Supervised Pixel-Based Land Cover Classification
Performing a land cover classification for Saginaw Woods using remote sensing imagery and field data.
As part of my field remote sensing course, I learned how to perform a supervised pixel-based land cover classification using a study area of Saginaw Woods, Michigan. We had two field survey days where we walked around the property and recorded ground-truthing points for various land cover types, such as deciduous/coniferous forests, wetlands, and developed land.
Based off the collected field data and 2022 NAIP imagery for the study area, I created training sample polygons that represented each level 1 land use land cover code. Using the Image Classification Wizard in ArcGIS Pro, I created a classification schema and ran a Random Trees classifier that produced a pixel-based land cover classification map.
To assess the accuracy of the classification, I used a confusion matrix. The method resulted in a user accuracy of 92% and a producer accuracy of 88%. For improved classification, I would collect better validation and sampling points.
