Australia-wide 30 m machine learning-derived canopy height models composites: best pick and median

This dataset is part of the OzTreeMap project and provides two new 30 m spatial resolution canopy height products for continental Australia: (1) the best-pick canopy height model (pick-CHM); and (2) the median canopy height model (med-CHM). Both products were generated and validated as part of the study titled “Accuracy of Machine Learning-Derived Canopy Height Models at Continental Scale.”\n\nThe pick-CHM is a composite model in which each 30 m pixel adopts the most accurate canopy height value among four publicly available machine learning-derived CHMs—Tolan et al. (2024), Lang et al. (2023), Potapov et al. (2021), and Liao et al. (2020)—based on the vegetation class (Scarth et al., 2019) that the pixel represents and our vegetation-specific accuracy assessment (see lineage). The med-CHM represents a pixel-wise median composite of the same four CHMs and achieved the highest overall accuracy when validated against 22,967 km² of reference airborne point cloud data across 16 Australian vegetation classes.\n\nBoth datasets are provided as single-band GeoTIFF rasters in EPSG:3577 (Australian Albers) coordinate reference system, with 30 m spatial resolution and float32 data type. These CHMs offer improved accuracy and spatial consistency compared to the individual global products supporting continental-scale applications in forest structure monitoring, carbon accounting, and ecosystem assessment.\n\nReferences\nLang, N., Jetz, W., Schindler, K., Wegner, J.D., 2023. A high-resolution canopy height model of the Earth. Nat Ecol Evol 7, 1778–1789. https://doi.org/10.1038/s41559-023-02206-6\n\nLiao, Z., van Dijk, A.I.J.M., He, B., Larraondo, P.R., Scarth, P.F., 2020. Woody vegetation cover, height and biomass at 25 m resolution across Australia derived from multiple site, airborne and satellite observations. Int. J. Appl. Earth Obs. Geoinf. 93, 102209. https://doi.org/10.1016/j.jag.2020.102209\n\nPotapov, P., Li, X., Hernandez-Serna, A., Tyukavina, A., Hansen, M.C., Kommareddy, A., Pickens, A., Turubanova, S., Tang, H., Silva, C.E., Armston, J., Dubayah, R., Blair, J.B., Hofton, M., 2021. Mapping global forest canopy height through integration of GEDI and Landsat data. Remote Sens. Environ. 253, 112165. https://doi.org/10.1016/j.rse.2020.112165\n\nScarth, P., Armston, J., Lucas, R., Bunting, P., 2019. A structural classification of Australian vegetation using ICESat/GLAS, ALOS PALSAR, and Landsat sensor data. Remote Sens. 11, 147. https://doi.org/10.3390/rs11020147\n\nTolan, J., Yang, H.-I., Nosarzewski, B., Couairon, G., Vo, H.V., Brandt, J., Spore, J., Majumdar, S., Haziza, D., Vamaraju, J., Moutakanni, T., Bojanowski, P., Johns, T., White, B., Tiecke, T., Couprie, C., 2024. Very high resolution canopy height maps from RGB imagery using self-supervised vision transformer and convolutional decoder trained on aerial lidar. Remote Sens. Environ. 300, 113888. https://doi.org/10.1016/j.rse.2023.113888\nLineage: A total of 26,987 LiDAR and photogrammetry point cloud tiles (1–4 km² each) were obtained from the Elevation and Depth (ELVIS) and Terrestrial Ecosystem Research Network (TERN) open repositories, representing a 5% stratified sample designed to match the distribution of Australia’s 16 vegetation structure classes (Scarth et al., 2019). For each tile, a 0.5 m canopy height model (CHM) was generated using the pit-free algorithm (Khosravipour et al., 2014), and individual tree crowns were delineated with the Dalponte segmentation algorithm (Dalponte & Coomes, 2016) using vegetation-specific optimized parameters (Pucino et al., 2025, under review). \n\nThe resulting point-cloud-derived CHMs served as reference data for evaluating the vertical accuracy of four publicly available satellite-based machine-learning or deep learning-derived CHMs: (1) Lang et al. (2023); (2) Liao et al. (2020); (3) Potapov et al. (2021); and (4) Tolan et al. (2024). All datasets were co-registered and resampled to 30 m resolution. Pixel-wise error metrics were computed, and a combined score defined for each vegetation class which publicly available dataset is the most accurate.\n\nThree new continental-scale 30 m CHMs were then produced: (i) a pixel-wise median composite; (ii) a vegetation-class-specific best-pick composite; and (iii) a deep-learning CHM derived from a multi-layer perceptron (MLP - not publicly available).\n\nNote: this document's Start Date and End Date indicate the nominal dates of the datasets we tested, not the publication dates of their associated articles.\n\nReferences\n\nDalponte, M., Coomes, D.A., 2016. Tree-centric mapping of forest carbon density from airborne laser scanning and hyperspectral data. Methods Ecol. Evol. 7, 1236–1245. https://doi.org/10.1111/2041-210X.12575\n\nKhosravipour, A., Skidmore, A.K., Isenburg, M., Wang, T., Hussin, Y.A., 2014. Generating Pit-free Canopy Height Models from Airborne Lidar. Photogramm. Eng. Remote Sens.\t 80, 863–872. https://doi.org/10.14358/PERS.80.9.863\n\nLang, N., Jetz, W., Schindler, K., Wegner, J.D., 2023. A high-resolution canopy height model of the Earth. Nat Ecol Evol 7, 1778–1789. https://doi.org/10.1038/s41559-023-02206-6\n\nLiao, Z., van Dijk, A.I.J.M., He, B., Larraondo, P.R., Scarth, P.F., 2020. Woody vegetation cover, height and biomass at 25-m resolution across Australia derived from multiple site, airborne and satellite observations. Int. J. Appl. Earth Obs. Geoinf. 93, 102209. https://doi.org/10.1016/j.jag.2020.102209\n\nPotapov, P., Li, X., Hernandez-Serna, A., Tyukavina, A., Hansen, M.C., Kommareddy, A., Pickens, A., Turubanova, S., Tang, H., Silva, C.E., Armston, J., Dubayah, R., Blair, J.B., Hofton, M., 2021. Mapping global forest canopy height through integration of GEDI and Landsat data. Remote Sens. Environ. 253, 112165. \n\nPucino, N., McVicar, T. R., Levick, S.R., van Dijk, A.I.J.M., 2025. Assessing optimization strategies for unsupervised individual tree crown detection and delineation to support continental-scale inventories: role of vegetation type and point cloud data density. Sci. of Remote Sens. (Under Review) \n\nScarth, P., Armston, J., Lucas, R., Bunting, P., 2019. A structural classification of Australian vegetation using ICESat/GLAS, ALOS PALSAR, and Landsat sensor data. Remote Sens. 11, 147. https://doi.org/10.3390/rs11020147\n\nTolan, J., Yang, H.-I., Nosarzewski, B., Couairon, G., Vo, H.V., Brandt, J., Spore, J., Majumdar, S., Haziza, D., Vamaraju, J., Moutakanni, T., Bojanowski, P., Johns, T., White, B., Tiecke, T., Couprie, C., 2024. Very high resolution canopy height maps from RGB imagery using self-supervised vision transformer and convolutional decoder trained on aerial lidar. Remote Sens. Environ. 300, 113888. https://doi.org/10.1016/j.rse.2023.113888\n\n

本数据集隶属于OzTreeMap项目，面向澳大利亚大陆提供两款全新的30米空间分辨率冠层高度产品：(1) 最佳选择冠层高度模型（best-pick Canopy Height Model, pick-CHM）；(2) 中值冠层高度模型（median Canopy Height Model, med-CHM）。两款产品均为题为《大陆尺度机器学习衍生冠层高度模型的精度评估》的研究中生成并验证的成果。

pick-CHM为合成模型，其每个30米像素将根据该像素所属的植被类型（Scarth等，2019）以及本研究针对植被类型的精度评估结果（详见谱系说明），从四款公开可用的机器学习衍生冠层高度模型（Tolan等，2024、Lang等，2023、Potapov等，2021、Liao等，2020）中选取精度最高的冠层高度值作为该像素的取值。

med-CHM则为上述四款冠层高度模型的逐像素中值合成产物，在针对澳大利亚16种植被类型的22967平方千米参考机载点云数据进行验证时，该模型实现了最高的总体精度。

两款数据集均采用EPSG:3577（澳大利亚阿尔伯斯）坐标参考系，以单波段GeoTIFF栅格格式存储，空间分辨率为30米，数据类型为32位浮点型（float32）。相较于单款全球产品，本研究生成的冠层高度模型具备更优的精度与空间一致性，可支撑大陆尺度的森林结构监测、碳核算及生态系统评估等应用。

参考文献
Lang, N., Jetz, W., Schindler, K., Wegner, J.D., 2023. 全球高分辨率冠层高度模型. 《自然-生态学与进化》, 7卷, 1778–1789页. https://doi.org/10.1038/s41559-023-02206-6

Liao, Z., van Dijk, A.I.J.M., He, B., Larraondo, P.R., Scarth, P.F., 2020. 基于多站点机载与卫星观测数据生成的澳大利亚25米分辨率木质植被覆盖度、高度及生物量. 《国际应用地球观测与地理信息学报》, 93卷, 102209页. https://doi.org/10.1016/j.jag.2020.102209

Potapov, P., Li, X., Hernandez-Serna, A., Tyukavina, A., Hansen, M.C., Kommareddy, A., Pickens, A., Turubanova, S., Tang, H., Silva, C.E., Armston, J., Dubayah, R., Blair, J.B., Hofton, M., 2021. 结合GEDI与Landsat数据绘制全球森林冠层高度. 《遥感环境》, 253卷, 112165页. https://doi.org/10.1016/j.rse.2020.112165

Scarth, P., Armston, J., Lucas, R., Bunting, P., 2019. 基于ICESat/GLAS、ALOS PALSAR及Landsat传感器数据的澳大利亚植被结构分类. 《遥感》, 11卷, 147页. https://doi.org/10.3390/rs11020147

Tolan, J., Yang, H.-I., Nosarzewski, B., Couairon, G., Vo, H.V., Brandt, J., Spore, J., Majumdar, S., Haziza, D., Vamaraju, J., Moutakanni, T., Bojanowski, P., Johns, T., White, B., Tiecke, T., Couprie, C., 2024. 基于机载激光雷达训练的自监督视觉Transformer（Vision Transformer）与卷积解码器生成的RGB影像超高分辨率冠层高度图. 《遥感环境》, 300卷, 113888页. https://doi.org/10.1016/j.rse.2023.113888

谱系说明：研究从高程与深度（Elevation and Depth, ELVIS）及陆地生态系统研究网络（Terrestrial Ecosystem Research Network, TERN）的公开数据库中获取了共计26987个激光雷达（LiDAR）与摄影测量点云瓦片（单瓦片面积1~4平方千米），该样本为按比例分层抽取的5%样本，其分布与澳大利亚16种植被结构类型（Scarth等，2019）的分布相匹配。针对每个瓦片，本研究采用无坑算法（Khosravipour等，2014）生成了0.5米分辨率的冠层高度模型（CHM），并基于针对植被类型优化的参数（Pucino等，2025，待刊），使用Dalponte分割算法（Dalponte与Coomes，2016）勾绘单木树冠。

由点云衍生得到的冠层高度模型被用作参考数据，用于评估四款公开可用的星载机器学习或深度学习衍生冠层高度模型的垂直精度：(1) Lang等（2023）；(2) Liao等（2020）；(3) Potapov等（2021）；(4) Tolan等（2024）。所有数据集均经过配准并重采样至30米分辨率。研究计算了逐像素误差指标，并为每种植被类型定义了综合得分，以确定哪款公开数据集的精度最优。

随后本研究生成了三款全新的大陆尺度30米分辨率冠层高度模型：(i) 逐像素中值合成模型；(ii) 基于植被类型的最佳选择合成模型；(iii) 由多层感知机（Multi-Layer Perceptron, MLP，未公开）衍生的深度学习冠层高度模型。

备注：本文档中的起始日期与结束日期代表本研究测试数据集的名义日期，而非相关文献的发表日期。

参考文献
Dalponte, M., Coomes, D.A., 2016. 基于机载激光扫描与高光谱数据的森林碳密度单木制图. 《生态学与进化方法》, 7卷, 1236–1245页. https://doi.org/10.1111/2041-210X.12575

Khosravipour, A., Skidmore, A.K., Isenburg, M., Wang, T., Hussin, Y.A., 2014. 基于机载激光雷达生成无坑冠层高度模型. 《摄影测量工程与遥感》, 80卷, 863–872页. https://doi.org/10.14358/PERS.80.9.863

Lang, N., Jetz, W., Schindler, K., Wegner, J.D., 2023. 全球高分辨率冠层高度模型. 《自然-生态学与进化》, 7卷, 1778–1789页. https://doi.org/10.1038/s41559-023-02206-6

Liao, Z., van Dijk, A.I.J.M., He, B., Larraondo, P.R., Scarth, P.F., 2020. 基于多站点机载与卫星观测数据生成的澳大利亚25米分辨率木质植被覆盖度、高度及生物量. 《国际应用地球观测与地理信息学报》, 93卷, 102209页. https://doi.org/10.1016/j.jag.2020.102209

Potapov, P., Li, X., Hernandez-Serna, A., Tyukavina, A., Hansen, M.C., Kommareddy, A., Pickens, A., Turubanova, S., Tang, H., Silva, C.E., Armston, J., Dubayah, R., Blair, J.B., Hofton, M., 2021. 结合GEDI与Landsat数据绘制全球森林冠层高度. 《遥感环境》, 253卷, 112165页.

Pucino, N., McVicar, T. R., Levick, S.R., van Dijk, A.I.J.M., 2025. 支撑大陆尺度森林清查的无监督单木树冠检测与勾划优化策略评估：植被类型与点云数据密度的作用. 《遥感科学》（待刊）

Scarth, P., Armston, J., Lucas, R., Bunting, P., 2019. 基于ICESat/GLAS、ALOS PALSAR及Landsat传感器数据的澳大利亚植被结构分类. 《遥感》, 11卷, 147页. https://doi.org/10.3390/rs11020147

Tolan, J., Yang, H.-I., Nosarzewski, B., Couairon, G., Vo, H.V., Brandt, J., Spore, J., Majumdar, S., Haziza, D., Vamaraju, J., Moutakanni, T., Bojanowski, P., Johns, T., White, B., Tiecke, T., Couprie, C., 2024. 基于机载激光雷达训练的自监督视觉Transformer（Vision Transformer）与卷积解码器生成的RGB影像超高分辨率冠层高度图. 《遥感环境》, 300卷, 113888页. https://doi.org/10.1016/j.rse.2023.113888