{f0e4c532-b1d8-4b96-8d1c-855103e5099c} eng; USA utf8 dataset EnviroAtlas Coordinator U.S. Environmental Protection Agency, Office of Research and Development-Sustainable and Healthy Communities Research Program, EnviroAtlas Geospatial Data Owner (919) 541-3832 109 T.W. Alexander Drive Research Triangle Park NC 27709 EnviroAtlas@epa.gov https://www.epa.gov/enviroatlas pointOfContact 2025-02-13 ISO 19115 Geographic Information - Metadata ISO 19115:2003/Cor.1:2006(E) EnviroAtlas -- St. Louis, MO -- Meter-Scale Urban Land Cover Data (MULC) Data (2016) 2019-06-19 publication U.S. EPA Office of Research & Development (ORD) - National Exposure Research Laboratory (NERL) Research Triangle Park NC publisher US Environmental Protection Agency (US EPA) (additional partners are listed in Supplemental) originator The EnviroAtlas St. Louis, Missouri Meter-Scale Urban Land Cover (MULC) dataset comprises 4188 km2 around the city of St. Louis and surrounding land in parts of eleven counties within Illinois and Missouri. These MULC data and maps were derived from several sources from multiple years: LiDAR (2008-2012); 1-m pixel, four-band (red, green, blue, and near-infrared) leaf-on aerial photography acquired from the United States Department of Agriculture (USDA) National Agriculture Imagery Program (NAIP, 2012, 2014-2016); leaf-off 6-inch pixel four-band imagery (2015) as well as ancillary vector data (e.g., roads, building footprints.). Eight land cover classes were mapped: Water, Impervious Surfaces, Soil/Barren, Tree/Forested, Grass/Herbaceous Non Woody Vegetation, Agriculture, and Wetlands (Woody and Emergent). Wetlands were delineated using the best available existing wetlands data, which was a National Wetlands Inventory (NWI) layer. An analysis of 745 completely random and 226 stratified random photo-interpreted land cover reference points yielded a simple overall user's accuracy (MAX) of 81% and an overall fuzzy user's accuracy (RIGHT) of 90% (see confusion matrices below). This dataset was produced in three phases by the University of Missouri and the East-West Gateway Council of Governments for the Missouri Resource Assessment Partnership (MoRAP) and the US EPA to support research and online mapping activities related to the EnviroAtlas. EnviroAtlas (https://www.epa.gov/enviroatlas) allows the user to interact with a web-based, easy-to-use, mapping application to view and analyze multiple ecosystem services for the contiguous United States. The dataset is available as downloadable data (https://edg.epa.gov/data/Public/ORD/EnviroAtlas) or as an EnviroAtlas map service. Additional descriptive information about each attribute in this dataset can be found in its associated EnviroAtlas Fact Sheet (https://www.epa.gov/enviroatlas/enviroatlas-fact-sheets ). The purpose of the Meter-scale Urban Land Cover (MULC) map is to provide a detailed snapshot of the urban environment and ecosystem for EnviroAtlas' analysis of multiple human and ecosystem health metrics and services. This analysis is typically performed by U.S. Census block group. Greenspace and impervious surface land cover classes figure prominently. Land cover maps are integral to sound, scientific decision making and effective stewardship. A high-resolution urban land cover map quantitatively and qualitatively represents the physical environment and ecosystem matrix of a city. Land cover varies at a fine scale in urban environments, and the existing NLCD (National Land Cover Database, www.mrlc.gov) land cover maps at 30 m spatial resolution are too coarse to measure many urban phenomena, and may not be sufficiently current in rapidly changing urban environments. There are many possible applications, for example: - Knowing the exact location and size of green and gray spaces; - Identifying highly impervious areas that would most benefit from additional trees and other vegetation to mitigate surface runoff and urban heat island effects; - Measuring the width of urban riparian buffers; - Locating potential wildlife corridors, gaps and linkages; - Planning park development for maximum benefits. The overall goal of EnviroAtlas is to employ and develop the best available science to map indicators of ecosystem services production, demand, and drivers for the nation. completed EnviroAtlas Coordinator U.S. Environmental Protection Agency, Office of Research and Development-Sustainable and Healthy Communities Research Program, EnviroAtlas Geospatial Data Owner (919) 541-3832 109 T.W. Alexander Drive Research Triangle Park NC 27709 EnviroAtlas@epa.gov https://www.epa.gov/enviroatlas pointOfContact asNeeded Ecology Ecosystem Environment Health Impact Land theme EPA GIS Keyword Thesaurus Agriculture Biophysical Communities Ecosystem Services EnviroAtlas Environment Environmental Atlas Green Space Land Land Cover Surface Water Water Wetlands theme EnviroAtlas Illinois Missouri St. Louis, MO place None https://edg.epa.gov/EPA_Data_License.html otherRestrictions Use Constraints: None. Please check sources, scale, accuracy, currency and other available information. Please confirm that you are using the most recent copy of both data and metadata. Acknowledgement of the EPA would be appreciated. otherRestrictions Access Constraints: None. PUBLIC https://project-open-data.github.io/schema/#accessLevel Standard Technical Controls eng; USA environment geoscientificInformation imageryBaseMapsEarthCover -90.990687 -89.744822 38.148596 38.948874 9999-01-01 This dataset was produced by the University of Missouri, the East-West Gateway Council of Governments for the Missouri Resource Assessment Partnership (MoRAP) and the US EPA. true Entity and Attribute Information Entity and Attribute Overview: For easier readability, this XML metadata can be opened in a text editor (e.g., Notepad). For more information about EnviroAtlas data, go to https://www.epa.gov/enviroatlas/enviroatlas-data. For information about the Missouri Resource Assessment Partnership (MoRAP), go to https://morap.missouri.edu/. These data were created by two institutions in three phases over the course of several years. The University of Missouri under the direction of Dr. David Diamond, created 55% of the land cover for the study area. EPA completed the other 45% in two phases and added agriculture and wetlands derived from ancillary data sources. This Meter-Scale Urban Land Cover (MULC) dataset for the city of St. Louis and surrounding land in parts of eleven counties within two states (IL, MO): St. Louis MO, Lincoln MO, Monroe IL, Madison IL, Franklin MO, Calhoun IL, Jefferson MO, St Charles MO, St Louis City MO, Warren MO, St Clair IL. Eight land cover classes were mapped: Water (10), Impervious Surfaces (20), Soil/Barren (30), Trees (40), Grass/Herbaceous (70), Agriculture (80), and Wetlands (Woody [91] and Emergent [92]). The primary sources used to derive this land cover layer were 2012 (St. Clair County, IL), 2014 (MoRAP MO and IL), 2015 (Madison and Monroe County, IL), and 2016 (Missouri) National Agriculture Imagery Program (NAIP) imagery (U.S. Department of Agriculture), 2015 6-inch leaf-off imagery (MoRAP MO and IL) and 2008-2012 LiDAR. Phase 1,EPA Processing Steps: St. Clair county was separately classified by the EPA in 2016 before the MoRAP classification. St. Clair County, IL land cover dataset includes data for the St. Louis Urban Statistical Area, as defined by the US Census Bureau, plus a 1-km buffer. The total area classified was 851 square kilometers, with the entire study area within St. Clair County, IL. The land cover classification was developed using 36 Digital Orthophoto Quarter Quadrangle ("quarter quads") acquired from the United States Department of Agriculture (USDA) National Agriculture Imagery Program (NAIP) in 2012. Each quarter quad had 1-meter pixel resolution and included four bands (visible RGB [red-green-blue] plus near-infrared). In addition to the NAIP imagery, LiDAR point cloud data were incorporated into the classification for the entire St. Clair County, IL study area. This LiDAR data collected consisted of two different data sets. These data were acquired in the form of processed rasters and LAS tiles. The St. Clair County LiDAR dataset and derivatives were acquired from Illinois Height Modernization Program, Illinois State Geological Survey (ISGS), and Illinois Department of Transportation, 2002-2013, Illinois LiDAR county database. Digital terrain model (DTM) and digital surface model (DSM) rasters were pre-processed 3-foot rasters. The DTM was subtracted from the DSM to get height above ground (HAG) for above ground features. A 1-meter intensity raster was created using LAS point cloud data for St. Clair County. The LAS tiles also had a nominal point spacing of 3 feet with a vertical accuracy of 12.5 cm. The St. Clair LIDAR dataset covered approximately 99% of the study area. Data was collected from March 3, 2012 to April 2, 2012. LAS to multipoint conversion using ArcTool 3D Analyst were used below: 1. Unclassified, 2. Ground, 3. Low Vegetation, 4. Medium Vegetation, 5. High Vegetation, 6. Building, 8. Model Key Points, and 9. Water. An ArcGIS model was used to develop a raster for water bodies in the study area by vectorizing the extent of the LAS dataset identified above. The model rasterized LAS points -> expanded the raster -> shrunk the expanded raster -> vectorized the shrunk raster -> eliminated voids in the vector file below a certain area threshold. This process created a dataset extent with voids in areas without returns such as water bodies. The Union tool in ArcGIS was used to fill in the water voids and select only desired water bodies. These polygons were then exported and rasterized. A shapefile with a total of 36 intersecting USGS quarter quadrangles (QQs) was overlaid on the study areas. Land cover classification was performed for each of these intersecting QQs, with each individual QQ buffered by 3 meters to prevent unclassified areas in the final product. In all of the QQs, both NAIP imagery bands and LiDAR data bands available were stacked to create a 7-band composite mosaic before classification. The bands consisted of the following: NAIP bands 1-4, NDVI (normalized difference vegetation index) band (created with NAIP), LiDAR height above ground (HAG), and LiDAR intensity. An NDVI raster was created using the NAIP imagery as follows: NDVI= (NIR-VIS)/(NIR+VIS), where NIR is near infrared reflectance, VIS is reflectance of visible red light, and NDVI is a dimensionless number ranging from –1 to 1. NDVI exploits the high NIR reflectance of vegetation to separate it from non-vegetation. In addition, the NDVI ratio helps to normalize variable illumination and atmospheric conditions across the entire image mosaic. To derive the LiDAR intensity band, the tiles in the LiDAR dataset were re-projected into the coordinate system of the NAIP imagery (NAD_1983_UTM_Zone_15N) and then clipped to the boundaries of the QQs. The LiDAR point data were then imported into ArcGIS and interpolated from vector to raster data using the LAS to raster tool. To derive the HAG, DTM was subtracted from the DSM as mentioned in the above section. DSM and DTM were rasters already processed through the Illinois Height Modernization Program. A total of eight land use / land cover (LULC) classes were mapped: Water, Impervious Surface, Soil and Barren, Tree and Forest, Grass and Herbaceous, Agriculture, Woody Wetland and Emergent Wetland. The four classes 1) Impervious surfaces, 2) Soil and barren, 3) Tree and forest and 4) Grass and herbaceous were classified using GeniePro 2.4 (www.observera.com), a machine-learning supervised classification software program. These layers were created by selecting groups of appropriate training pixels. Solution files were then created in GeniePro, which were then used to map the individual classes throughout the study area. A solution algorithm was evolved by running approximately 5000 iterations to create a classification algorithm. Errors were corrected using hand-digitization tools in GeniePro. LiDAR data was used where available in classifying Trees, Grass, Soil, and Impervious Surface. Classes were classified in GeniePro in the following sequence: 1) Trees/Grass-Herbaceous/Non-vegetation, 2) Soil/Impervious. The entire water layer was produced with LiDAR data through the use of the technique described in the previous paragraph. After each step in the above sequence, all the individual classifications were mosaicked together to form a complete class. ArcGIS tools were then used to mask the completed classification out of the imagery before beginning creation of the next land cover type in the sequence. Therefore water/trees/grass were all masked out before classifying soil/impervious. Upon completion of the classification process using GeniePro, the Agriculture class was created. Common Land Unit (CLU) data was available for the state of Illinois. CLU data (shapefile) were visually inspected and edited with hand digitizing tools in ArcGIS. Although this process was more tedious and time consuming than using GeniePro, the results produced a high level of accuracy. Agricultural data created through CLU and county land parcel data was then merged and projected to NAD_1983_UTM_Zone_15N to create a final vector layer for the entire study area. Agricultural parcel vector data was then converted to raster to create the agriculture class for St. Clair County, IL. Before the incorporation of wetland data the existing classes were mosaicked using the following hierarchy and assigned cell values: 1) water - 10, 2) trees/forest - 40, 3) impervious surface – 20, 4) agriculture - 80, 5) grass/herbaceous - 70, 6) soil/barren - 30. Wetland data for the study area was acquired from the National Wetland Inventory. Polygons corresponding to woody and emergent wetlands were extracted. Classified pixels falling within the boundaries of these polygons were extracted from the 6-class classification. Pixels initially classified as "grass and herbaceous" were reclassified as "emergent wetlands" and assigned a cell value of 92; pixels initially classified as "tree and forest" were reclassified as "woody wetlands" and assigned a cell value of 91. An accuracy assessment was performed on the 6-class classification (Woody and Emergent wetland classes were excluded). The area used fell within the the St. Clair County portion of the St. Louis Urban Statistical Area plus 1-km buffer. Phase 2, MoRAP Processing Steps: University of Missouri under the direction of Dr. David Diamond and the East-West Gateway Council of Governments as part of the Missouri Resource Assessment Partnership (MoRAP) created 1-meter per pixel land cover data for the urban areas of the St. Louis community using object based classification in eCognition landcover mapping software(http://www.ecognition.com/). EPA added agriculture and wetlands taken from ancillary data sources and translated land cover from seven MoRAP classes into seven MULC classes. The MoRAP study area was derived from boundaries of LiDAR data 1km2 tiles that circumscribe areas with more than 35% urban landcover as classified in National Landcover Dataset (NLCD, 2011), including seven counties over a 2300 km2 area (St. Louis MO, Monroe IL, Madison IL, Jefferson MO, St Charles MO, St Louis City MO, St Clair IL). Classification was developed using 2015 4-band leaf-off imagery collected 6-inch resolution, normalized difference vegetation index (NDVI) and built-up area index (BAI) derived from 2015 imagery, 2014 4-band NAIP leaf on imagery, and LiDAR from a variety of years covering 2008-2012 to create vector objects of elements within the data like roads, homes, trees, and water (called image objects in eCognition) at 1m resolution. The four banded imagery included red, blue, green and near infrared reflectance. NDVI is calculated by NDVI= (NIR-R)/(NIR+R) where NIR is near infrared band and R is red band. BAI is calculated by BAI= (B-NIR)/(B+NIR) where B is blue band and NIR is near infrared band. Through eCognition software, each image object had 105 variables related to it such as band reflectance, LiDAR height, object texture and object shape. 8000 analyst-selected objects were used to train the classification which was based on the random forest algorithm. The training objects represented at least 900 samples for each of seven land cover classes: Urban/Impervious, Open Water, Row Crops, Grassland, Evergreen Woody Vegetation, Deciduous Woody Vegetation, and Barren/Sparsely Vegetated. Classification was iterative with successive refinements made by adding training data until the level of accuracy increased. Post classification systematic review was conducted at 1:4000 and 1:10000 scale and misclassified objects were reclassed to proper class where needed, primarily green rooftops reclassed from tree cover to impervious. Phase 3, EPA Processing Steps: Additional classification was conducted in 2019 after the MoRAP classification in surrounding areas that make up the EnviroAtlas SLMO community boundary. The classification consisted of 1037 km2 area surrounding the MoRAP urban mask out to 1km boundary of the community census block groups. This phase 3 area includes sections from all 11 counties (St. Louis MO, Lincoln MO, Monroe IL, Madison IL, Franklin MO, Calhoun IL, Jefferson MO, St Charles MO, St Louis City MO, Warren MO, St Clair IL). The most up to date NAIP and LiDAR data was used in this classification. The Missouri LiDAR LAS tiles were obtained from the Missouri Spatial Data Information Service (MSDIS, http://msdis.missouri.edu/), 2010-2011, and Illinois LiDAR LAS tiles from the Illinois Geospatial Data Clearinghouse (https://clearinghouse.isgs.illinois.edu/), 2012, 2014. All LiDAR data collected in feet was converted to meters. DEM and DSM rasters were created from LiDAR point cloud data using the LAS dataset to raster conversion tool. The DEM was subtracted from the DSM to get height above ground for above ground features. A 1-meter intensity raster was created using LAS point cloud data for each county. 1-meter HAG and intensity rasters were mosaicked together and clipped to the SLMO community boundary. Areas of unusable data were present in the HAG and intensity output mosaics, as a result of county boundary locations and areas of overlap during LiDAR collection. These areas were clipped and replaced with a differing county's data where necessary. The HAG and intensity mosaics were then clipped to the areas within the community boundary not classified by MoRAP, mostly located outside urban areas. Unclassified no data pixels and small pixels present within the dataset were removed to aid in processing and will be re-added and averaged into the dataset in future steps. NAIP imagery was obtained from the USDA ArcGIS REST server (https://gis.apfo.usda.gov/arcgis/rest/services/NAIP), Missouri 2016 60cm and Illinois 2015 1m. The Missouri and Illinois NAIP imagery was clipped to the boundaries of the QQs within the greater SLMO community boundary using ModelBuilder. The tiles of the NAIP imagery were re-projected into the coordinate system of the LiDAR dataset (NAD_1983_UTM_Zone_15N). To form a cohesive 1m mosaic to match the 1-meter resolution of the LiDAR data, the 0.6m Missouri NAIP imagery was resampled to 1m. Only the NAIP imagery necessary to fill in the gaps from MORAP LC was extracted for the final 1m NAIP mosaic. To match the LiDAR HAG and intensity layers, unclassified no data pixels and small pixels present within the dataset were removed. The imagery bands and LiDAR data bands available were stacked to create a 7-band composite mosaic before classification. The bands consisted of the following: NAIP bands 1-4 (red, green, blue, and NIR), NDVI band (created with NAIP in method described in Phase 1), LiDAR height above ground (HAG), and LiDAR intensity. The LiDAR point data were imported into ArcGIS and interpolated from vector to raster using the LAS to raster tool. HAG was derived as previously described in Phase 1 using interpolated DEM subtracted from the DSM. Object based random forest classification was run on ArcPro 2.3 on nine sections of the 7-band composite that make up unbroken tiles of the missing area surrounding the MoRAP and St. Clair classification. Each section included 1km border into the already classified area which was used to extract training segments and better blend the multiple classifications together. Training data was collected from areas based on object based segments that were purely classified by MoRAP or St. Clair as a single MULC class (excluding wetlands and agriculture). A total of 32000 training samples were collected (6400 samples per class respectively). The resulting classified product was run through iterative threshold refinement using ancillary Navteq 2015/2016 roadline data, Microsoft 2018 building footprint data (https://github.com/microsoft/USBuildingFootprints), 2015/2016 USDA Cropland Data Layer (CDL, https://nassgeodata.gmu.edu/CropScape/), 2008 USDA Common Land Unit (CLU) agriculture data, and waterbody data from National Wetland Index 2015/2016, as well as previously used LiDAR derived HAG to correct misclassification between tree/forest class and grass. Navteq roadline data was buffered using the average lane size of US highways, 3.65676m, multiplied by To and From Lane counts where available and 3.65676m multiplied by Lane Category values where missing; additionally roadlines categorized as ramps were considered single lane roads and buffered by 3.65676m. Where roadlines were considered pedestrian accessible (AR_PEDEST==Y), the buffer width was increased by an additional 2.438m. The final buffered product was used as a mask in ArcTool Raster Calculator to reclassify water and soil classes with an NDVI value between -0.45 and 0 to impervious class. Due to misclassified driveways, a buffer mask was created using Microsoft building footprint and Navteq roadline data to reclassify soil to impervious. The Microsoft building foot prints were converted to points at the center of each polygon, the nearest point along Navteq roadlines for each building point was calculated using the Near ArcTool, points were appended to the same feature class and then converted from points to lines based on related IDs to create a line feature of estimated driveways. The estimated driveway centerlines were buffered by 10m, the soil class within the buffer that was not within 2m of NWI water bodies or agriculture polygons were reclassified to impervious if NAIP red band >190 in Raster Calculator. To correct misclassed water and areas where water was partially misclassed as impervious, a series of conditional statements were used to relate NWI waterbody data to the water class in Raster Calculator. The water class was converted from raster to polygon and buffered by 38.1m (about the distance of widest sand bar). For buffered areas that intersect with NWI waterbody data but not Navteq or Agriculture data, soil and impervious classes were reclassed as water when NDVI<=-0.25, reclassed to soil when NDVI is greater than -0.25 and less than or equal to 0.2, and reclassed to grass when NDVI is greater than 0.2. For buffered areas that intersect Navteq and Microsoft building footprint but not NWI water bodies or does not intersect any ancillary data and has an area less than 5000m2, water was reclassed to impervious when NDVI ranged from -0.45 and -0.25, reclassed to soil when NDVI ranged from -0.25 and 0.2 and reclassed to grass when NDVI ranged from 0.2 and 0.3. To correct areas where grass was misclassified as forest and forest was misclassified as grass, two steps were taken. First, where forest class had HAG values <0.2m and the segment from random forest classification majority HAG was < 0.2, forest was reclassed to grass. To define segment majority as <0.2 or >=0.2, HAG was converted to binary and then Zonal Statistics ArcTool was used to define majority using the segmented data as zones. In the second step to correct misclassifications, all grass with a HAG above or equal to 0.2m was reclassified as forest. To correct soil misclassified grass, grass with NAIP green band >180 was reclassified to soil. Finally, during a systematic review of classification, hand-digitized edits were created to fix mostly misclassified soil to impervious and misclassified shadowed areas to correct class. 2008 USDA Common Land Unit (CLU) vector data and 2015/2016 USDA Cropland Data Layer (CDL), and NAIP and Google imagery were used as guides to determine the areas most likely to be used for cultivated row crops for this classification and to update MoRAP. The data was used to create agriculture class polygons before running threshold based rules on the classification but only applied after corrections had been made. CLU data is not classified but delineate agricultural land boundaries based on land use features such as roads and water bodies and, as a result, remain relatively constant over time. CDL data has lower resolution (30m), but more up-to-date land cover classification based on supervised classification of satellite imagery. Majority defined classes were assigned to CLU vector data based on the 2015/2016 CDL landcover resampled to 1m resolution using Zonal Statistics majority tool. CLU vector data assigned with agriculture classes from the CDL were recoded into the MULC Agriculture (80) classification through reclassification of soil and grass MULC classes to agriculture masked by the CLU vectors, leaving trees, shrub, impervious and water within fields as they were classified. MoRAP and St. Clair county classifications were updated to the NAIP 2016/2015 imagery through two systematic corrections followed by a thorough review of entire classification where areas in need of an update were circled and classified using object based random forest classification in ArcPro 2.3 and a 5-band composite of NAIP 2016/2015 red, green, blue, NIR band and NDVI derived from NAIP. 305 training samples were hand selected by segments for the 153km2 of areas in need of classification (31 Water, 92 Impervious, 39 Soil/Barren, 64 Tree/Forest, and 78 Grass/Herbaceous samples respectively). After review, corrections were made mostly to the water class through hand digitizing. In the MoRAP urban areas, forest with NDVI <81 were reclassed to impervious, excluding 2m within water class. In St. Claire county, soil was reclassed to grass where NAIP green band was >155. After corrections were made, the agriculture class was added using the method describe above paragraph. Woody and emergent wetland classes were also added to entire SLMO community using 2015/2016 NWI wetland data following a similar method described for St. Clair county classification. ***-------------------------------------------*** Accuracy Assessment - An accuracy assessment was conducted on the completed land cover classification that included the area classified by MoRAP plus the Agriculture class added by EPA. Wetlands were not included in the accuracy assessment due to difficulties identifying those areas in the imagery per EnviroAtlas accuracy assessment methods. An Analyst performed photointerpretation of 2015/2016 NAIP aerial photography used in the classification. Seven hundred random reference points (100 per land cover class) were interpreted and labeled using a fuzzy classification approach (Gopal and Woodcock 1994) for the MoRAP and Phase 3 area. This permitted the Analyst to assign a confidence value to the photo interpreted label at each reference point. Confidence is expressed as an integer from 1 to 5: 1: Absolutely wrong: classification value was unacceptable (Very Wrong); 2: Understandable but Wrong: classification value was not good. There was something about the site that made the answer understandable, but there was clearly a better answer. Classification would pose a problem for users of the map. (Not Right); 3: Reasonable or Acceptable: Maybe not the best possible classification but it was acceptable; the classification did not pose a problem to users of the map. (Right); 4: Good Answer: Would be happy to find this classification given on the map (Very Right); 5: Absolutely Right: No doubt about the match. (Perfect) After the initial assessment, additional samples were generated for any class with less than 50 reference samples and for the St. Clair classification. Using stratified random sampling, sufficient sample points were generated such that all classes had a minimum of fifty reference samples (most classes had approximately 100 reference samples) as well as 26 points per class for St. Clair county (equaling 17% of total points, St. Clair accounts for 17% of total community area). 226 stratified random points were added and included in the final accuracy assessment for a total of 921 points. The final accuracy assessment was tabulated in confusion matrices for the non-fuzzy method ("MAX") and fuzzy method ("RIGHT"), both presented below. The MAX interpretation was correct if the classified land cover matched the interpreter's highest score, illustrating that the interpreter found this class to be the most appropriate for that location. The RIGHT interpretation was correct if the classified land cover matched any class the interpreter gave a value of 3 or greater, illustrating that the interpreter found the classification to be acceptable but another class may have been more appropriate. Therefore, the MAX is a more conservative view and RIGHT is a more liberal view. This fuzzy method allowed for uncertainty in the analyst's photo interpretation due to complex land cover characteristics. For example, for a point located within a pixel on land comprised of patchy grass and soil, the interpreter may have assigned a 4 for grass and a 3 for soil. This was accounted for by the RIGHT results, while MAX only accounted for the highest value recorded by the interpreter. The following confusion matrices summarize the accuracy assessment MAX and RIGHT results. For easier readability, this XML metadata can be opened in a text editor (e.g., Notepad) and the confusion matrix may be copied from text editor to an Excel spreadsheet. Confusion matrices can also be viewed in original formatting by opening in ArcCatalog: MAX Agricult Grass_Herb Impervious SoilBarren TreeForest Water Row Total User's Accuracy Agricult 108 6 0 7 0 1 122 0.885246 Grass_Herb 14 128 18 8 21 1 190 0.673684 Impervious 0 7 126 5 9 0 147 0.857143 SoilBarren 1 7 6 58 1 4 77 0.753247 TreeForest 1 31 6 2 262 4 306 0.856209 Water 0 0 0 0 3 76 79 0.962025 Row Total 124 179 156 80 296 86 921 nan Producer's Accuracy 0.870968 0.715084 0.807692 0.725 0.885135 0.883721 nan nan Overall Accuracy 0.823018 nan nan nan nan nan nan nan K_Hat 0.776816 nan nan nan nan nan nan nan K Variance 0.000252 nan nan nan nan nan nan nan RIGHT Agricult Grass_Herb Impervious SoilBarren TreeForest Water Row Total User's Accuracy Agricult 118 3 0 1 0 0 122 0.967213 Grass_Herb 3 158 12 5 12 0 190 0.831579 Impervious 0 4 135 1 7 0 147 0.918367 SoilBarren 0 3 3 69 1 1 77 0.896104 TreeForest 1 20 4 1 277 3 306 0.905229 Water 0 0 0 0 3 76 79 0.962025 Row Total 122 188 154 77 300 80 921 nan Producer's Accuracy 0.967213 0.840426 0.876623 0.896104 0.923333 0.95 nan nan Overall Accuracy 0.904452 nan nan nan nan nan nan nan K_Hat 0.879204 nan nan nan nan nan nan nan K Variance 0.000151 nan nan nan nan nan nan nan Classification errors may stem from multiple sources. Some are due simply to mixing land cover and land use categories in the analysis. For example, Agriculture-Grass or Agriculture-Soil confusion are mostly artifacts. Agriculture land use is also Soil land cover while barren, and Grass-Herbaceous land cover while growing row crops. Grass-Soil is the most common confusion due to intermixing and the presence of brown, senescent grass or sparse grass. An underlying assumption is that most non-arid region soil (not barren rock) is capable of supporting some Grass-Herbaceous vegetation at some point during the year. Grass-Tree confusion and Soil-Impervious confusion are also relatively common. This is perhaps a less problematic error than Grass-Impervious, or Tree-Impervious. Grass and Tree are both living vegetation, versus non-living Impervious surfaces. Shrub intermixing with Grass and Soil is likely introduced from prioritizing height above ground data derived from LIDAR during classification. References: Gopal, S. and Woodcock, C. (1994). Theory and Methods for Accuracy Assessment of Thematic Maps Using Fuzzy Sets. Photogrammetric Engineering and Remote Sensing 60(2), 181-188. O'Neil-Dunne J, MacFaden S, Royar A. (2014). A Versatile, Production-Oriented Approach to High-Resolution Tree-Canopy Mapping in Urban and Suburban Landscapes Using GEOBIA and Data Fusion. Remote Sens. 2014, 6, 12837-12865. http://www.mdpi.com/2072-4292/6/12/12837. U.S. Department of Agriculture. National Agriculture Imagery Program imagery. Farm Service Agency. Aerial Photography Field Office: U.S. Department of Agriculture Web page, http://www.fsa.usda.gov/FSA/apfoapp?area=home&subject=prog&topic=nai. U.S. Fish and Wildlife Service. Illinois and Missouri National Wetlands Inventory digital data. Accessed 2018. http://wetlands.fws.gov/. University of Missouri. Missouri Resource Assessment Partnership Web page. Accessed 2018. https://morap.missouri.edu/. Microsoft Github Repository. USBuildingFootprints. Accessed 2018. https://github.com/microsoft/USBuildingFootprints USDA National Agricultural Statistics Cropland Data Layer. Illinois (2015) and Missouri (2016) Cropland Data Layer. Accessed 2018. https://nassgeodata.gmu.edu/CropScape/ Missouri Spatial Data Information Service. MOLiDAR LAS File Download Tool Webpage. Accessed 2017. http://msdis.missouri.edu/ Illinois Geospatial Data Clearinghouse. Illinois Height Modernization (ILHMP): LiDAR Data Webpage. Accessed 2017. https://clearinghouse.isgs.illinois.edu/ Entity and Attribute Detail Citation: https://www.epa.gov/enviroatlas/enviroatlas-fact-sheets EnviroAtlas Coordinator U.S. Environmental Protection Agency, Office of Research and Development-Sustainable and Healthy Communities Research Program, EnviroAtlas Geospatial Data Owner (919) 541-3832 109 T.W. Alexander Drive Research Triangle Park NC 27709 EnviroAtlas@epa.gov https://www.epa.gov/enviroatlas distributor https://enviroatlas.epa.gov/download/SLMO_MULC_tif.zip download Download The URL providing direct access to the downloadable dataset. download https://enviroatlas.epa.gov/arcgis/rest/services/Supplemental/Landcover_AllCommunities/ImageServer ESRI:ArcGIS API The endpoint of a web service to access the dataset (REST endpoint, WMS GetCapabilities URL, or a SOAP WSDL endpoint). webMapService https://www.epa.gov/enviroatlas/enviroatlas-data information Documentation URL for accessing related documents such as data dictionary, technical information about a dataset, developer documentation, etc. information https://www.epa.gov/enviroatlas information Home Page Alternative landing page used to redirect user to a contextual, Agency-hosted "homepage" for the Dataset or API information dataset Horizontal Positional Accuracy Report Data were collected using methods that are accurate to within 2-5 meters (EPA National Geospatial Data Policy [NGDP] Accuracy Tier 2). For more information, please see EPA's NGDP at https://www.epa.gov/geospatial/geospatial-policies-and-standards) Features represented have not been tested for completeness Features represented have not been tested for completeness Tests for integrity have not been performed. This lineage was generated by the FGDC CSDGM to 19139 transformation The original MULC raster was developed. See the Overview section for the complete processing workflow. 2013-11-05T00:00:00 Two major classification errors were corrected: 1) road layers overlaying tree canopy and 2) houses misclassified as trees. See the Overview section for details 2017-01-01T00:00:00 unknown 2029-02-13 This metadata was automatically generated from the FGDC Content Standard for Digital Geospatial Metadata standard (version FGDC-STD-001-1998) using the 2024-02-09T11:11:00 version of the FGDC CSDGM to ISO 19139 transform. EnviroAtlas Coordinator U.S. Environmental Protection Agency, Office of Research and Development-Sustainable and Healthy Communities Research Program, EnviroAtlas Geospatial Data Owner (919) 541-3832 109 T.W. Alexander Drive Research Triangle Park NC 27709 EnviroAtlas@epa.gov https://www.epa.gov/enviroatlas pointOfContact