predicting volume gains in this system. Additionally, we discuss the application of ALS as part of a phenotyping platform to inform operational decision making and forestry policy in British Columbia.