Understand Vibrant Planet’s baseline data and models that help you get to implementation faster
To understand your landscape’s potential risks and opportunities to plan for resilience, you need a detailed view of your landscape. One of the first and most essential steps in this process is gathering three-dimensional (3D) structure data of your vegetation health and density. Vibrant Planet uses dozens of publicly-available and widely-used datasets to build this in-depth view of your landscape, including LiDAR, and data from the USDA, National Aerial Imagery Program (NAIP), LANDFIRE, USGS, and USFS.
When possible, our foundational 3D vegetation map uses recent LiDAR. LiDAR is one of the most powerful remote sensing datasets to gather 3D vegetation detail, but is often expensive
($.50-$1 per acre) and can be immediately out of date as soon as vegetation changes – whether from disturbance events such as wildfire, or treatments like mechanical thinning. Because of these limitations, Vibrant Planet uses machine learning algorithms, trained on LiDAR, to build a Synthetic Canopy Height (Synthetic CHM) that supplements additional high-quality and real-timelandscape assessment data. Our Synthetic CHM applies high spatial and temporal resolution imagery to calculate individual trees and their height, as well as other helpful metrics including health, wood product value, wildlife habitats, and above ground biomass.
| + | + | = |
 |
 |
 |
| Satellite imagery inputted to algorithm | Resulting canopy height data produced by Synthetic CHM |
Current and specific view of vegetation |
This allows you to annually (or more frequently) reassess your landscape at the tree and house-level. Use this information to help speed up the arduous and ongoing process of understanding your landscape so you can move on to analyzing the best areas for a risk reduction or ecological enhancement treatments.
Science Corner: Synthetic CHM Deep-Dive
Vibrant Planet’s main source of forest structure training data comes from USGS 3DEP lidar data that is flown across the US (U.S. Geological Survey, n.d.), which has regional coverage at disparate temporal resolution. The point clouds all must follow the USGS 3DEP standards (NGP Standards
and Specifications, n.d.). Where lidar data is unavailable, Vibrant Planet has built machine learning models (UNet and ZoeDepth) to predict a canopy height model, percent canopy cover above 2 m, and percent canopy cover between 2-8 m products from aerial imagery.
For Vibrant Planet’s UNet model, National Agriculture Imagery Program (NAIP) is the imagery dataset used along with other spatial data that include topographic information (ex: slope, aspect, curvature, elevation) and satellite multi-spectral images (Sentinel-2-L2A). For prediction, the most recent available NAIP and satellite imagery tiles are used. For model training however, it is important that the date of the NAIP imagery is reasonably near to the date of the lidar dataset acquisition. Multiple datasets are used as input to the deep learning model:
- National Agriculture Imagery Program (bands RGB-IR)
- 3DEP DEM 10 meter product (elevation, slope, aspect, curvature)
- Sentinel-2 Level-2A 2022 monthly median product (all bands)
Now that you understand one component of the underlying landscape data to help you evaluate risk, in the next section, you’ll learn how we segment the landscape based on similar qualities.