Aboveground biomass estimates from UAV LiDAR improved via contextual learning in a Norway spruce forest

Abstract

Forest structure analysis and biomass prediction systems are key tools for advancingforest trait-based ecology and management. Surveys using Unmanned Aerial Vehicles(UAV) and Light Detection and Ranging (LiDAR) systems have contributedto this field with increased accuracy in tree phenotyping. Moreover, methods combiningUAV LiDAR surveying and machine learning (ML) have also emerged to enhanceestimates of single tree traits. Here, we utilizeda UAV LiDAR system to survey a Norway spruce forest in Davos, Switzerland, where adetailed field-based inventory served as ground truth data. Our objectives were (i) togain insights into variation and gradients of tree height and (ii)to evaluate whether such insights may prove useful as contextual information toimprove predictions of stem diameter and tree-level biomass. We segmented the pointcloud data scene into individual canopies and treated the LiDAR derived tree height asthe variable of interest.We then used local indicators of spatial association to detect the significant localcontext, and defined tree neighborhoods within the forest. Then,we extracted metricsfromthe neighborhoods and introduced them in a ML regression experiment toevaluate predictions of individual tree diameter.The focus was on comparing performance of tree diameter predictions betweenregression models that either consider neighborhood metrics (i.e. context-awaremodels), or not. Next, AGB was estimated from the tree height derivedfrom theUAV LiDAR survey, the predicted tree diameter and allometry. The benefits ofcontext awareness were assessed in terms of accuracy gained in estimating AGB. Weobtained results of different machine learning methods(i.e. AdaBoost, Lasso and Random Forest) and evaluated these based on nestedcross-validation. We applied this approach to two separate tree data sets within thesame site, one being clustered and continuous, the other discontinuousand scattered in separate sampling plots. In both cases, we found evidence ofenhanced AGB prediction performance in context-aware regressions, where the RMSEwas reduced by 4.0% and by 9.1%, respectively.These findings indicate that gradients in tree heights across the ecosystem may proxyfor local microclimate, edaphic conditions and biotic factors that influence tree growth,which can be leveraged to enhance predictions of AGB. The method proposed is fullynative to UAV LiDAR data.