Modelling canopy cover with satellite data

Great news: there’s a new canopy cover article online!

I’ve tested plenty of different methods for estimating canopy cover, but the test with satellite images that was supposed be part of my doctoral thesis was delayed and delayed. I guess that I was still busy with other things on the day when Janne sent me an excel file with the SPOT 5 reflectances from our Hyytiälä sample plots. Out of curiosity, I calculated the RSR spectral index from the reflectances and plotted the values against the lidar-based canopy cover estimates that I had. Surprisingly, it was a really beautiful exponential curve! The earlier results that I had seen from the boreal zone were not nearly as impressive. I think that the reason for the very good accuracy was that we used lidar-based canopy cover estimates instead of quickly measured, inaccurate field values.

 

So nice!

It still took more than a year until I finally started writing the paper last winter in Norway. I tried to keep things simple – one problem, one data set for modelling and another for validation, and two spectral indices to compare. After some reviews, the paper was accepted quicker than I thought. It is always nice when you can keep things simple, but unfortunately all of our problems are not as easy. 

Hyytiälä from satellite

Crown volume estimation

It’s been a long time since my previous post. Meanwhile I have been mostly working with non-canopy problems, especially lidar-based forest inventory. I’ve also had some teaching and supervision responsibilities, and a six-month visit to Norwegian University of Life Sciences, where I (for a change) had plenty of time for research. Thus I finally got out the first paper based on our 2011 field campaign: crown volume estimation using airborne lidar data.

The roots of this study go back to 2005 when I collected data for my master’s thesis at Metla’s Suonenjoki research station. Pola and Miina had employed three students to work there, and one of us, Sanna, made some curious measurements using an instrument called angle measurer (a.k.a. “miinaharava”).  It is a T-shaped stick that can be used to measure crown widths at different heights, enabling estimation of vertical canopy profiles and crown volume.  The data was needed for developing spectral reflectance models – some papers have been published in AFM and Silva Fennica.

The angle measurer 

Example of a crown profile.

Sanna’s work came back to my mind later when I learned what a colleague of mine at the UEF, Jari Vauhkonen, was studying. Jari worked with detecting individual tree crowns from airborne lidar data, and was the first person to test the alpha shape method in the prediction of tree attributes. Alpha shapes are a method that can be used to combine a set of 3D points into one geometrical shape, which is defined by the alpha parameter.  The volume of the shape can be calculated based on the triangulation of its individual points (=lidar echoes). Thus, when the echoes represent a tree crown, its volume can be estimated automatically. Our idea was to validate these estimates using the angle measurer.

Thus Juha and Laura measured the crown volumes of 89 trees during the 2011 field campaign. 77 of these trees were detected from the lidar data and used in the analysis. The echo segmentation phase was somewhat laborious, as my automated algorithm did not always delineate the tree perfectly and plenty of manual work was needed to remove this error source (and actually this had to be done several times because of some personal blunders and errors in the lidar data preprocessing). Based on the delineated echoes, Jari calculated the crown volumes and we simply plotted them against field-measured values. The results showed that the lidar-based volumes were clearly smaller than field values, mainly because there were not enough echoes from the lower part of the crown. Yet the results were better than those obtained using a general model for the crown dimensions  and assuming an ellipsoidal crown shape. Full paper can be read here.

This simple experiment is anyway a step into a direction that is very interesting to me – using features derived directly from the lidar point cloud instead of predictions based on forestry databases. Most of the existing theoretical forest models use traditional forest attributes such as tree density, height, and basal area, which are increasingly estimated using lidar, so why not directly use lidar point cloud features instead? One problem is that the lidar features can be sensitive to scanner settings. Nevertheless, in my opinion more this kind of investigations should be made in the near future.

Needle work

We have been lacking a spectral database on boreal tree species for a long time. Most data available have been measured for North American species or for ‘mats of needles’ i.e. not for single needles. Another problem has been that the previous spectral databases lack a detailed description of the structural and biochemical properties of the tree leaves and needles needed in many ecophysiological applications. 

 Collecting samples in the forest with 16-meter long scissors.

Finally, to fill the large gap in our knowledge, Petr et al. toiled in Hyytiälä last summer for several weeks measuring the optical properties of the most common (and nearly only..) tree species in Finland: Scots pine, Norway spruce and Silver birch. Measuring the reflectance and transmittance of single needles is really tough due to their small size and twisted shape. There are no commercial gear available and in-house solutions have to be developed for holding the small samples. The task was further complicated by our ambitious plans: we wanted (whenever possible) to measure separate the adaxial and abaxial sides of the foliage elements and both for shaded and sunlit crown positions. 

Yes, it does take a lot of patience and a peaceful lab environment to prepare the samples for the spectroradiometer (and other) measurements. Another challenge was acting very quickly: the needles are, after all, alive and their spectra may begin to change if they have been detached from the branch for a long time. Quick fingers (accustomed to needle work or playing a musical instrument) were definitely an asset in operating the tweezers. Bare fingers were not, of course, allowed to touch the samples and contaminate them.

 

Detaching young and very soft spruce needles for spectral measurements.

The hard work paid off, and we are now proud to present a carefully measured data set on the reflectance and transmittance spectra of needles and leaves of boreal tree species. And just a couple of weeks ago, the paper presenting the data was published! The data are now freely available and can be downloaded from the SPECCHIO database.

For more info, see Petr’s paper:

Lukeš, P., Stenberg, P., Rautiainen, M., Mõttus, M. & Vanhatalo, K.M. 2013. Optical properties of leaves and needles for boreal tree species in Europe. Remote Sensing Letters, 4(7): 667-676.