Downstream Analysis¶
samstars writes individual crown polygons.
Those polygons can be used as crown-level units for later analysis, including species classification, biomass estimation, crown metrics, and change analysis.
Species classification is a separate downstream step using these crown geometries.
Crown Output¶
The main segmentation output is:
1 | |
The GeoPackage layer is named crowns.
It contains one feature per predicted crown:
| Field | Description |
|---|---|
id |
Crown identifier within the output file. |
source |
Source tile or proposal record used to create the crown. |
label |
Internal instance label from polygon refinement. |
score |
Refinement confidence score when available. |
geometry |
Crown polygon in the output CRS. |
If you combine outputs from multiple runs, add a run identifier before merging files so crown identifiers remain unambiguous.
Species Classification¶
A common downstream workflow is:
- Run
samstars.segment(...)to producecrowns.gpkg. - Extract crown-level predictors from imagery, CHM, vegetation indices, texture layers, or other rasters.
- Train or apply a species classifier outside
samstars, for example with XGBoost, scikit-learn, or another modeling library. - Join predicted species labels and confidence scores back to the crown polygons.
Use a table with one row per crown and a stable key, usually id for one output file or run_id plus id for merged runs.
Join Species Predictions¶
After classification, read the crown GeoPackage into a GeoPandas GeoDataFrame and join a prediction table to it.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 | |
Recommended downstream fields:
| Field | Description |
|---|---|
species |
Predicted species name, code, or class label. |
species_score |
Classifier confidence or probability for the assigned class. |
species_model |
Optional classifier name or version. |
species_source |
Optional source of the species label, such as model, field label, or manual edit. |
Feature Tables¶
Feature extraction is project-specific. The crown polygons can be used with any raster or vector processing library that supports polygon summaries.
Typical crown-level predictors include:
- image-band means, medians, and standard deviations
- vegetation indices summarized inside each crown
- CHM height summaries, such as maximum, mean, and percentiles
- texture metrics computed from imagery or canopy-height rasters
- crown geometry metrics, such as area and perimeter
Quality Checks¶
Before using crown polygons for classification or other analyses, check:
- the output CRS matches the rasters used for feature extraction
- very small or low-confidence crowns are handled consistently
- training and prediction areas are kept separate during model assessment
- merged outputs have unique crown identifiers
- species labels are joined to the correct segmentation run