Thanks to all of you who tuned in to my Using Free Government Data and Remote Sensing to Create a More Powerful GIS webinar last week! We had a great turnout and many good questions. One topic that I briefly covered in the webinar had to do with impervious surfaces and how to calculate that using NAIP data. Some of the attendees asked some questions about this topic, including some more detail about what rules I used to calculate it using the and opportunities for calculating impervious surfaces using other types of data as well.
As a refresher, impervious surfaces are paved or hardened surfaces that do not allow water to pass through, and also refers to the general inability of a surface to allow water to percolate through. Impervious surfaces are mainly artificial structures–such as pavements (roads, sidewalks, driveways and parking lots) that are covered by impenetrable materials such as asphalt, concrete, brick, and stone–and also rooftops fall into this category. Soils compacted by urban development are also highly impervious.
In ENVI, one of the easiest methods to identify impervious surfaces is through the Feature Extraction module, which uses an object based image analysis method to find and extract feature from an image. ENVI uses all spectral, texture, and spatial values in the image to find contiguous regions of similar values to group into segments. I have access to all of the attributes in the image – for every segment outline, ENVI calculates spectral, textural, and spatial attributes that I can view, threshold against, and set parameters for to build a set of rules that define my feature or features. In my example I selected the average of band 3 attribute, which is a brightness threshold, as most of the impervious surfaces in the image I was using were bright white. I also used the texture mean attribute, as the texture or roughness of the roads, driveways, and roofs is going to be very different from the texture of the surrounding vegetation and fields. Finally, I added an area parameter, to rule out perhaps any very small features or to isolate features of a certain size. These three parameters worked quite well to identify the impervious surfaces in that three-band NAIP image as you can see in the picture below.
However, if you have access to some of the other data that I mentioned in the webinar, such as Landsat or ASTER, you can take advantage of the near infrared band of data and add that to your analysis. Using the near infrared and red bands of data, ENVI can calculate a vegetation index within the feature extraction workflow. What that means is that you can easily threshold against any vegetation in the image, allowing you to focus on the rest of the non-vegetated features that you’re interested in.
Have any of you tried thresholding for or against vegetation to improve the quality of your results?