Spatial distribution of maximum canopy height and heights percentiles in Canada at 250m spatial resolution

The canopy height maps were built to be included as covariates in the model to predict AGB in forest areas of Canada. We created wall-to-wall height metrics using ATL08 LiDAR products from the ICESat-2 satellite. The data was download for one-year period (October 2018 to October 2019). Points were filtered regarding solar background noise and atmospheric scattering, totaling 49,959 points distributed over the entire Canada. These points were associated with 10 ancillary variables primarily corresponding to structure information, such as seasonal Sentinel-1 VV and VH polarization, annual PALSAR-2 HH and HV polarization, annual clumping index, and also the MODIS NDVI summer season. Afterwards, the random forest algorithm was used to extrapolate ATL08 parameters and develop regression models with the abovementioned spatially continuous variables. The maximum height and height percentiles (h85 and h95) were estimated with an R² of approximately 0.61.<br>

本研究构建冠层高度图，将其作为协变量应用于加拿大林区地上生物量（Aboveground Biomass, AGB）的预测模型。本研究利用ICESat-2卫星的ATL08激光雷达（LiDAR）产品，生成了无缝隙全覆盖的冠层高度指标。本次研究所用数据的下载时段为2018年10月至2019年10月，共计一年。针对太阳背景噪声与大气散射效应，本研究对数据点进行筛选，最终得到分布于加拿大全境的49959个有效数据点。将上述数据点与10项辅助变量进行关联，这些变量主要涵盖林分结构相关信息，具体包括哨兵1号（Sentinel-1）卫星的季节性VV、VH极化数据、PALSAR-2的年度HH、HV极化数据、年度聚集指数，以及MODIS夏季归一化植被指数（NDVI）。随后，本研究采用随机森林算法对ATL08参数进行外推，并基于上述空间连续变量构建回归模型。最终估算得到的最大高度及高度百分位数（h85与h95）的决定系数（R²）约为0.61。<br>