Reflectance Calibration with Phantom 4M Multispectral data

DJI Phantom 4M(ultispectral) camera system (channels 1-5 and RGB camera)

There was some confusion in the past on how to process the Phantom 4M Multispectral data to “reflectance” level. Here a short summary for Agisoft Metashape 1.7.2 with a 50% reflectance panel from Micasense. In general: the P4M sun illu. sensor works for normalizing the within flight campaign variations while the panel reflectance measurement is used to make different multispectral data measurements from different campaigns comparable by converting DNs to reflectance. Since vegetation surfaces do not fully reflect light in lambertian style this does not always fully work, but we can get close when the illumination directions of the different flight campaigns are comparable and when relative change is needed. Although … reflectance calculation is usually done to get a value for absolute comparison of spectral information with reference spectra or other datasets.

  1. Import all Geotiff folders with the multispectral tiff files. Choose Add Photos as “Multi-camera system”.
  2. Locate the reflectance panel tiffs using “Tools-> Calibrate Reflectance”. Locate the panel files automatically or move the panel tiffs manually into an image folder and rename the folder to “cal-img”. For each channel insert the right reflectance values (0.518 f.e.) that are specific for your panel and your band setup (Micasense Rededge band specs are close to P4M camera specs but not identical so using the RE values with a Micasense panel should work somehow but here more accurate values might be needed). In general the sun sensor can be deactivated when the flight was fully sunny, but should be activated when illum. has changed. For longer flights with multiple battery changes I usually keep this ticked (but panel measurements between flights is suggested). Ok the “reflectance calibration” and
  3. check that masking of the calibration area of your panels is selected (the rest of the image area within the tiffs should be masked out) in every channel (change channel under “Tools->Set primary channel” to check the mask for all channels).
  4. Increase brightness level (Tools-> SetBrightness) to about 800% (only for visualization).
  5. Set “Tools->Primary Channel” to NIR or RE if you work with vegetation surfaces.
  6. proceed with workflow processing:
    1. Align all tiffs using your preferred setup (High/80000/8000/AdaptiveCameramodelFit no – if camera meta data should be used – see preferences/),
    2. Build Dense Cloud (Ultra High, no filtering, can be reduced to depth maps only if DEM is to be derived from depth maps),
    3. Build DEM (quality: ultra high, UTMxy N ETRS89 …, build from dense cloud),
    4. Build Ortho Mosaic
  7. Check that “Tools->Camera Calibration” shows a tick at “Bands-> normalize band sensitivity”.
  8. Under “Tools->Raster Transform” do a normalization of 16uint to 0-1 reflectance using “Input Band” B1 -> B1/32768 and repeat for all other channels. This is important for the back-scaling into unscaled reflectance. Agisoft scales 100% reflectance to the middle of the 16Bit space: 65536/2 = 32768 and this follows basically the Micasense recommendation and processing guidelines in Atlas Cloud .
  9. Export the Orthomosaik using “Export to Tiff” with the “index value” option under “Raster transform”! Now check the TIF file – it should show 32R Bit values between 0 and 1 for all channels.

Overall this seems to work. A comparison with spectrometer values to follow. However there are some indications from my initial testing in March 2021 that repeated flights under the same conditions do not lead to comparable reflectance values. Here some more testing is clearly needed with reference panels in the field. I measured up to 20% reflectance change after remapping a few days later and this is not related to vegetation change. Its difficult to assess where these changes came from. There might be some illumination changes at work here since some TIFs where captured under overcast conditions and the two flight campaigns did not take place at the same time of the day. I removed these from the TOP mosaicing and reflectance calibration process. NDVI calculations however seem to be very comparable and show expected (locally with ground reference confirmed) regreening change effects. Differences between the NDVI calculations on likely unchanged surfaces never exceed the 0.02 range. Some comments on the Agisoft Support pages also hint towards the vignetting correction in Agisoft Metashape that might have an effect on the reflectance calibration. More test flights under comparable conditions are needed.

Parts of this information was summarized by Agisoft here: . This tutorial describes processing for MicaSense RedEdge MX data but it was also suggested by Agisoft Support for P4M data.

Fig.: NIR Tif data of the 50% calibration target (Micasense Panel).
Fig. : P4M true ortho photo mosaic (TOP) of the March 2021 leaf-off beech stand status with early bloomer vegetation signatures (RGB: NIR-red-green).
Fig.: Reflectance values for P4M channel 1-5 for March (wild garlic – Allium Ursinum dominated understorey vegetation signature).

Leave a Reply

Your email address will not be published. Required fields are marked *