Native Canopy And Precision-Forestry Products
Purpose
This guide explains P79.7: how FEMIC derives a native-resolution canopy height model (CHM) from aligned DSM/DEM rasters, summarizes canopy height and cover, compares CHM evidence with VRI inventory fields, and ingests optional individual-tree products.
FEMIC does not implement a new individual-tree segmentation algorithm. Tree products are consumed from existing public sources where coverage and quality permit.
CHM From DSM And DEM
femic lidar canopy \
--dem data/lidar/lidarbc/dem/layer_5/2024/dem.tif \
--dsm data/lidar/lidarbc/dsm/layer_1/2024/dsm.tif \
--output-dir data/lidar/canopy \
--vertical-datum "Canadian Geodetic Vertical Datum of 2013"
The DSM and DEM must share exactly the same grid, CRS, transform, resolution, and vertical datum. FEMIC rejects misaligned or different-CRS inputs rather than resampling them.
The CHM is DSM - DEM. Negative heights are clamped to zero by default and
counted in QA. Use --no-clamp-negative only for diagnostics.
Outputs:
<output-dir>/
chm.tif
canopy_structure.json
canopy_qa.json
canopy_qa.csv
vri_canopy_comparison.csv # when --vri is supplied
Canopy Structure
The canopy QA report uses schema femic.lidar.canopy_qa.v1.
canopy_structure.json records:
valid/nodata pixels and area;
min, mean, median, p90, p95, p99, and max height;
cover percentages at 2, 5, 10, and 20 m.
The CHM and structure are native-grid products. No reprojection, resampling, smoothing, or threshold promotion occurs.
VRI Comparison And Uncertainty
Use --vri to compare polygon-level CHM evidence with VRI height and cover:
femic lidar canopy \
--dem data/lidar/dem.tif \
--dsm data/lidar/dsm.tif \
--output-dir data/lidar/canopy \
--vri data/input/tfl_6/vri_2025_r1_poly_tfl6.gpkg \
--vri-height-field proj_height_1 \
--vri-cover-field crown_closure \
--vri-id-field feature_id
The comparison uses CHM p95 as the height metric and CHM cover at >=2 m as
the cover metric. It records per-polygon rows, bias, MAE, and RMSE.
This is uncertainty evidence, not a field-accuracy claim. VRI height and crown closure are inventory estimates, and CHM cover at a 2 m threshold is not equivalent to VRI crown closure.
Individual Tree Products
Ingest an existing point product:
femic lidar trees \
--product data/lidar/trees/lidarbc_trees.gpkg \
--output-dir data/lidar/trees
Required fields are tree_id and height_m. Optional fields are
crown_radius_m and species. The command writes a JSON completeness
summary using schema femic.lidar.individual_tree.v1 with duplicate IDs,
missing heights, height/crown distributions, and species counts.
Block summaries:
femic lidar trees \
--product data/lidar/trees/lidarbc_trees.gpkg \
--blocks data/input/analysis_blocks.gpkg \
--block-id-field block_id \
--output-dir data/lidar/trees
Tree points outside all blocks are excluded and reported. Tree and block CRS must already match.
TFL 6 Pilot Evidence
A 1 km TFL 6 AOI used LidarBC 1 m DSM/DEM tiles:
CHM mean 8.62 m, median 7.71 m, p90 18.96 m, p95 21.78 m, max 55.42 m.
Cover: 70.94% at >=2 m, 60.09% at >=5 m, 41.03% at >=10 m, 7.84% at >=20 m.
95 negative pixels were clamped to zero.
VRI comparison across 7 polygons: height bias -2.93 m, MAE 4.08 m, RMSE 6.16 m; cover bias +23.87 percentage points, MAE 39.47 points.
The LidarBC feature service has no CHM/canopy layer and no individual-tree layer. CHM was derived; individual-tree ingestion remains optional.
The cover disagreement is large enough that CHM cover at 2 m must not be treated as VRI crown-closure equivalence. Height p95 is more comparable but still variable.
Limitations
No field-accuracy claim is made.
CHM depends on DSM/DEM vintage and vertical-datum agreement.
The feature service does not currently publish an individual-tree layer.
No new segmentation algorithm is implemented.
Point-cloud processing is optional and GPU is not required.