Norfolk Island landcover mapping from airborne LiDAR

A landcover classification produced from airborne LiDAR and spaceborne datasets to assist in the mapping and management of woody weeds on Norfolk Island.\nLineage: We applied Random Forest modelling for classifying the distribution of landcover types, and mapping their occurrence probabilities spatially, from remotely sensed datasets. Terrain and vegetation structural metrics derived from airborne LiDAR formed the primary predictor variables, together with satellite imagery from Landsat-8 (multi-spectral) and ALOS-2 (synthetic aperture radar). The resulting classification outputs were robust (i.e., in terms of the F1 score and the class specific precision and recall statistics, scoring between 70-90%) for the four woody weeds of primary interest (Red guava, Hawaiian holly, African olive, and Cotoneaster). LiDAR derived metrics representing elevation above sea level, canopy height and structural variability accounted for the highest variable importance scores in the model. Primary outputs include probability maps and GIS layers for the distribution of the four woody weeds that were the target of the study.

本数据集为基于机载激光雷达（airborne LiDAR）与星载数据集生成的土地覆盖分类成果，旨在协助诺福克岛（Norfolk Island）开展木本杂草的制图与治理工作。
数据生成流程：本研究依托遥感数据集，采用随机森林（Random Forest）建模方法完成土地覆盖类型分布分类，并空间化绘制各类土地覆盖的发生概率分布。本研究的核心预测变量包括由机载激光雷达提取的地形与植被结构特征量，以及Landsat-8多光谱卫星影像与ALOS-2合成孔径雷达（synthetic aperture radar）影像。最终生成的分类结果针对本研究重点关注的四种木本杂草（红番石榴、夏威夷冬青、非洲橄榄、栒子属植物）表现稳健：以F1分数（F1 score）及类别专属精确率（precision）、召回率（recall）统计值来看，其得分区间为70%-90%。模型中变量重要性得分最高的特征量为机载激光雷达提取的海拔高程、冠层高度与结构变异性指标。本数据集的核心输出成果为本次研究目标的四种木本杂草的分布概率地图与GIS图层（GIS layers）。