Dakota Agros

Research GIS workbench

XYZ alignment & georeferencing Plot bloom notes & model comparisonOpen workspace

1 · Project & layers

Project IDFlight IDImport intoInput GeoJSON CRSImport GeoJSON

Use WGS84 unless the file explicitly uses another CRS. Imports are copied; source files are never edited.

Active selection layer

2 · Attribute selection & review

Field

Selection method

Review selected objects

Map clicks select features in the active layer. Shift-click adds/removes a feature. No selection automatically deletes data.

3 · Select by location & buffers

Reference layerRelationshipTest geometryDistance (m)Buffer distance (m; negative shrinks)

Click at least three map vertices, then Finish. Drawn zones also support holdout-footprint inspection.

4 · Measurements & distributions

Object area uses calibrated source pixel counts or polygon geometry. A bounding box is never treated as the object mask. Compactness is available only for polygons.

Histogram fieldOptional local raster for zonal statisticsRaster band (1-based)

Statistics describe each saved geometry; for box candidates they describe the box, not an inferred canopy mask. Raster values retain their source units.

5 · Proximity review

Check

Pairs are flags for review. Nearby objects may be distinct plants. No automatic merging or biological inference.

6 · Reviewed plot reports

Include accepted objects onlyJoin plot metadata JSON

Metadata is an array of rows with unique plot_id plus treatment, replicate and optional row. Objects matching multiple plots are flagged and excluded; plot edges are included when unambiguous.

7 · Saved recipes & edit history

Open recipe Allow replay on a different dataset

Recipes select only. Review before editing. A source fingerprint and reference-layer fingerprint record the original context.

8 · Spatial validation & test lock

Assign selected plots toCheckpoint ancestry / provenance note

The lock protects test labels and layout from this workbench's edits. Geometry checks cannot establish checkpoint independence. A fork gets a new project ID and does not overwrite the locked export.

9 · Flight comparison

Earlier report JSONLater report JSONFlag absolute count change (%) I checked alignment against the same plot layout

Reports must share project, layout, metadata and review policy. Changes can reflect detection or imagery differences; this does not track individual plants.

10 · Tagged-chip training audit

Import tagged chips as Objects or Chip footprints. Histograms, attribute summaries and spatial selection work on the active layer. These tools prioritize review; they do not manufacture ground truth.

Group summary byLoad map feature binding JSONNative chip size (pixels)

Uses the same feature builder as Heliovision training and detection. Direct-point models require matching point-cloud inputs and 128-pixel chips. At most 200 chips per audit.

I inspected the entire selected chips and reviewed every target object, including missed objects.

Queue uses review status, confidence, coverage and target rarity, with spatial diversity. Only explicit training members qualify. Verify all targets in a chip before treating it as a hard negative. Export selected attributes or GeoJSON to retain review suggestions.

11 · Point-cloud plot height

Height notes use original point elevations, never canopy raster averages. Load plot polygons, then provide a classified LAS/LAZ with class-2 ground in a projected meter CRS. Automatic ground classification produces provisional measurements.

Classified point cloud Ground estimation

The alley-ground option assumes approximately planar trial terrain. It filters ground support without changing cloud labels or canopy elevations. For triangle methods, the distance limit applies to every triangle vertex; the 21_811 candidate used 8 m to span crop beds. Settings require review for each field.

New output folder Point Z units to meters Maximum ground-support distance (m) Mode bin width (m) Canopy-top XY bin spacing (m) Inset from plot edges (m)

Reports include mean, median, binned mode, min, max, P95 and P99. Point summaries include all retained non-ground points; top summaries retain the highest actual point per occupied XY bin. Neither is an average of individually measured plants. Unsupported and negative heights are counted separately. Large clouds can take several minutes.

Objects: green · plots: blue · zones: orange · chip footprints: gray · selected: yellow · test plots: purple. Wheel to zoom; drag to pan.

Attributes