Module 2: Land Use/Land Cover and Ground Truthing

    During this week's lab, I learned about land use/land cover (LULC) classification, and I accessed the accuracy of LULC layers using ex situ ground truthing methods. 

    For the first part of the lab assignment, I created an LULC dataset using satellite imagery of Pascagoula, MS. During the process, I realized how tedious the process can be, but I did learn a few tricks that saved me a significant amount of time. The first tip is the use of the clip function when modifying the features of a feature class. Before mapping each class take time to determine the major LULC classes. Map those areas first as large areas (don't exclude areas that differ from the primary LULC class). Once you have the large area mapped. Go through and map the areas that differ from the main LULC class of the area. Once you have the smaller areas mapped, you can select them and clip the larger area so they become separated. Another helpful tip was to use the snapping function when drawing the polygons. This ensures that the polygons directly border one another. Preventing a gap between the polygons would be nearly impossible without this function. Lastly, it is important to determine how invested you want to be with your classification. The map that I generated is somewhat coarse since I was only classify to Level II LULC classification. Higher classifications will require greater time to classify, so make sure that you take the time to consider the pros and cons of an LULC product that it is more detailed. 

    For the second part of the lab assignment, I explored ground truthing and assessing the accuracy of an LULC layer. I did this by randomly generating points throughout the study area. I then visited those points in google maps and used the satellite imagery and the street view to determine the true classification of the point. Next, I compared the observed classification and the mapped classification and noted any differences. I used this information to assess the overall accuracy of my LULC layer. Ultimately, I determined that my LULC layer had an accuracy of 80% which is not bad. The error for some classes was greater than others (e.g. cropland). A random stratified sampling method would have provided more insight into the accuracy for each class.

Below is a copy of my map displaying the LULC classes and my accuracy assessment.