GEDI forest transects
3 Jan 2021
We describe here how the Jura Mountains mapping website uses GEDI LIDAR (light detection and ranging) transects (PDF).
But first, some guidance for users.
Images giving the profiles of the density of biomass - technically the Plant Area Volume Density (PAVD) - along transects are made available by selecting "Transects" in the map menu. The map should then be rotated (use Alt + Shift keys) and zoomed to the resolution that makes the length of the transect (the blue line below the transects PAVD image) equal to the PAVD profile shown in the image.
OpenStreetMap technology is rarely applied at the local project-level as opposed to the global level. The Jura Mountains mapping website aims to overcome this shortcoming by piloting open-source OpenStreetMap project-level applications.
A recent development for application at the global level involves the Global Ecosystem Dynamics Investigation (GEDI) high-resolution earth-observation LIDAR data.
GEDI Level 1 and Level 2 products have been made available since late-2019. They allow the canopy height and vegetation density of forested areas to be estimated.
Level 3 products that grid spatially interpolated Level 2 footprint estimates of canopy cover, canopy height, leaf-area, and vertical foliage profiles were planned for release in mid-2020, and are presumably still planned for release.
Level 4 products - the highest GEDI product level - will eventually use calibrated models to derive estimates of above-ground biomass density footprints.
These footprints will then be used to estimate biomass within 1 km square cells in order to monitor the Earth's carbon cycle at the global level as a component of carbon accounting aimed at reducing the level of atmospheric carbon dioxide which drives climate change.
The three GEDI lasers mounted on the International Space Station fire 242 times each second to produce eight beam tracks which are spaced approximately 600 m apart in the cross-track direction (see specifications). Two beams are full-power beams and a third laser beam is split into two "coverage" beams. Every other shot of each of the beams is optically dithered across-track to give an additional beam tracks with irregularly spaced shots. Each beam samples the Earth at points that are approximately 60m apart along the beam track's orbit, resulting in almost continuous along-track coverage. Each illuminated footprint reflects LIDAR energy with a diameter of about 30 m and a vertical measurement accuracy of about 2o to 3o mm.
The focus at the project-level pilot is to evaluate the application of GEDI products to the built environment that comprises urban and nature-based infrastructure which blend vegetation and man-made structures.
As a first step, Level 2 GEDI metrics for transects through the Jura Mountain forests are being examined. The aim is to see if the vertical foliage profile can be used not only to monitor logging and other forestry activity but also to spatially quantify the built environment.
Web mapping uses OpenLayers since map rotation is needed to align images of PAVD profiles with GEDI transects. Drop-down menus are used now to select profiles, but this will be changed at some stage by using pop-up images.
Implementation is fairly straightforward. As regards data processing, the starting point is to map vegetation profiles along sections of the GEDI beam tracks that span forests in the Jura Mountains.
Images of PAVD profiles are obtained using a Jupyter Notebook published by the US Land Processes Distributed Active Archive Center (LP DACC).
As the Notebook explains, using a terminal (search for "cmd" , the Command Prompt in Windows 10), create a Conda environment and install and run the Notebook:
- conda create -n geditutorial -c conda-forge --yes python=3.8 h5py shapely geopandas pandas geoviews holoviews
- conda activate geditutorial
- conda install jupyter notebook
- jupyter notebook
Installing Conda can be a bit troublesome. For Windows 10 with Python already installed, Miniconda is adequate. After downloading and following instructions, let the installer set the PATH variable, which can checked with the terminal command:
- echo %PATH%
GEDI Level 2B data (as HDF5 formated files) then needs to be downloaded. The DAAC tutorial Notebook demonstration file is: GEDI02_B_2019170155833_O02932_T02267_02_001_01.h5
As explained in the Notebook, a NASA Earthdata account is needed and the .h5 file must be downloaded or copied into the same directory as the Notebook (for a simple installation, the Notebook and the .h5 file will be in c:/Users/admin where admin is a Windows 10 administrator account).
In running the Notebook, an error will probably be thrown when the PAVD profile of a single shot is printed, so the single-shot PAVD profile is not displayed - but it can be saved as a .png image by adding to the relevant Notebook cell:
Second, care must be taken in specifying the GeoJSON file that is used to define the boundary of the beam track that will be sampled. The tutorial explains that a GeoJSON file for the Redwood National Park must be downloaded from Earthdata (or other sources - see below) to the same directory as the Notebook and the .h5 file.
The Notebook in fact uses the maximum and minimum dimensions of an envelope that encapsulates the GeoJSON. It is usually wise to define the sampled area as a simple polygon. However, the GeoJSON file must be a denoted as a MultiPolygon. Fortunately the GeoJSON can be expressed in Web Mercator EPSG 4326 units (i.e., in terms of coordinates given by say Google maps).
Once the GeoJSON is loaded in the Notebook, the full GEDI orbit is plotted. By zooming the orbit to the area outlined by the GeoJSON file one can select the beam shot that will become the centre of the PAVD profile. Clearly, selecting a point that is at the centre of the intersection between the GeoJSON polygon and the orbit is the most sensible approach. However, by default (which can be changed), the PAVD profile is generated for a scan that covers 50 laser shots on either side of the selected shot. So different parts of the demarcated area can be sample by selecting different shots. The selected beam shot must then be entered in the Notebook.
The last section of the Notebook exports the section of the GEDI beam swathe that was used to calculate the PAVD profile across the area designated by the GeoJSON and centred on the selected shot. The swathe has (as shown on the Jura mountains GEDI map) two full-power beams with a third laser beam split into two coverage beams. Each of the beams is optically dithered across-track to give four additional beam tracks. The beam number for one of the eight beams - usually a full-power beam - must be entered in the Notebook.
The Plant Area Volume Density (PAVD) - the derivative of the cumulative Plant Area Index - that measures the volume density distribution of biomass is recognised as a sensitive measure of forest dynamics.
In the figures, the PAVD is plotted against height in order to show the vertical distribution of biomass in forested areas. Several strata with high PAVD values are generally of interest, namely biomass concentrated as follows:
- bottom - understory (trunks, shrubs and juvenile trees);
- middle - sub-canopy (branches and leaves of sub-canopy trees and trunks of the canopy trees in the understory);
- top - tree canopy (canopy branches and leaves).
- high top, middle or bottom densities can extend over a considerable distance (200-300 m);
- immature forests have densities that vary litle with height;
- some mountain tops appear to have exceptionally dense sub-canopies (snow-laden trees?).
Ground truth will establish whether or not these features, especially the first, reflect forestry practice.
The impact of topography on PAVD is well known, so it is probably not surprising that canopy heights are underestimated considerably for very steep slopes. One assumes that this means that the vertical distribution of biomass appears concentrated into a small height. At what point this effect may impact the identification of man-made structures is unclear.
This concern reflects a more general issue relating to high-resolution LIDAR earth observation of mixtures of man-made structures and vegetation that is discussed below.
Jura Mountain forested and non-forested areas (generally mountain pastures with scattered trees) are not unsurprisingly easily resolved. GEDI transects also clearly resolve man-made features down to the scale of 2m wide earthen tracks that are not tree covered.
The best way to identify GEDI orbits that intersect the area of interest (technically in terms of the Jupyter Notebook, the input bounding box) is to use the GEDI Finder. In the case of the Jura Mountains:
The response is a list of linked files equivalent to the link for the .h5 file used for the Notebook:
Publically available GEDI data started to be available in April 2019 and the latest data is for April 2020.
GEDI has the highest resolution and densest sampling of any LIDAR instrument in orbit to date. It has a nominal mission length of two years, starting in early-2019. The GEDI Level 2A data product demonstrates the enormous amount of data that is generated. The product is based on 156 layers for each of the eight beams, and includes ground elevation, canopy top height, relative return energy metrics (describing canopy vertical structure, for example), and many other interpreted products from the return waveforms.
Applications to date have tended to aim to improve maps of forest height and biomass across the globe. A notable derived service is Global Forest Canopy Height that can be downloaded. A programme to establish calibration plots has been set up. It seeks coincident small-footprint (at least 25 m in diameter) LIDAR data and field-measured plot inventory data. GEDI's success in measuring globally the mean biomass in 1 km cells will probably depend on the outcome of this calibration programme. Technologies based on Entwine and Caesium data storage and display and LOPoCS point-cloud streaming from pgpointcloud-extended PostgreSQL databases to serve biomass mapping will also be crucial.
With regard to the application of OpenStreetMap technology at the local project level, the main issue is whether or not GEDI and its successors will help in spatially quantifying the built environment in terms of vegetation and surface characteristics. The hypothesis here is that the construction and operation of an infrastructure project should be assessed, at least in part, in terms of the project's impact on its microclimate.
Several approaches for classifying vegetation and surface characteristics are used, notably GIT, LULC, LCZ, HERCULES and USVT. GIT (Green Infrastructure Typology) is suitable for use at the local scale with 50-100m data grids, but needs calibration, especially for mixed types of land cover. While GEDI's 30m diameter shots can help, the fundamental problem is that high-resolution GIT-type classification currently requires expensive airborne remote sensing data. Replacement by low-cost space-borne LIDAR data would be a game changer.
Aside from making use of the perhaps unexpected high accuracy of GEDI-generated digital terrain models over water (PDF) and in urban areas (PDF), and recognising the average GEDI geolocation error of about 8 to 11 m (PDF), the main GEDI focus is not the shot level but the 1000m mesh level, primarily to estimate mean biomass using models.
It is estimated that globally and certainly in the northern hemisphere, most of the Earth's 1000m grid cells will be intercepted by two GEDI beam swathes. There are numerous options for interpolating point data to continuous grids depending on the nature of the dataset. As a starting point, a kriging methodology similar to the one outlined in the GEDI Level 3 Algorithm Theoretical Basis Document (PDF) is being used. This is a good option for natural and continuous forest types but probably not for mixed land cover types (called fragmented or urban/suburban forests), vegetated built-up areas and nature-based infrastructure. These will probably need advanced interpolation methods such as co-kriging or fusion with auxiliary data having continuous spatial coverage such as satellite imagery and possibly OpenStreetMap data.
Microclimate classification is becoming the basis for infrastructure project design and selection. If and how GEDI addresses meshing will therefore probably impact the way this classification could be carried out. GEDI height data currently do not discriminate trees from buildings so the Global Forest Canopy Height service for example masks urban areas on the basis of a vegetation index and the gridded Global Human Settlement Layer (GHLS) built-up area dataset. Planned are more sophisticated approaches that would normally include a broader range of land cover masks (e.g., inland water, barren land, imperviousness, and wetland). Filtering data for built infrastructure and vegetation may be another approach. Similar methods are probably applicable for climate mitigation carbon accounting for project planning and certification and policy making at the non-forest project-level, where trees accompanying built infrastructure are an important biomass contributor (PDF).
A similar discussion can take place for project-level biodiversity accounting based on earth observation. So after making a carbon account for one of our vegetation-rich villages, Jura Mountains mapping may turn its project-scale ecology accounting towards biodiversity accounting, perhaps in terms of functional diversity (PDF).
3 January 2021