Cross-Sensor Calibration

Turn raw NEON hyperspectral flight lines into corrected ENVI cubes, sensor-matched products, tidy Parquet, and a QA panel—in one restart-safe command.

Run your first tile (3 steps)

# 1) Create an output base
BASE=output_demo && mkdir -p "$BASE"

# 2) Process one flight line (replace IDs as needed)
cscal-pipeline \
  --base-folder "$BASE" \
  --site-code NIWO \
  --year-month 2023-08 \
  --product-code DP1.30006.001 \
  --flight-lines NEON_D13_NIWO_DP1_L020-1_20230815_directional_reflectance \
  --max-workers 2 \
  --engine thread  # swap for "ray" once the optional dependency is installed

# 3) Open the QA image (and PDF report)
open "$BASE/NEON_D13_NIWO_DP1_L020-1_20230815_directional_reflectance/NEON_D13_NIWO_DP1_L020-1_20230815_directional_reflectance_qa.png"
open "$BASE/NEON_D13_NIWO_DP1_L020-1_20230815_directional_reflectance/NEON_D13_NIWO_DP1_L020-1_20230815_directional_reflectance_qa.pdf"

New here? See the Quickstart.

Something broke? Jump to Troubleshooting.

---

Why teams use cross-sensor-cal

• Restart-safe orchestration – rerun the same command after a crash; completed stages are skipped automatically.
• Consistent artifacts – every tile produces ENVI cubes, tidy Parquet, and QA outputs named predictably.
• Built-in QA and calibration – each run emits a single-page QA panel, a multi-page QA PDF, and a JSON metrics file. Brightness adjustments between Landsat and MicaSense are applied from a versioned JSON and recorded across every QA artifact for full traceability.
• Python & CLI parity – drive the pipeline from scripts or automate it in schedulers with the same arguments.

Install in minutes

Choose a clean virtual environment (see the detailed Environment page for Conda tips and GDAL notes).

venv + PyPIFrom source

python -m venv .venv && source .venv/bin/activate
pip install -U pip
pip install cross-sensor-cal

git clone https://github.com/earthlab/cross-sensor-cal.git
cd cross-sensor-cal
python -m venv .venv && source .venv/bin/activate
pip install -U pip
pip install -e .[dev]

Understand the pipeline

1. Download NEON HDF5 tiles.
2. Export ENVI cubes and apply topographic + BRDF correction.
3. Convolve to other sensors, flatten to Parquet, and finish with QA outputs (PNG, PDF, JSON).

See the Pipeline Stages overview for purpose, inputs, outputs, and pitfalls at every step.

Next steps

Goal	Start with
Rerun the full flow on multiple tiles	CLI & examples
Inspect Parquet outputs in notebooks	Parquet preview
Tune configuration for HPC or CI	Reference → Configuration
Validate artifacts and schemas	Reference → Schemas & Reference → Validation
Extend to a new sensor or reader	Reference → Extending
Something looks off	Troubleshooting

---

Quality Assurance Overview

All processed tiles automatically produce a QA panel, a multi-page QA PDF, and JSON metrics.

✅ Good: reflectance within [0, 1.2], low ΔReflectance, high mask coverage.
⚠️ Needs Review: 1–2 metrics outside thresholds.
❌ Fail: large brightness shifts, missing wavelengths, or high spectral error.

Each QA artifact lists any Landsat→MicaSense brightness coefficients that were applied so you can audit sensor harmonization settings quickly.

Read full QA interpretation → and the JSON metrics reference.

Release highlights

• Per-flightline master table written as <prefix>_merged_pixel_extraction.parquet
• QA panel <prefix>_qa.png is emitted after the merge during full runs