top of page

Spot-tailed earless lizard

Map3.jpg

Methodology

Baseline land cover maps were generated through automated categorization of very high resolution imagery (0.6 meters pixel size) provided by the National Agricultural Imaging Program (NAIP).

NAIP imagery utilized for the mapping process was obtained for the year 2018 (with the exception of Webb, Nueces and Ward counties for which we used data from years 2015 and 2020). We applied an optimized Geographic Object-Based Image Analysis (GEOBIA) approach for ingesting, processing, and classifying NAIP imagery into land cover and land use classes over a large area in a time and computationally efficient way. We pre-processed a total of 1,471 NAIP scenes for thirteen counties in central and south Texas by reducing the spectral dimensionality of NAIP imagery using principal component analysis (PCA), texture analysis, and edge detection. Objects created through image segmentation were then used to implement a random forest algorithm for classification with minimal post-processing corrections. For each county, we collected at least 1,200 sampling points mostly through image interpretation. A preliminary validation for Tom Green and Irion counties yielded 94.7% overall accuracy and a Kappa statistic of 92.5% from area-based cross-validation.
The user’s and producer’s accuracies across the mapped classes were higher than 85% for most classes. 

The products here presented are Level 1 (no post-processing or corrections). This version is not the final version, therefore, sharing the maps outside the organization is not encouraged. 

The design and application of the methodology was led by PhD candidate Mukti Subedi at Texas Tech University. For more information, contact the GST lab's PI Dr. Carlos Portillo-Quintero at carlos.portillo@ttu.edu.

 

01_Chapter1_outline_01.png
bottom of page