## Methods and Protocols The 1-meter land cover map was generated primarily using the 2015 NAIP imagery. It has eight land cover classes and achieved an overall accuracy of 94%. This detailed land cover map was created in the eCognition software using the object-based classification approach. The first step was NAIP image segmentation, where pixels with similar spectral and spatial characteristics were grouped into parcel sized or finer objects. Next, hierarchical rule-based expert systems were developed to classify the objects into specific land cover classes based on their spectral, morphological, and contextual attributes. Various auxiliary GIS data were utilized to improve classification accuracy. They include the 2015 cadastral parcel data, the 2014 USGS Lidar data (1-meter), the 2014 Microsoft/OpenStreetMap Building Footprint data, the 2015 Street TIGER/Line, and the previous NAIP-based land cover map for the year of 2010. Specifically, building footprints and Lidar data significantly improved the boundary detection of above-ground objects. Compared to the 2010 NAIP-based land cover map, buildings and tree canopy shapes were classified more accurately. Finally, post-classification manual editing was applied to minimize classification errors. ## Land Cover Class Description | Class ID | Class Name | Description | |----------|----------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------| | 1 | Building | This class contains all types of buildings in residential, commercial, and industrial areas. | | 2 | Asphalt | This is the dark color impervious surface. This class mainly contains road and parking lots. | | 3 | Bare Soil & Concrete | This class includes bare soil, sand, gravel, and rock in urban and desert environments. It also contains light color impervious surface such as concrete. | | 4 | Tree & Shrub& | This class contains trees and shrubs that are higher than 1.5-meter. | | 5 | Grass& | This class is mainly turf grass. It may also contain shrubs that are lower than 1.5-meter. | | 6 | Water | This class includes different types of water bodies, such as lake, pond, canal, and swimming pools. | | 7 | Active Cropland | This class is cropland covered by green vegetation. | | 8 | Fallow | This class is cropland without green vegetation. | ## Accuracy Assessment Table | Class | Correctly Classified count | Reference count | Classified count | Producer's Accuracy | User's Accuracy | |----------------------|----------------------------|-----------------|------------------|-----------------------|-------------------| | Building | 49 | 49 | 50 | 100.00% | 98.00% | | Asphalt | 48 | 52 | 50 | 92.30% | 96.00% | | Bare Soil & Concrete | 45 | 55 | 50 | 81.80% | 90.00% | | Tree & Shrub& | 46 | 50 | 50 | 92.00% | 92.00% | | Grass& | 45 | 49 | 50 | 91.80% | 90.00% | | Water | 49 | 50 | 50 | 98.00% | 98.00% | | Active Cropland | 48 | 49 | 50 | 98.00% | 96.00% | | Fallow | 46 | 46 | 50 | 100.00% | 92.00% | | Overall Accuracy | | | 94.00% | | | * Producer's accuracy = Correctly Classified count / Reference count * User's accuracy = Correctly Classified count / Classified count * Overall accuracy = Sum of Correctly Classified count/ Sum of Reference (or Classified) count * &: Lidar data only covered 21% of the study area. Within the Lidar data extent, tree and grass achieved higher accuracy >95%. Beyond Lidar coverage, the tree versus grass classification was less accurate due to the lack of surface height information. As a result, tree and grass achieved accuracies of 90% - 92% over the entire study area.

This method step describes provenance-based metadata as specified in the LTER EML Best Practices.

This provenance metadata does not contain entity specific information.