{8e426180-b263-48b0-ac28-514adc076f03} 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 -- Minneapolis/St. Paul, MN -- Meter-Scale Urban Land Cover (MULC) Data (2010) 2016-12-05 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) originator The Minneapolis-St. Paul, MN EnviroAtlas Meter-scale Urban Land Cover (MULC) data were generated from four-band (red, green, blue, and near infrared) aerial photography provided by the United States Department of Agriculture (USDA) National Agricultural Imagery Program (NAIP). The NAIP imagery for the state of Minnesota was collected during the summer and fall of 2010. Lidar data and relevant ancillary datasets contributed to the classification. Eight land cover types were classified: water, impervious surface, soil and barren land, trees and forest, grass and herbaceous, agriculture, woody wetland, and emergent wetland. An accuracy assessment of 644 completely random and 62 stratified random photointerpreted reference points yielded an overall User's Accuracy of 83 percent. The boundary of this data layer is delineated by the US Census Bureau's 2010 Urban Statistical Area for Minneapolis-St. Paul, MN plus a 1-km buffer. This dataset was produced by the US EPA to support research and online mapping activities related to 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). Ecosystem goods and services are critically important to human health and well-being, but they are often overlooked due to lack of information. Using EnviroAtlas, many types of users can access, view, and analyze diverse information to better understand the potential impacts of various decisions. EnviroAtlas data is available at two spatial extents: the contiguous United States and select communities. The National Land Cover Dataset (NLCD) provides land cover for the entire contiguous U.S. at 30-meter resolution. However, the density of an urban landscape requires higher resolution in order to perform meaningful analysis. Meter Scale Urban Land Cover (MULC) was developed to fulfill that need. By comparison, there are 900 MULC pixels for every one NLCD pixel. Anticipated users of MULC include city and regional planners, water authorities, wildlife and natural resource managers, public health officials, citizens, teachers, and students. Potential applications of this data include the design of wildlife corridors and riparian buffers, stormwater management, urban heat island mitigation, access to recreation and green space, urban forestry, and urban landscape ecology. MULC data present a "birds-eye" view that can help identify important features, patterns and relationships in the urban landscape. 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 Minnesota Minneapolis/St. Paul, MN 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 -93.7854091301 -92.9610329068 44.5899921157 45.3094922328 2010-01-01 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). The MULC data for Minneapolis, MN were generated from digital image processing and object-based image analysis (OBIA) of aerial photography, LiDAR data, and relevant ancillary datasets. The aerial imagery is from the United States Department of Agriculture's (USDA) National Agriculture Imagery Program (NAIP) and was collected in the summer and fall of 2010. The photography is distributed as Digital Orthophoto Quarter Quadrangles (DOQQ), also referred to as "quarter quads." Each quarter quad contains four spectral bands — blue, green, red and near-infrared — with 1-meter pixel resolution and 8-bit pixel depth. A quarter quad is approximately 7.7 km north-south and 6.1 km east-west, comprising approximately 47,000,000 pixels. All quarter quads remained in their original UTM 15N, NAD83 projection throughout the classification workflow to preserve data integrity. A Normalized Difference Vegetation Index (NDVI) raster was derived from the NAIP tiles and band stacked in with the 4-band NAIP along with LiDAR derivatives height above ground (HAG) and pulse intensity. LiDAR data was obtained from the Minnesota Geospatial Commons (https://gisdata.mn.gov/dataset/elev-lidar-metro2011) and were collected in the spring and fall of 2011. The LiDAR extent covered the entire study area. For most of the study area the data was collected at a density of 1.5 to 2 points per square meter. Parts of Minneapolis and St. Paul were collected at 8 points per square meter. The 2008 USDA Common Land Unit (CLU) shapefiles were used to classify agricultural land use. The CLU polygons delineate permanent boundaries for a common land cover and land management practice. Agriculture pixels were identified by comparing the CLU layer with NAIP imagery and selecting polygons that overlaid land that appeared to be in some stage of row cropping. ENVI 5.2 Feature Extraction Plugin (FX) (https://www.harris.com/solution/envi) was used to perform a rule-based classification consisting of the following land cover classes: Unclassified (0), Water (10), Impervious Surface (20), Soil and Barren (30), Trees and Forest (40), Grass and Herbaceous (70), Agriculture (80). This schema was intentionally modeled after NLCD land cover. The FX tool uses an object-based approach to classify imagery, where an object (also called a segment, or polygon) is a group of pixels with similar spectral, spatial, and/or texture attributes (https://harrisgeospatial.com/docs/ExtractFeatures.html). The segments in the image ideally correspond to real-world features. The workflow involves segmenting the image into polygons of "like" pixels, computing multiple attributes, and building rules to classify features of interest. Each rule contains one or more additional attributes such as area, length, or texture that are constrained to a chosen range of values (https://harrisgeospatial.com/docs/rule_based_classification.html). Successful results depend heavily on good segmentation, followed by effective rule building. A "good" segmentation ideally creates polygons corresponding to homogeneous features of interest (e.g. trees, buildings, roads, etc). The following is a detailed description of the multi-step classification procedure. Step 1: Individual NAIP tiles were mosaicked into a single raster. Step 2: Created return intensity and HAG rasters from LiDAR point cloud. Return intensity was filtered within the LiDAR dataset and then rasterized. Creating the HAG raster involved several steps. LiDAR data were processed into a digital surface model (DSM) and a digital terrain model (DTM). The DSM consists of all points in the dataset, resulting in surface elevation grid that includes objects that are above the ground, such as trees and buildings. The DTM consists of only ground points, resulting in an elevation grid of bare earth without built structures or vegetation. A HAG raster was created by subtracting the DTM from the DSM. This raster represents the height of objects on the surface of the earth where ground points are equal to zero. Step 3: Creating a tree canopy raster First, a constant raster was created for pixels with HAG > 2 meters. That raster encompassed only trees and built structures because soil, grass, and other impervious surfaces will have HAG values of zero, or between 0 and 2 meters in the case of herbaceous vegetation. Next, building footprints and water were combined into a separate layer and masked out of the constant raster. What remains should be mostly tree/forest pixels. This raster was expanded by 3 meters and then shrunk by 3 meters to fill small gaps (one or two pixels) created by the high resolution lidar. This method was not effective in part of the Minneapolis LiDAR dataset that was acquired when deciduous trees had no leaves (i.e. "leaf-off") because few to no returns existed. The NAIP imagery, though, is always collected "leaf-on" and this led to confusion of tree/forest areas as grass/herbaceous in the contrasting areas. This artifact was addressed in step 7 after rule-based classification was completed. Step 4: Deriving NDVI from NAIP imagery 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. Step 5: Band stacking and segmentation pre-processing NAIP (blue, green, red, and near infrared), NDVI, HAG, intensity, and tree canopy rasters were composited into an 8 band raster. This full extent raster was then clipped by the extent of each quarter quad. These boundaries were then divided again resulting in a final total of 200 tiles. The purpose of creating smaller tiles was to make it less processing intensive for image segmentation in ENVI. Step 6: ENVI rule-based classification with feature extraction (i.e. OBIA) Through a process of trial and error, segmentation and merge levels of 35 and 70, respectively, were found to be optimal for separating land cover features into image segments (https://www.harrisgeospatial.com/docs/fxrulebasedtutorial.html). ENVI calculates statistics for the spectral, spatial, and textural attributes of each segment. These attributes are then used in rule set development. Spatial attributes include area, length, compactness, solidity, roundness, elongation, and rectilinearity. Spectral attributes include average, minimum, and maximum band values. Textural attributes include the average texture or textural variance of a given band. After segmenting the image, rules sets were used to assign segments to the following land cover classes: Water, Impervious Surface, Soil and Barren, Trees and Forest, Grass and Herbaceous, and Shadow. Each rule contains one or more attributes that are constrained to a specific range of values. Some attributes were weighted to give them greater affect in a rule set. Rules were also based on a spectral separability analysis, which determined whether the standard deviation of spectral values overlapped between classes. Bands that did not have overlapping values were particularly effective and important components to rule sets. The majority of rule sets were created with the following attributes: Water - NIR band segment spectral mean, segment area, blue band spectral range, and segment elongation. Trees - NIR band segment spectral mean, texture (coarser texture helps differentiate from grass), and HAG > 2 meters. Hierarchy applied to favor trees over impervious surface. Grass - NIR band spectral mean, texture (smoother texture helps differentiate from trees), HAG < 2 meters. Shadow - spectral mean of blue and green bands. Impervious surface - NIR band spectral mean. Soil - NIR band spectral mean, elongation (below certain level helps filter out roads), HAG (below certain value helps filter out light impervious rooftops), segment area (segments above certain size helped filter out driveways and sidewalks). Hierarchy in the classification was based on a value assigned in the threshold field. For example, if it were determined that trees should have priority over impervious surface in a classification, a threshold value was assigned to both trees and impervious surface to classify trees on top of impervious surface. Step 7: Post-processing the OBIA After completing the classification for all tiles, the output rasters were mosaicked into a single image. All pixels of light impervious surface classified as soil were reclassified to impervious surface by overlaying building footprints and using a conditional statement in ArcGIS. An artifact present after the rule-based classification was the prevalence of impervious surface along the edge of water bodies. To correct this, impervious pixels intersecting with the National Hydrography Dataset (NHD) were reclassified as Water. In some places the NHD layer overlapped legitimate impervious areas such as bridges, which were corrected by hand-editing. All shadow intersecting road centerlines buffered by 5 meters and all shadow intersecting building footprints were classified as impervious. The Eliminate tool in ArcGIS was used to merge the remaining shadows into bordering polygons with the largest shared boundary. Tree-grass confusion was prevalent in vegetated areas where LiDAR data were collected during "leaf-off" conditions. These areas were remediated by supervised classification in GeniePro 2.4, a machine learning pixel-based classification software (observera.com/solutions/genie-pro/). Training pixels were collected for trees, grass, and background; when possible, the same solution was used for multiple areas. The corrected, classified areas were exported to a GeoTIFF and mosaicked on top of the primary land cover raster. Step eight: Classifying agricultural land use Common Land Unit (CLU) polygons created by the US Department of Agriculture were acquired through a publicly available GIS source (www.geocomm.com) and used as an overlay to identify agricultural fields. The CLU polygons serve as a guide for locating agricultural fields. Note: this dataset is comprised of boundaries only and does not include attribute information in order to protect privacy. Based on visual inspection, polygons demarcating agricultural fields were selected manually and converted to a raster layer. Note: these pixels were also classified as either soil (fallow), grass (row crop) or tree (orchard) and saved as an additional, secondary class. Step nine: Addressing confusion of Soil and Barren with Impervious Surface. Soil and Barren and Impervious Surface classes have a high confusion rate due to their spectral and textural similarity. Hand editing was selectively performed as a partial remedy. ***-------------------------------------------*** Accuracy Assessment A second, independent analyst photointerpreted a random sample of six hundred reference points using a fuzzy classification approach (Gopal and Woodcock, 1994). Ancillary image data such as Google Satellite and Street Views, and Bing Aerial and Birdseye views, were used as appropriate to substantiate the interpretation based on the NAIP imagery. Uninterpretable points (e.g., dark shadow) were noted and discarded. As follows, the independent analyst used a scale of 1 to 5 for the fuzzy assessment. 1: Absolutely wrong: classification value is unacceptable (Very Wrong); 2: Understandable but Wrong: classification value is not good. There is something about the site that makes the answer understandable, but there is 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 is acceptable; the classification does 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) If any land cover class ended up with less than fifty reference points a stratified random sample of additional points was generated and interpreted. This step typically applies only to the Soil and Barren class, as was the case in Minneapolis, and less frequently to Water and Agriculture. Thus, in the end, all classes had a minimum of fifty reference samples. The final accuracy assessment resulted in confusion matrices for a non-fuzzy method (MAX) and a fuzzy method (RIGHT), both presented below. The fuzzy method allowed for uncertainty in the analyst's photo interpretation due to complex land cover characteristics. For example, a point located within a pixel on land comprised of patchy grass and soil can be 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. 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 55 1 0 1 0 0 57 0.964912 Grass_Herb 1 167 7 0 27 2 204 0.818627 Impervious 0 4 116 4 3 2 129 0.899225 SoilBarren 0 8 7 29 4 2 50 0.58 TreeForest 0 8 2 0 141 0 151 0.933775 Water 0 0 0 0 0 53 53 1 Row Total 56 188 132 34 175 59 644 Producer's Accuracy 0.982143 0.888298 0.878788 0.852941 0.805714 0.898305 Overall Accuracy 0.871118 K_Hat 0.835488 K Variance 0.000286 RIGHT Agricult Grass_Herb Impervious SoilBarren TreeForest Water Row Total User's Accuracy Agricult 56 1 0 1 0 0 58 0.965517 Grass_Herb 1 163 7 0 23 2 196 0.831633 Impervious 0 4 116 4 3 2 129 0.899225 SoilBarren 0 8 7 29 4 2 50 0.58 TreeForest 0 8 2 0 148 0 158 0.936709 Water 0 0 0 0 0 53 53 1 Row Total 57 184 132 34 178 59 644 Producer's Accuracy 0.982456 0.88587 0.878788 0.852941 0.831461 0.898305 Overall Accuracy 0.877329 K_Hat 0.843645 K Variance 0.000273 Classification errors may stem from multiple sources, including sub-pixel mixing, land use and land cover hierarchies, and land cover with similar attributes. Sub-pixel mixing of land cover occurs when a pixel covers an edge between two land cover types, such as grass from a lawn and impervious surface from a sidewalk. An error can also be recorded between a primary land cover class and a secondary land use class. This would happen if a vegetated pixel is classified as Agriculture and then interpreted as Grass and Herbaceous. The most common confusion occurs between grass and soil. This is due to the prevalence of dry grass during image acquisition that typically occurs during warmer, drier season. These conditions are ideal for NAIP imagery collection because they are conducive to achieving the goal of cloud-free leaf-on imagery. Another common confusion results from land cover with similar attributes; for instance, between Impervious Surface and Soil and Barren, and Grass and Herbaceous and Trees and Forest. This is less problematic for users because the confusion occurs between the similar groups of living vegetation and non-vegetated abiotic surfaces. 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. 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/MSPMN_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 See the Overview section for the complete processing workflow. 2015-11-01T00:00:00 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