Global and regional Canopy Height Maps (CHM). Created using machine learning models on high-resolution worldwide Maxar satellite imagery.

DATA OVERVIEW

Mapped attributes are

• above ground biomass, AGB;
• downed wood biomass, i.e., the sum of coarse and fine woody debris, DWB;
• canopy bulk density, CBD;
• canopy height,CH;
• canopy base height, CBH;
• and canopy fuel load, CFL.

Models were fit using auxiliary information that included lidar data from 20 acquisitions in Oregon and climate data. Measurements in plots of the Forest Inventory and Analysis program (FIA) were used to obtain plot-level ground observations for predictive modeling. Tree and transect measurements in FIA plots were respectively used to obtain plot-level values of AGB and DWB. To obtain plot-level values of CBD, CH, CBH and CFL, tree measurements in FIA plots were processed with FuelCalc. Plot level auxiliary variables were obtained intersecting the axiliary information layers with the FIA plots. Predictive models were random forest models in which a parametric component was added to model the error variance. The error variance was modeled as a power function of the predictive value and was used to produce uncertainty maps. A different model was fit for each variable and the resulting models were used to obtain maps of synthetic predictions for all areas covered by the 20 lidar acquisitions. The modeled error variance was used to generate uncertainty maps for the predictions of each response variable. Model accuracy was assessed globally (for the entire dataset) and separately for each one of the 20 lidar acquisitions included in the dataset.

Results from the accuracy assessment can be found in Appendix A and Appendix B of Mauro et al. (2021).

Each variable has two associated maps. These maps are named using the following convention where VARIABLE is the acronym for each variable (AGB, DWB, CBD, CH, CBH or CFL):

Predictions of forest attributes:

VARIABLE.tif

Standard deviation of modeled errors:

SD_VARIABLE.tif

### There are two additional rasters. The first one, year.tif is necessary to obtain the reference year for each lidar acquisition. The second one, forest_mask.tif provides a forest vs non-forest mask. Forested areas are coded as 1s and non-forested areas with no-datas. This mask is a resampled subset of the PALSAR JAXA 2014 ‘New global 25m-resolution PALSAR mosaic and forest/non-forest map (2007-2010) - version 1’ from the Japan Aerospace Exploration Agency Earth Observation Research Center (www.eorc.jaxa.jp/ALOS/en/palsar_fnf/fnf_index.htm). Its reference year is 2009. Models to predict forest attributes were created using ground observations in forested areas. For many applications it is advisable to use the provided mask to excluded non-forested areas from analyses. This can be done, for example, multiplying the desired raster by the forest mask. Exceptions to this may occur in relatively open forested lands where the mask eliminates areas that actually sustain forest. In those areas, the use of an add-hoc forest mask might be more appropriate. ### Reference year: year.tif ### Forest mask: forest_mask.tif ###

UNITS:

For a given variable, both predictions and standard deviation of model errors have the same units. These units are:

• Variable (Abreviation): Units

• Above ground biomass (AGB): Mg/ha

• Downed wood biomass (DWB):Mg/ha

• Canopy bulk density (CBD): Kg/m3 (Kilogram per cubic meter)

• Canopy height (CH): m

• Canopy base height (CBH): m

  Canopy fuel load (CFL):Mg/ha

COORDINATE REFERENCE SYSTEM:

The reference system for all maps is EPSG 5070

USAGE

These data are made freely available to the public and the scientific community in the belief that their wide dissemination will lead to greater understanding and new scientific insights.

Please include the following citation in any publication that uses these data:

Mauro, F., Hudak, A.T., Fekety, P.A., Frank, B., Temesgen, H., Bell, D.M., Gregory, M.J., McCarley, T.R., 2021. Regional Modeling of Forest Fuels and Structural Attributes Using Airborne Laser Scanning Data in Oregon. Remote Sensing 13. https://doi.org/10.3390/rs13020261

This dataset represents global tree heights based on a fusion of spaceborne-lidar data (2005) from the Geoscience Laser Altimeter System (GLAS) and ancillary geospatial data. See Simard et al. (2011) for details.

LANDFIRE's (LF) Forest Canopy Height (CH) describes the average height of the top of the vegetated canopy. CH measurement units are meters * 10 and extracted from Existing Vegetation Height (EVH). CH is assigned the midpoint of the EVH forested classes at non-disturbed locations. These products are provided for forested areas only.

These files are rasters of tree canopy heights derived from 23 sets of aerial lidar collected during 2014-2018 in Texas. Canopy heights are expressed in meters. These data were used to model golden-cheeked warbler habitat.

This dataset characterizes the global distribution, biomass, and canopy height of mangrove-forested wetlands based on remotely sensed and in situ field measurement data. Estimates of (1) mangrove aboveground biomass (AGB), (2) maximum canopy height (height of the tallest tree), and (3) basal-area weighted height (individual tree heights weighted in proportion to their basal area) for the nominal year 2000 were derived across a 30-meter resolution global mangrove ecotype extent map using remotely-sensed canopy height measurements and region-specific allometric models. Also provided are (4) in situ field measurement data for selected sites across a wide variety of forest structures (e.g., scrub, fringe, riverine and basin) in mangrove ecotypes of the global equatorial region. Within designated plots, selected trees were identified to species and diameter at breast height (DBH) and tree height was measured using a laser rangefinder or clinometer. Tree density (the number of stems) can be estimated for each plot and expressed per unit area. These data were used to derive plot-level allometry among AGB, basal area weighted height (Hba), and maximum canopy height (Hmax) and to validate the remotely sensed estimates.

Although the random forest algorithm has been widely applied to remotely sensed data to predict characteristics of forests, such as tree canopy height, the effect of spatial non-stationarity in the modeling process is oftentimes neglected. Previous studies have proposed methods to address the spatial variance at local scales, but few have explored the spatial autocorrelation pattern of residuals in modeling tree canopy height or investigated the relationship between canopy height and model performance. By combining Light Detection and Ranging (LiDAR) and Landsat datasets, we used spatially-weighted geographical random forest (GRF) and traditional random forest (TRF) methods to predict tree canopy height in a mixed dry forest woodland in complex mountainous terrain. Comparisons between TRF and GRF models show that the latter can lower predefined extreme residuals, and thus make the model performance relatively stronger. Moreover, the relationship between model performance and degree of variation of true canopy height can vary considerably within different height quantiles. Both models are likely to present underestimates and overestimates when the corresponding tree canopy heights are high (>95% quantile) and low (

Data Use
License: CC-BY-NC-SA 4.0
Recommended Citation: Wang, H., Seaborn, T., & Wang, Z. (2021). Data from: Modeling tree canopy height using machine learning over mixed vegetation landscapes [Data set]. University of Idaho. https://doi.org/10.7923/VJ7D-KS92

This dataset characterizes canopy heights of mangrove-forested wetlands globally for 2015 at 12-m resolution. Estimates of maximum canopy height (height of the tallest tree) were derived from the German Space Agency's TanDEM-X data that produced global digital surface models. Also provided are Lidar estimates of canopy height based on the GEDI instrument, which were used for training and validation of the TanDEM-X estimates of forest height. The coverage of these data follows Global Mangrove Watch's mangrove extent maps. These spatially explicit maps of mangrove canopy height can be used to assess local-scale geophysical and environmental conditions that may regulate forest structure and carbon cycle dynamics. Maps revealed a wide range of canopy heights, including maximum values (>60 m) that surpass maximum heights of other forest types. Maps are provided in cloud optimized GeoTIFF format, and mangrove heights for individual GEDI tiles are compiled in a comma separated values (CSV) files.

This web map is a subset of Global Canopy Height 2020 Image Layer. Many maps provide details of land classification types such as grassland, scrub, or tree cover. However, the total amount of biomass in these areas can vary greatly. Better mapping of the canopy height is important for understanding potential biodiversity, ecosystem function, or loss of carbon biomass due to deforestation, development, or fires. Scientists at ETH Zurich created a method to estimate canopy height using the best available LiDAR from space and airborne sensors, including the Global Ecosystem Dynamics Investigation (GEDI) on board the International Space Station.These LiDAR observations were used as ground truth data to train a deep convolutional neural network to regress canopy height from Sentinel-​2 images at a 10 m spatial resolution. The accuracy of the vegetation height is ± 5 m.This map was financed in a partnership with Barry Callebaut AG, the world's largest chocolate maker, to better allow companies to track deforestation in their supply chains. The intent is objective, highly automated tool to guide sustainable agribusiness.More about this project may be read at ETH Zurich website here: https://prs.igp.ethz.ch/research/current_projects/automated_large-scale_high_carbon_stock.html

multispectral imagery from Sentinel-2

Height of the top of canopy above bare earth (Canopy Height Model (CHM)); data are mosaicked over AOP footprint; mosaicked onto a spatially uniform grid at 1 m spatial resolution in 1 km by 1 km tiles. Data are provided in geotiff format.

LANFIRE’s (LF) Remap Forest Canopy Height (CH) describes the average height of the top of the canopy for a stand. In disturbed locations CH is calculated from linear regression equations derived from Forest Vegetation Simulator (FVS) plot data output, but at non-disturbed locations it is assigned the midpoint of Fuel Vegetation Height (FVH) forested classes. In some instances, LF Remap assumes the potential burnable biomass in the tree canopy has been accounted for in the surface fuel model. For example, young or short conifer stands where the trees are represented by a shrub type fuel model will not have canopy characteristics. LF Remap Annual Disturbance products are incorporated into CH to provide informed changes by disturbance type, severity, and time since disturbance (TSD). Annual Disturbance products provide a pre-disturbance scenario represented by LF Remap existing vegetation products, the reporting of pre-disturbance scenarios helps to calculate CH. Vegetation adjustments are then modeled in disturbance areas based on disturbance type and severity. CH is then used in the calculation of Canopy Bulk Density (CBD) and Canopy Base Height (CBH). CH supplies information for fire behavior models, such as FARSITE (Finney 1998), that can determine; the starting point for embers in the spotting model, wind reductions, and the volume of crown fuels. CH also has capable fuels functionality, capable fuels calculate TSD assignments for disturbed areas using an “effective year.” For example, year 2020 fuels may be calculated for the year 2020. This new process considers all the existing disturbances included in LF Remap and adjusts the TSD for these to the effective year (2020 in this example), making the products "2020 capable fuels." More information about capable fuels can be found at https://www.landfire.gov/lf_remap.php.

LANDFIRE's 2023 Update (LF 2023) Forest Canopy Height (CH) describes the average height of the top of the canopy for a stand. CH is used in the calculation of Forest Canopy Bulk Density (CBD) and Base Height (CBH). CH supplies information for fire behavior models, such as FARSITE (Finney 1998), that can determine the starting point of embers in the spotting model, wind reductions, and the volume of crown fuels. To create CH, LANDFIRE's Existing Vegetation Height (EVH) product must be produced first. EVH is a continuous scaled product which assigns height to all life forms in the LF data, this product is created using an image-based process (within the Conterminous United States (CONUS)) to assess canopy structure for areas disturbed in the past twenty years. CH is then derived from EVH by assigning bins of 13 height classes for fuel production and use in fire behavior software. In LF 2023, fuel products are created with LF 2016 Remap vegetation in areas that were un-disturbed in the last twenty years. To designate disturbed areas where CH is modified, the aggregated Annual Disturbance products from 2014 to 2023 in the LF Fuel Disturbance (FDist) product are used. All existing disturbances between 2014-2023 are represented in LF 2023, and the products are intended to be used in 2024 (the year of release). When using any product from the LF 2023 fuel product suite, users should consider adjusting fuel layers for disturbances that occurred after the end of the 2023 fiscal year (after October 1st, 2023). Disturbances that occurred after the end of the 2023 fiscal year are not accounted for within LF 2023 fuel products. Learn more about LF 2023 at https://www.landfire.gov/data/lf2023.

In this study, the accuracy of forest canopy height estimation was assessed using Synthetic Aperture Radar (SAR), including backscatter and Polarimetric SAR (PolSAR), as well as optical indices derived from optical imagery, and Random Forest (RF) and Support Vector Machine (SVM) models were applied by using canopy heights derived from Light Detection and Ranging (LiDAR) as a reference for validation. Accurate measurement of canopy height is critical for effective forest management, biodiversity conservation, and climate change analysis, so this study attempted to address the challenges posed by traditional measurement methods, which are time-consuming and limited in scope. SAR with its all-weather, day and night imaging capability, has the distinct advantage of being able to continuously monitor forest canopy dynamics over a wide area, thus overcoming the spatial time and cost constraints of ground-based observations. Approaches in this study involved pre-processing of SAR and LiDAR data to reduce inherent data inaccuracies, as well as calculating optical indices to facilitate indirect estimation of canopy height. This study provided a comparative assessment of the performance of RF and SVM models using various data integrations, highlighted the higher accuracy was achieved through the synergistic combination of PolSAR and optical indices. The results showed that the data-integrated approach improved the accuracy of canopy height estimation, with the RF model performing slightly better than the SVM model in terms of prediction under the optimal data configurations of the two models in this study. These findings support the advanced application of incorporating remote sensing techniques, validated against LiDAR benchmarks, as a viable strategy for refining forest canopy height estimation, thereby providing insights for forest management and ecological modelling programs.

Method:This lidar derivative provides information about vegetation height. The 3-foot resolution raster was produced from the 2020 Quality Level 1 classified lidar point cloud, which was provided by Sanborn Map Company, Inc. Tukman Geospatial developed the CHM from the classified point cloud using the following processing steps in LasTools:Create Tiles (lastile)Height Normalize the Point Cloud (lasheight)Set points classified as buildings to 0 heightThin the remaining points, taking the highest point in a 1.5 x 1.5 foot area (lasthin)Convert the thinned point cloud to a DEM (las2dem)

The data was developed based on a horizontal projection/datum of NAD83 (2011).

Lidar was collected in early 2020, while no snow was on the ground and rivers were at or below normal levels. To postprocess the lidar data to meet task order specifications and meet ASPRS vertical accuracy guidelines, Sanborn Map Company, Inc., utilized a total of 25 ground control points that were used to calibrate the lidar to known ground locations established throughout the project area.

An additional 125 independent accuracy checkpoints, 70 in Bare Earth and Urban landcovers (70 NVA points), 55 in Tall Grass and Brushland/Low Trees categories (55 VVA points), were used to assess the vertical accuracy of the data. These check points were not used to calibrate or post process the data.Uses and Limitations: The CHM provides a raster depiction of the highest vegetation returns for each 3x3 foot raster cell across Santa Clara County. The layer is useful for myriad vegetation and forest-related analysis and is an important input to the automated processes used to develop the Santa Clara fine scale vegetation map. However, this data product was produced based on a rapid, fully automated point cloud classification and was not manually edited. As such, it includes some ‘false positives’ – pixels with a canopy height in the raster that aren’t vegetation. These false positives include noise from water aboveground non-vegetation returns from bridge decks, powerlines, and edges of buildings.

Related Datasets: This dataset is part of a suite of lidar of derivatives for Santa Clara County. See table 1 for a list of all the derivatives. Table 1. lidar derivatives for Santa Clara CountyDatasetDescriptionLink to DataLink to DatasheetCanopy Height ModelPixel values represent the aboveground height of vegetation and trees.https://vegmap.press/clara_chmhttps://vegmap.press/clara_chm_datasheetCanopy Height Model – Veg Returns OnlySame as canopy height model, but does not include lidar returns labelled as ‘unclassified’ (uses only returns classified as vegetation)https://vegmap.press/clara_chm_veg_returnshttps://vegmap.press/clara_chm_veg_returns_datasheetCanopy CoverPixel values represent the presence or absence of tree canopy or vegetation greater than or equal to 15 feet tall.https://vegmap.press/clara_coverhttps://vegmap.press/clara_cover_datasheetCanopy Cover – Veg Returns OnlySame as canopy height model, but does not include lidar returns labelled as ‘unclassified’ (uses only returns classified as vegetation)https://vegmap.press/clara_cover_veg_returnshttps://vegmap.press/clara_cover_veg_returns_datasheet HillshadeThis depicts shaded relief based on the Hillshade. Hillshades are useful for visual reference when mapping features such as roads and drainages and for visualizing physical geography. https://vegmap.press/clara_hillshadehttps://vegmap.press/clara_hillshade_datasheetDigital Terrain ModelPixel values represent the elevation above sea level of the bare earth, with all above-ground features, such as trees and buildings, removed. The vertical datum is NAVD88 (GEOID18).https://vegmap.press/clara_dtmhttps://vegmap.press/clara_dtm_datasheetDigital Surface ModelPixel values represent the elevation above sea level of the highest surface, whether that surface for a given pixel is the bare earth, the top of vegetation, or the top of a building.https://vegmap.press/clara_dsmhttps://vegmap.press/clara_dsm_datasheet

This data set contains Level-2 geolocated surface elevation and canopy height measurements collected by the NASA Land, Vegetation, and Ice Sensor (LVIS) Facility, an imaging lidar and camera sensor suite.

A tree crowns layer was derived from 2018 NAIP and 2019 LiDAR, and then each tree crown polygon was populated with the 95th percentile nDSM (height above ground) values from LiDAR collected in 2014 and in 2019. Object-based image analysis techniques (OBIA) were employed to extract potential tree crowns including the area of the crown and trees using the best available remotely sensed and vector GIS datasets. OBIA systems work by grouping pixels into meaningful objects based on their spectral and spatial properties, while taking into account boundaries imposed by existing vector datasets. Within the OBIA environment a rule-based expert system was designed to effectively mimic the process of manual image analysis by incorporating the elements of image interpretation (color/tone, texture, pattern, location, size, and shape) into the classification process. A series of morphological procedures were employed to insure that the end product is both accurate and cartographically pleasing. Following the automated OBIA mapping a detailed manual review of the dataset was carried out at a scale of 1:2000 and all observable errors were corrected.

This layer represents distance between the ground and the top of the canopy. Canopy height is a proxy for aboveground biomass and the amount of foliage that may be consumed in a canopy fire. Since LANDFIRE doesn't support a NoData value, all NoData pixels in canopy fuel metrics were set to 0 in the Landscape files. (e.g., canopy cover was set to 0 in all NoData locations). Topographic data and surface fuel model remain unaltered.

Goddard’s LiDAR, Hyperspectral, and Thermal Imager (G-LiHT) mission utilizes a portable, airborne imaging system that aims to simultaneously map the composition, structure, and function of terrestrial ecosystems. G-LiHT primarily focuses on a broad diversity of forest communities and ecoregions in North America, mapping aerial swaths over the Conterminous United States (CONUS), Alaska, Puerto Rico, and Mexico.The purpose of G-LiHT’s Canopy Height Model Keyhole Markup Language (KML) data product (GLCHMK) is to provide LiDAR-derived maximum canopy height and canopy variability information to aid in the study and analysis of biodiversity and climate change. Scientists at NASA’s Goddard Space Flight Center began collecting data over locally-defined areas in 2011 and that the collection will continue to grow as aerial campaigns are flown and processed. GLCHMK data are processed as a Google Earth overlay KML file at a nominal 1 meter spatial resolution over locally-defined areas. A low resolution browse is also provided showing the canopy height with a color map applied in JPEG format.

DATA OVERVIEW

Mapped attributes are

• above ground biomass, AGB;
• downed wood biomass, i.e., the sum of coarse and fine woody debris, DWB;
• canopy bulk density, CBD;
• canopy height,CH;
• canopy base height, CBH;
• and canopy fuel load, CFL.

Models were fit using auxiliary information that included lidar data from 20 acquisitions in Oregon and climate data. Measurements in plots of the Forest Inventory and Analysis program (FIA) were used to obtain plot-level ground observations for predictive modeling. Tree and transect measurements in FIA plots were respectively used to obtain plot-level values of AGB and DWB. To obtain plot-level values of CBD, CH, CBH and CFL, tree measurements in FIA plots were processed with FuelCalc. Plot level auxiliary variables were obtained intersecting the axiliary information layers with the FIA plots. Predictive models were random forest models in which a parametric component was added to model the error variance. The error variance was modeled as a power function of the predictive value and was used to produce uncertainty maps. A different model was fit for each variable and the resulting models were used to obtain maps of synthetic predictions for all areas covered by the 20 lidar acquisitions. The modeled error variance was used to generate uncertainty maps for the predictions of each response variable. Model accuracy was assessed globally (for the entire dataset) and separately for each one of the 20 lidar acquisitions included in the dataset.

Results from the accuracy assessment can be found in Appendix A and Appendix B of Mauro et al. (2021).

Each variable has two associated maps. These maps are named using the following convention where VARIABLE is the acronym for each variable (AGB, DWB, CBD, CH, CBH or CFL):

Predictions of forest attributes:

VARIABLE.tif

Standard deviation of modeled errors:

SD_VARIABLE.tif

### There are two additional rasters. The first one, year.tif is necessary to obtain the reference year for each lidar acquisition. The second one, forest_mask.tif provides a forest vs non-forest mask. Forested areas are coded as 1s and non-forested areas with no-datas. This mask is a resampled subset of the PALSAR JAXA 2014 ‘New global 25m-resolution PALSAR mosaic and forest/non-forest map (2007-2010) - version 1’ from the Japan Aerospace Exploration Agency Earth Observation Research Center (www.eorc.jaxa.jp/ALOS/en/palsar_fnf/fnf_index.htm). Its reference year is 2009. Models to predict forest attributes were created using ground observations in forested areas. For many applications it is advisable to use the provided mask to excluded non-forested areas from analyses. This can be done, for example, multiplying the desired raster by the forest mask. Exceptions to this may occur in relatively open forested lands where the mask eliminates areas that actually sustain forest. In those areas, the use of an add-hoc forest mask might be more appropriate. ### Reference year: year.tif ### Forest mask: forest_mask.tif ###

UNITS:

For a given variable, both predictions and standard deviation of model errors have the same units. These units are:

• Variable (Abreviation): Units

• Above ground biomass (AGB): Mg/ha

• Downed wood biomass (DWB):Mg/ha

• Canopy bulk density (CBD): Kg/m3 (Kilogram per cubic meter)

• Canopy height (CH): m

• Canopy base height (CBH): m

  Canopy fuel load (CFL):Mg/ha

COORDINATE REFERENCE SYSTEM:

The reference system for all maps is EPSG 5070

USAGE

These data are made freely available to the public and the scientific community in the belief that their wide dissemination will lead to greater understanding and new scientific insights.

Please include the following citation in any publication that uses these data:

Mauro, F., Hudak, A.T., Fekety, P.A., Frank, B., Temesgen, H., Bell, D.M., Gregory, M.J., McCarley, T.R., 2021. Regional Modeling of Forest Fuels and Structural Attributes Using Airborne Laser Scanning Data in Oregon. Remote Sensing 13. https://doi.org/10.3390/rs13020261