Predicted Spatially Complete Zoning Map of North Carolina

Spatially-complete zoning map of North Carolina, USA. The <strong>results </strong>folder contains results of a machine learning (random forest) model predicting 3 core district zones (residential, non-residential, and mixed use) and 13 sub-district zones (open space, industrial, commercial, office, planned use, high-density residential, medium-high-density residential, medium-density residential, medium-low-density residential, low-density residential, agricultural residential, mixed use, and downtown). Results are provided as 30-m rasters (.tif) with each value corresponding to a zoning district. Table containing zone district ID (number) and zone district name (character string) is included in <strong>zone_classification.csv</strong>. Final (spatially complete statewide maps) can be found in the <strong>final_predicted </strong>folder. This folder includes Statewide core district results in <strong>NC_predicted_core.tif</strong> and statewide sub-district results in <strong>NC_predicted_sub.tif</strong>. Zoning was generalized and reclassified into 3 core district zones and 13 sub-district zones (described above). Reclassified zoning data, collected from 39 counties in North Carolina is provided in the <strong>observed </strong>folder with core districts in <strong>core_district_observed_zones.tif</strong> and sub-districts in <strong>sub_district_observed_zones.tif</strong>. Also in this folder is <strong>zoning_implementation_NC.csv</strong> which includes links to the source data (zoning map and zoning ordinance) for all collected data. Two models were created to predict zones under different data availability scenarios (i.e., scenarios that assume different levels of data availability). Predictions labeled “within_county” utilized the within-county model which predicts zoning districts in areas where zoning data is partially available for that county. To approximate scenarios of incomplete data accessibility, 20% of the data was randomly removed from training and reserved for independent performance assessments. Predictions labeled “between-county” utilized the between-county model which predicts zoning districts in areas where zoning data is inaccessible. To approximate this scenario, multiple between-county model iterations were computed by randomly removing entire counties from the training dataset and computing performance metrics on the removed (test) counties. Predictions are provided for both core districts and sub-districts (described above). Results from these models can be found in the <strong>predicted </strong>folder. This folder contains four subfolders: <strong>core_district_within_county</strong>, <strong>sub_district_within_county</strong>, <strong>core_district_between_county</strong>, and <strong>sub_district_between_county</strong>. Within each of these folders are predicted maps 30-m raster (.tif), performance reports including precision, recall, and f1 score overall and per district (.csv), and accuracy maps (3-km grid shapefile [.shp, .shx, .prj, .dbf]) with values corresponding to the proportion of misclassified pixels within a grid cell. Multiple randomized testing county samples were conducted for the between-county models. Each random sample is labeled <strong>r*_</strong> where * is replaced with a number between 1 and 15.