Some of the datasets I’ve been working with lately include:
- NAIP, 1 meter, 4 band imagery – A colleague classified 3.5 county’s worth of NAIP images into between 4-7 categories and handed it to me to reclassify into “trees” and “not trees” pixels. Though I was not asked to do an error analysis, I loathe using classified imagery without a formal error analysis, so I did one. With 20 randomly chosen pixels in each county (since they were classified separately) checked by-eye to see if they were correctly identified or not, we got a 94% concurrence. That is an excellent error rate. Another error-check that should be done, however, is to randomly choose 20 non-forest pixels in each county to determine concurrence since the original error-analysis was heavily weighted toward “tree” pixels given the huge percentage of trees in the study area. That will be one of my next tasks if I have the time to undertake it.
- NOAA CCAP, 30 meter, landcover – This dataset covers the coastal regions of the U.S. but was problematic for my project’s needs in that it has a “Palustrine Forested” category whereas we wanted to know specifically what type of forest (coniferous, deciduous, mixed) that those pixels represented. The NOAA people were very responsive and sent me the Landsat mosaics that were used to produce each of the four CCAP year’s worth of data (1992, 1996, 2001, and 2006) so that I could mask out those palustrine forested pixels and reclassify them using a supervised classification. While there is very little way to error-test the results because the data are at least 5 years old, some visual assessment of the 2006 results showed a decent amount of concurrence with what we know to be true on-the-ground right now.
- Regions – I’m currently involved in a fun project involving by-eye digitizing, at a high resolution, some logically drawn regions (some might call these “territories”) based on demographics and existing political boundaries, but weighted more toward demographics and travel corridors when they cross political boundaries. This is a very fun exercise in the sense that it gives a level of geographic awareness that is only possible when immersed in such a task.
So…what data have you been working with lately?