Skip to content

Configuration

When do I need this? When you want to override defaults for correction/convolution or run headless in CI. Includes examples and links to producing/consuming stages.

Purpose

Explain optional overrides that change how Pipeline Stages behave—especially Stage 3 (correction), Stage 4 (convolution), and Stage 6 (merge).

Inputs

  • YAML or JSON files loaded via --config
  • Environment variables read by orchestrators
  • Inline CLI overrides such as --sensor or --max-workers

Outputs

A normalized configuration object passed into each stage; affects artifact naming in Outputs.

Brightness configuration

Brightness adjustments between sensor systems (e.g., Landsat→MicaSense) are defined via small JSON files shipped with the package:

  • Location: cross_sensor_cal/data/brightness/*.json
  • Loader: cross_sensor_cal.load_brightness_coefficients(system_pair)

Example (landsat_to_micasense.json):

{
  "system_pair": "landsat_to_micasense",
  "unit": "percent",
  "bands": {
    "1": -7.40,
    "2": -2.75,
    "3": -6.94,
    "4": -10.12,
    "5": -6.65,
    "6": -2.74,
    "7": -1.11
  }
}

These coefficients are applied to Landsat convolution products as: L_adj = L_raw * (1 + coeff / 100) and are surfaced in:

  • The brightness summary table (per-band values).
  • The QA JSON (brightness_coefficients).
  • The multi-page QA PDF (Page 3).

How to change coefficients

  1. Create a new JSON file under cross_sensor_cal/data/brightness/.
  2. Set "system_pair" and a "bands" mapping of 1-based band indices to percent adjustments.
  3. Call load_brightness_coefficients("<your_system_pair>") from your pipeline extension or configuration.

Run it

cscal-pipeline --base-folder out --config configs/niwo_gpu.yaml \
  --site-code NIWO --year-month 2023-08 \
  --flight-lines NEON_D13_NIWO_DP1_L020-1_20230815_directional_reflectance

from cross_sensor_cal.config import load_config

cfg = load_config("configs/niwo_gpu.yaml")
print(cfg.pipeline.stages["convolution"].sensors)

Pitfalls

  • Missing sensor definitions lead to empty Stage 4 outputs—double-check sensors: blocks.
  • Keep path references relative to the working directory used during Stage 2–7 execution.
  • For CI, pin random seeds where applicable to keep QA metrics stable.