dedal-calibration: reflectance calibration of DJI Mavic 3M multispectral images

dedal-calibration converts DJI Mavic 3M multispectral images (green, red, red edge, near-infrared) to reflectance, so values can be compared between images, flights, dates and fields. It is the calibration used by DEDAL (Distributed Ecosystem for Drone Analytics in Life-sciences, https://dedal.jic.ac.uk), the John Innes Centre's platform for drone-based crop phenotyping, both in its web platform and in its cluster pipeline.

Each image is corrected for dark level, vignetting, exposure and gain from its own metadata and normalised by the drone's sunlight-sensor reading; a reflectance panel photographed on the day sets the absolute scale per band. The panel and the panel shot are found automatically (no QR code or template), and readings that are saturated, non-uniform or outliers are rejected. Calibrated images keep all their original metadata and are ready for OpenDroneMap with its radiometric calibration switched off. Every run writes a report with the panel used, the factor per band, checks against DJI's factory calibration, the camera's typical factors, the sun elevation and the panel-shot timing, and an outcome: accepted, flagged (with reasons) or not calibrated. Panels are described in files with their certificate details; detection settings and decision thresholds can be changed per run, and are then recorded and flagged.

Validation (193 flights over 17 fields, 2024–2026): a second panel shot of the same day is predicted with a median error of 3.3–4.2% per band (1,241 pairs on 89 days), about ten times better than without the sunlight sensor. Field reflectance through a season was about twice as steady as with the established Pix4D workflow, and plot NDVI agreed with Pix4D at r = 0.91 (median).

Method: https://dedal.jic.ac.uk/calibration/. Contact: dedal@jic.ac.uk. DJI Mavic 3M only.

• Is supplement to: DOI 10.5281/zenodo.15846333 
• Is documented by: URL https://dedal.jic.ac.uk/calibration/