Public Lab Research note

Usability of Public Lab's NDVI and Kite/Balloon Mapping for Remote Sensing in Academics/Research

by Roolark | January 01, 2014 02:01 01 Jan 02:01 | #9919 | #9919

What I want to do

Traditional remote sensing of vegetation and snow, using Landsat and MODIS imagery, is useful for small scale, regional areas. Higher resolution satellite imagery from Spot / GeoEye is expensive, and the temporal resolution is limited to (mostly) set schedules.

Initially, it was my plan to try and determine the changes in vegetation and vegetation health from ski valleys in areas that receive only natural snow, versus areas that receive fairly regular, annual artificial snow. The problem is, with Landsat imagery, I am unable to get fine enough spatial resolution. This has led me to Public Labs, and while I no longer have the time to work on my initial proposal studying artificial snow and vegetation, I do have plenty of time to work on determining whether Balloon and Kite imagery, using a canon A2200 or similar camera, modified for NIR, can produce higher spatial resolution imagery, with a good enough spectral accuracy, to compare with Landsat, for use in future earth research.

Data Requirements: Landsat 8 imagery of areas with differing vegetation, and hopefully additional imagery with snow.

Balloon / Kite imagery, using the Infragram filter from Public Labs, taken in the same areas covered by the Landsat 8 imagery.

Analysis Tools: QGIS + semi-automatic classification plugin (free, open-source software) and ERDAS Imagine (commercial, industry/academic standard).

Process: 1. With QGIS and the classification plugin, perform USGS level 1 and level 2 classifications of the Landsat 8 imagery. Different classification algorithms will be used to provide comparisons for accuracy. The final output of this process will include the spectral signatures of the classified land covers, classified raster images with matching vector layers, and an accuracy/error assessment of the classifications.

  1. Using QGIS, I will perform the same classification process on the Balloon / Kite imagery. Final outputs of this process will be the same as with the Landsat imagery.

  2. A single reference vector file, with polygons known to be properly classified, will be used for both the Landsat and the Balloon / Kite data, to determine accuracy/error.

  3. Repeat steps 1-3 using ERDAS Imagine

Expected outcome: The modified infragram camera from the balloon and kites should produce higher spatial resolution imagery and lower spectral resolution imagery than the Landsat 8 data sets. The higher spatial resolution should make it possible to discern differences in vegetation and snow types with better accuracy than the 30 meter Landsat data. The lower spectral resolution could or could not impact the ability to use semi automatic classifications with acceptable (95% or better RMS error) error. If it is found that classification error is acceptable using the balloon/kite imagery and an infragram camera, then this could be a solution for conducting research in an academic setting where results need to be consistent with current standards and costlier tools.

My attempt and results

Questions and next steps


Hi Roolark -

This sound like an interesting project. I hope you keep us posted with your progress. I think there is a lot of value in pairing Landsat imagery with DIY aerial photography. There is no doubt that DIY aerial cameras can be used in scientific research if rigorous methods are used. Effectively what you are trying to do is downscale or disaggregate the information in the Landsat signal to get information at a finer scale. If you accept that the Landsat data is of high quality then your task is to determine how the energy in a pixel is distributed within a pixel. As you mention, an important step in this process is to accurately align the aerial images so they are coincident with the Landsat image. In addition to using automated classification on the photos you might want to try visual/manual interpretation.

In my experience it is still quite cumbersome to use automated classification methods with uncontrolled aerial photos. If you want to try automated classification it will be best if you can constrain as many image acquisition variables (flying height, camera orientation, time of day/solar angle, exposure settings...) as possible. The low spectral resolution of the PLOTS system is (nearly) the least of the limitations.

All the best,


Reply to this comment...


Thank you for your interest and input. There are some details I'm still working on putting together, to include in the research note. I haven't been able to obtain reasonable imagery yet because the wind has been uncharacteristically calm here in Albuquerque (for kites) and I don't yet have a balloon.

I plan to collect my aerial imagery around the same times and dates as the Landsat 8 imagery I will be using. I will also be performing the appropriate dark-object-subtraction atmospheric correction on the Landsat imagery.

The reference vector point/polygon classification file I plan to use, for comparison to both the Landsat and kite/balloon imagery will consist of both heads-up digitizing AND in situ observations. With the field observations, I may be able to borrow our department's field spectrometer too, which would be very useful in measuring the spectral accuracy of the NDVI (and NDSI) infragram on vegetation and snow for features large enough to be discernible. Alternatively, I could use a portable (offline) implementation of the PLOTS desktop spectrometer for this purpose... although that may require an accuracy assessment of that product, in order to use it for accuracy assessments of the others?

I have a reasonably accurate GPS device that I will be attaching to my imaging rig, and I can coordinate the seconds between measurements to closely match the CHDK image intervals, in order to discern location and elevation of the rig. Unfortunately, I don't know a really great way to determine exactly what the flight altitude of each picture will be, but the GPS should give me at least a rough estimation.

If you have any other suggestions, please let me know, I'm all ears. This is for my departmental honor's thesis, and as I mentioned in my research note, my ultimate goal would be to use these tools (if practicable) for conducting further environmental research for graduate school.

Is this a question? Click here to post it to the Questions page.

Reply to this comment...

Collecting you imagery so it coincides with Landsat 8 imagery is a good idea. MODIS acquires images at roughly the same time as Landsat so that's another good reference source for calibrating the Landsat and aerial images. Using a field spectrometer to calibrate you camera for reflectance measurements is also a good idea. I'm not sure how you could use the PLOTS spectrometer for that since it only provides relative measurements but maybe someone else has pointers.

For the camera height the absolute height isn't as important as taking all of your photos from roughly the same altitudes above ground level. Geotagging your photos with a GPS is a good idea though. If you take a photo of your GPS receivers clock on the screen you can use a program like gPicSync ( to geotag your photos.

When you go out and take photos I expect you'll make some new discoveries and will have additional questions. Feel free to post your questions to this forum or contact me via email.

Reply to this comment...

Login to comment.