GEDI’s Elevation and Relative Height Metrics Explained
In this section:
- Descriptions of ground elevation, relative height metrics, and canopy top height datasets.
- Review of sensor parameters and environmental conditions that may influence performance of these metrics.
Watch the visualization of vegetation structure data
Source: Forest height visualization from NASA Scientific Visualization Studio.
The raw waveforms are geolocated and positioned relative to the Earth’s ellipsoid based on GEDI’s orbital positioning—this creates the L1B waveform geolocated to the Earth’s surface. At this stage, the waveforms are not yet processed to identify ground or canopy height. The data then undergo processing to determine ground and vertical profile locations– the results are found in the L2A product. The processing algorithms determine the ranging points for the lowest and highest returns within each waveform.
Several of the algorithm setting groups defined in previous sections, each have distinct waveform processing techniques for determining the lowest detected mode, highest return, and waveform mode detection, and therefore may result in different ground elevation, relative height, and top canopy heights.
especially if there are complex amplitude values or mode widths to resolve (Hofton et l., 2019; Fayad et al., 2020). Other sensor parameters, such as beam type, sensitivity, the extracted signal-to-noise ratio along with the inherent capabilities of the resolution of the laser pulse can affect the ability to resolve the ground elevation from other energy returns.
Factors that may influence the accuracy of the elevation metrics in addition to the sensor and waveform interpretation parameters may vary across surfaces. Geolocation errors can particularly affect the measurement, in addition to complex topographies. Dense vegetation structures and thick canopy cover, or atmospheric interferences like clouds can influence the waveforms ability to completely penetrate the surface feature and therefore reach the ground.
Ground Elevation
The lowest mode (`elev_lowestmode`), or the last detected peak, of the waveform is considered to be the ground detection (Dubayah et al., 2020). The geolocation of this return, which involves assigning the latitude, longitude, and elevation is interpolated based on the offset from the start of the received waveform.
GEDI offers the unique ability to estimate the ground elevation under vegetated landscapes. Mapping elevation within forests can provide crucial insight to ecosystem zones and climatic conditions, habitat extent, and vegetation classification. Elevation is also commonly used in general land cover maps, and can be used to map urban structures (Ma et al., 2024). Several studies have explored how well GEDI’s ground elevation captures inland waterbody levels and to detect floods, which could be crucial for disaster risk management (Fayad et al., 2020; Ma et al, 2024; Urbazaev et al., 2022).
Source: Vegetation structure tutorial figure from this training demonstrating plotting the corresponding TanDEM-x 90m InSAR elevation against GEDI’s ground and canopy top returns. (RH100 height is excluded).
Table 1: Several representations of the ground elevation are included in the GEDI data product, whether directly derived from GEDI or as an ancillary dataset. RH0 can also represent the ground elevation. Each GEDI shot includes datasets for the two DEMs from TanDEM-X 90m, or SRTM 30m elevation data.
| `elev_lowestmode` | m | elevation of center of lowest mode relative to reference ellipsoid |
|---|---|---|
| rh0 | m | Relative height metric at 0% cumulative energy returned |
| `digital_elevation_model` | m | TanDEM-X elevation at GEDI footprint location |
| `digital_elevation_model_srtm` | m | SRTM elevation at GEDI footprint location |
Relative Height Metrics (RH0-100)
Relative height metrics provide a quantitative meaning of how much of the laser energy has been returned below a certain height above the ground. Key to understanding these metrics, is that while the value is representing a height relative to the estimated ground and maximum height (canopy height), it is relative to the interpretation of how much of the waveform energy was returned at that height. The 100 intervals capture the cumulative returns at every 1% interval, where every 1% added, is representing the height at which that cumulated percent of the energy corresponds to. At RH0, we would expect 0% of the energy to have been returned from this elevation, since it represents the ground. RH100 represents the height above the ground below which a given percentage (100%) of cumulative waveform energy has been returned, presumably the maximum height of the surface. The rest of the RH metrics are percentiles of returned waveform energy, providing insight into the different canopy layers. RH50 would be the height of median energy returned. Negative RH values can occur, especially in sparse vegetation, which indicates that a significant portion of the waveform energy was reflected from below the fitted (selected) ground elevation for that waveform (Li et al., 2024).
Table 2: RH metrics as defined in the L2A Data Dictionary. The RH metrics range from RH0-RH100.
| `rh` | m | Relative height metrics at 1 % interval | | :—- | :—- | :—- |
Source: The GEDI Mission.
RH90, RH95, and RH98 can also be used to represent the upper canopy heights and are frequently used given their high correlations to observed upper canopy heights (Lahssini et al., 2022). Both RH100 and canopy top height are often replaced with a proxy for maximum canopy height–RH98. This is recommended because studies have shown that RH98, in practice, is more reliable and less noisy (affected by outliers or other conditions), than the highest return value and RH100 (Lang et al., 2022; Ngo et al., 2023). Depending on the context, singular RH metrics (like RH0 or RH95, 98, or 100) can be used for ground and max surface height mapping (Elliott et al., 2024; Islam et al., 2024). Other examples include all 100 metrics, every 5th, every 10th, or a combination of selected RH variables, like those determined to be the “most important” in Machine Learning modelling (Tommaso et al., 2023; Hoffren et al., 2023; Li et al., 2024). Aggregating RH metrics statistically or creating ratios is another option (Cooley et al., 2024).
Canopy Top Height
The calculated canopy top height comes from the elevation of the highest return or `elev_highestreturn`. Canopy height is computed by subtracting the elevation of the highest detected return from the elevation of the lowest mode, or ground elevation (Dubayah et al., 2020). The canopy top height calculation is essentially the same as RH100, since the representation of the height difference between the elevations of the lowest mode and highest return, since the highest return would be expected to represent the height below which 100% of the energy in the waveform was returned.
Table 3: Representations of the top canopy height included in GEDI products.
| `elev_highestreturn` | m | elevation of highest detected return relative to reference |
|---|---|---|
| rh100 | m | Relative height metric at 100% cumulative energy returned |
Other GEDI product levels relating to elevation and height
Besides the L2A footprint level elevation and relative height metrics, elevations are also provided as statistically gridded GeoTIFFs for the L3 Gridded Land Surface Metrics and fused product the Global Vegetation Height Metrics from GEDI and ICESat-2.
Source: An example land cover map of Liberia using tree height information from GEDI to improve annual land cover maps. NASA, alongside Conservation International and the Liberian government through the Environmental Protection Agency generated these improved maps with biodiversity information and field studies to detect changes in ecosystem types over time to inform sustainable planning. (NASA Visualization Studio).