Using remote sensing to model leaf area index as an indicator of ecosystem recovery across heterogeneous landscapes in Yukon, Canada

Leaf area index (LAI) is an important indicator of productivity within reclamation programs. It is traditionally quantified using ground-based measurements at sample plots, an approach that is timeconsuming, can provide subjective information, and is spatially constrained. Alternative and complimentary cost-efficient methods are therefore of value. Remotely piloted aircraft systems (RPAS) are a pragmatic alternative that can utilize multiple sensors to collect data over reclamation sites and derive spectral or structural metrics. These metrics have been widely used in physical and regression models to estimate LAI, however, most studies focus on mature forests or agricultural settings, and rarely on more complex heterogeneous study sites which present considerably greater challenges for modelling LAI. Furthermore, the accuracy of physical and regression models are seldom compared. This study evaluates and compares physical (Beer-Lambert), parametric (linear and exponential regression), and non-parametric random forest models to estimate LAI over six heterogeneous study sites using data acquired from three different RPAS sensors: (1) a multispectral optical sensor used to derive spectral indices, (2) an rgb optical sensor used to generate a 3D point cloud using structure from motion (SfM), and (3) a lidar sensor used to generate a 3D point cloud. The results indicate that physical models can estimate LAI with greater accuracy (RMSE 0.35 – 0.38) than parametric regression models (RMSE 0.53 – 0.58) and non-parametric random forest models (RMSE 0.43 – 0.49). While physical models are slightly more effective, all models demonstrate sufficient spatiotemporal transferability to be used for monitoring productivity at reclamation sites.Non UBCUnreviewedOthe