Sub-pixel Analysis Works Great with Landsat 8

Now that Landsat 8 data is available to all, we can get down to figuring out what we can do with it. If you are not familiar with the spectral capabilities of Landsat 8 data, I highly recommend you check out thes excellent posts by Charlie Loyd at MapBox: Putting Landsat 8's Bands to Work and Processing Landsat 8 Using Open-Source Tools.

With this spectral richness in mind, my colleague, Ben Kamphaus, has recently started a crusade to convince Landsat 8 users that they are not limited to traditional, discrete classifications that assign each pixel to one class of materials. Sub-pixel techniques that estimate the abundance of different materials within each pixel of an image have been around for decades now. They have been used successfully with Landsat data fora myriad of purposes, including detecting invasive vegetation species,monitoring impervious surfaces, estimating the abundance of urban vegetation,modeling forest structure, and mapping minerals. In fact, an argument can be made that sub-pixel analyses are best any time one is interested in materials that are frequently mixed with other materials at the resolution of the data. With Landsat’s 30 m resolution, this tends to be the case. Consequently, Landsat 8 and earlier Landsat data are perfect candidates for sub-pixel analyses.

I believe that an important roadblock to using sub-pixel techniques is simply that they are less understood than traditional classification methods. They do tend to involve more complicated mathematics,and they can require the user to make more decisions. And yet they provide major advantages, including the ability to find things that are smaller than a pixel. Moreover, there are some fairly easy-to-use, automated tools available to simplify the user’s experience while ensuring good results. In ENVI, a tool worth exploring is SMACC, which stands for Sequential Maximum Angle Convex Cone. SMACC is an unsupervised, iterative algorithm for finding and mapping end member spectra from spectral data. It was developed by Spectral Sciences Inc., and works beautifully with Landsat 8 data to find what’s in the scene and how separable it is from other surface materials.

What do you need to find within your Landsat 8 pixels?