20210402 09492700 1.0 FALSE CopyRaster D:\DBConnections103\DEV_QA_DBConnections\QAGISPublishing_gisras1_ro.sde\gisras1.OC.LandCover2010 D:\LandCover2010.tif # # 255 NONE ColormapToRGB # NONE NONE TIFF NONE LandCover2010.tif file://D:\LandCover2010.tif Local Area Network 13304893.861130 13469697.016017 340130.169570 508604.584412 1 002 Projected GCS_North_American_1983 Linear Unit: Foot (0.304800) NAD_1983_StatePlane_Michigan_South_FIPS_2113_Feet_Intl <ProjectedCoordinateSystem xsi:type='typens:ProjectedCoordinateSystem' xmlns:xsi='http://www.w3.org/2001/XMLSchema-instance' xmlns:xs='http://www.w3.org/2001/XMLSchema' xmlns:typens='http://www.esri.com/schemas/ArcGIS/10.7'><WKT>PROJCS[&quot;NAD_1983_StatePlane_Michigan_South_FIPS_2113_Feet_Intl&quot;,GEOGCS[&quot;GCS_North_American_1983&quot;,DATUM[&quot;D_North_American_1983&quot;,SPHEROID[&quot;GRS_1980&quot;,6378137.0,298.257222101]],PRIMEM[&quot;Greenwich&quot;,0.0],UNIT[&quot;Degree&quot;,0.0174532925199433]],PROJECTION[&quot;Lambert_Conformal_Conic&quot;],PARAMETER[&quot;False_Easting&quot;,13123359.58005249],PARAMETER[&quot;False_Northing&quot;,0.0],PARAMETER[&quot;Central_Meridian&quot;,-84.36666666666666],PARAMETER[&quot;Standard_Parallel_1&quot;,42.1],PARAMETER[&quot;Standard_Parallel_2&quot;,43.66666666666666],PARAMETER[&quot;Latitude_Of_Origin&quot;,41.5],UNIT[&quot;Foot&quot;,0.3048],AUTHORITY[&quot;EPSG&quot;,2253]]</WKT><XOrigin>-107138600</XOrigin><YOrigin>-97193800</YOrigin><XYScale>37448250.484633707</XYScale><ZOrigin>-100000</ZOrigin><ZScale>10000</ZScale><MOrigin>-100000</MOrigin><MScale>10000</MScale><XYTolerance>0.0032808398950131233</XYTolerance><ZTolerance>0.001</ZTolerance><MTolerance>0.001</MTolerance><HighPrecision>true</HighPrecision><WKID>2253</WKID><LatestWKID>2253</LatestWKID></ProjectedCoordinateSystem> 8 FALSE LZW 3 TIFF TRUE continuous unsigned integer 255 20210402 13390200 20210402 13390200 Please see document listed in Supplemental Information section for details. 535 Griswold Street 20210402 ArcGIS Metadata 1.0 Please see document listed in Supplemental Information section for details. Offline Raster Dataset EnterpriseLandCover2010ImageService Southeast Michigan Council of Governments raster digital data This dataset is a composite of the 5-class Green Infrastructure Land Cover, Building Footprints, and the 2-class Full Canopy datasets. The model to compile them accounts for features below the canopy by overlaying the GI classification with the Full Canopy. This dataset will enable the user to know where buildings, pavement, or open space exist beneath canopy. It is designed to be ingested into American Forests' CITYgreen software.A Green Infrastructure map for the 7-county SEMCOG region in southeastern Michigan was developed to support land cover and land use modeling. 4-band ADS-80 digital aerial imagery at 1-meter resolution was classified to a Green Infrastructure Level 1 classification scheme with the following classes: 1) Impervious Surface, 2) Open Space, 3) Trees, 4) Urban: Bare and 5) Water. The image was classified using Classification and Regression Tree techniques (CART analysis) and raster modeling.A standard accuracy assessment was performed on the final land cover. The overall accuracy statistic is 97.45%. Adjusted from the SEMCOG composite raster file, the data was reprojected to the Oakland County projection. This dataset is a composite of the 5-class Green Infrastructure Land Cover, Building Footprints, and the 2-class Full Canopy datasets. The model to compile them accounts for features below the canopy by overlaying the GI classification with the Full Canopy. This dataset will enable the user to know where buildings, pavement, or open space exist beneath canopy. It is designed to be ingested into American Forests' CITYgreen software.A Green Infrastructure map for the 7-county SEMCOG region in southeastern Michigan was developed to support land cover and land use modeling. 4-band ADS-80 digital aerial imagery at 1-meter resolution was classified to a Green Infrastructure Level 1 classification scheme with the following classes: 1) Impervious Surface, 2) Open Space, 3) Trees, 4) Urban: Bare and 5) Water. The image was classified using Classification and Regression Tree techniques (CART analysis) and raster modeling.A standard accuracy assessment was performed on the final land cover. The overall accuracy statistic is 97.45%. Adjusted from the SEMCOG composite raster file, the data was reprojected to the Oakland County projection. Southeast Michigan Council of Governments (SEMCOG) 2011 Please see document listed in Supplemental Information section for details. Suite 300 Michigan Oakland County USA Full Canopy Tree Canopy Green Infrastructure Level 1 Landcover CityGREEN Composite transportation climatologyMeteorologyAtmosphere environment structure farming imageryBaseMapsEarthCover ISO 19115 Topic Categories transportation Michigan Full Canopy Oakland County Tree Canopy climatologyMeteorologyAtmosphere environment structure Green Infrastructure Level 1 Landcover USA CityGREEN farming Composite imageryBaseMapsEarthCover This data has enhanced access or data sharing restrictions according to the Enhanced Access Policy (http://www.oakgov.com/it/gis/documents/EnhancedAccessPolicy.pdf). Please contact the One Stop Shop at 248.858.0720 for more information. This data has enhanced access or data sharing restrictions according to the Geospatial Data Access, Distribution, and Use Policy (http://www.oakgov.com/it/gis/documents/GISDataPolicies.pdf). Please contact the One Stop Shop at 248.858.0720 for more information. Version 6.2 (Build 9200) ; Esri ArcGIS 10.7.1.11595 -83.557492 -82.868996 42.026149 42.458209 For supplemental metadata information look on the web at: http://www.oakgov.com/it/gis/documents/metadata/SEMCOG_composite_raster.pdf. 1 -83.694211 -83.074225 42.893623 42.426085 Last metadata review date: 20121031 Data have been independently reviewed with respect to the SEMCOG 2010 imagery, and corrections made as a result of that review. Overall Map Accuracy: 97.45% Minimum Mapping Unit for Assessment: 0.025 acres User's Accuracy: Impervious Surface: 100% Open Space: 99.35% Trees: 98.51% Urban: Bare: 79.45% Water: 100% Producer's Accuracy: Impervious Surface: 93.28% Open Space: 96.84% Trees: 100% Urban: Bare: 96.67% Water: 100% Sanborn's experience tells us that no single software tool can do all the things that are required to create a good automated classification. We have a methodology that incorporates the most appropriate tools for the right output. We have written procedures that combine these tools effectively so that the output from one can be input into the other properly. We use a combination of statistical and remote sensing software packages, such as Feature Analyst, Erdas Imagine, See5, and Definiens Professional and other tools as necessary. Sanborn will classify the imagery into impervious and pervious classes, and segment the imagery into image objects that are then classified into the selected classification scheme. The See5 software creates a classification algorithm that ingests training samples of the land cover, creates a hierarchical classification ruleset, and can be applied to the entire project area. Once the classification ruleset is developed it is fairly simple to apply it at little cost to a much larger area. The SEMCOG 2010 imagery was captured during substantially leaf-off conditions, therefore if the classification were done purely using this imagery, the full extent of the canopy would be underestimated. The LiDAR data will help to mitigate this effect, as it will show a larger canopy extent when processed to create a Digital Canopy Model. The LiDAR data were also captured during leaf-off conditions, so even though the use of the LiDAR DCM will help give a fuller picture of the canopy, there will remain some underestimation. To mitigate this problem, Sanborn proposes an option to add an additional canopy/non-canopy classification derived from leaf-on National Agricultural Imagery Program (NAIP) imagery. This classification, when used in conjunction with the classification specified by the RFP, will give users a true picture of the state of the region’s Green Infrastructure. There is always a degree of confusion in the spectral classification; hence spatial models will be used to clean up misclassifications based on adjacency and ancillary datasets. These models are most effective when working with misclassification of shadows. When classifying using 8-bit imagery, shadows are often confused with water and woody vegetation, and these can be corrected based on surrounding pixels. Sanborn uses spatial models which account for a feature’s context when deciding whether or not to reclassify it. In this dataset, the tree canopy is prioritized over impervious. This is the most straightforward way to classify an aerial image, since the aerial sensor “sees” the tree crown rather than what is beneath. Most of the previous land cover datasets were mapped in this manner with the exception of the Alliance of Downriver Watersheds classification. In that case, impervious surfaces were prioritized over canopy. Sanborn has retained from previous production a version in which tree canopy was prioritized, and this can be used to create a more consistent classification. To detect change in the landscape, Sanborn used the GeoCDX software application, developed by the University of Missouri. Using this methodology, spectral signatures and other features were compared between the two dates of imagery under consideration to create a potential change map. The change detection methodology compares several different aspects of both sets of imagery, 2008 and 2010. These attributes include, but are not limited to, spectral value, as well as spatial-spectral features, such as texture and linear correlation (good for finding roads, sidewalks, etc.). Data driven window parameters allow GeoCDX to consider pixel neighborhoods, which can aid in the discrimination of false change and true change, which most often occurs in larger, contiguous patches. Once the potential change map is complete, the area is broken into tiles, each tile is ranked according to the amount of potential change it contains, and the change-signatures are analyzed to group similar types of land cover change patterns. The change data are then ready for review by an imagery analyst and labeling of the change. The GeoCDX leverages an efficient and simple software web interface that allows an imagery analyst to cycle through the tiles according to the types of change and the amount of change detected by the routine. During this process, the analyst labels the type of change present in the tile for later use in creating the new classification. The amount of change can be customized and can be set for various minimum mapping units. Once the change labels are complete, Sanborn will use them to refine the potential change map, removing any “false positives” and classifying the true change to its proper 2010 class. This process is guided by the streamlined output provided by the GeoCDX system. The labels and delineation created in this step was thoroughly checked for quality and consistency by the project’s QC analyst. Full - Canopy Dataset 1) Initial canopy processing was performed using training data from land cover classification described above. To enhance the coverage of the canopy data regionally, additional training sites were selected where needed (where there were inconsistencies in NAIP imagery across areas) and these were processed separately. The resulting canopy was modeled into the baseline canopy map. These models were saved and used on nearby areas with similar imagery characteristics. 2) Digital canopy models based on LIDAR data were used to better define canopy from non-canopy and particularly for distinguishing large trees from shrubs. 3) Individual production areas were edgematched. 4) QC was performed and any issues were resolved 5) Data was provided to SEMCOG for review and any valid issues identified by SEMCOG were also addressed. 6) All counties were edgematched into a SEMCOG coverage and then clipped to the various boundaries as needed. Approximately 18% of the SEMCOG project area had been previously mapped using 2008 USGS imagery. The following process was used for the 2008 classification. 2008 ADS-40 imagery at 1 meter resolution was acquired for the area and imported into ERDAS Imagine for mosaicking and tiling. Image derivatives were made, including band ratios and texture of the NIR band. 6 of 18 mosaicked areas were selected for their diversity of cover types and training samples for each of the target cover types were delineated. These samples were then used to train the CART classifier (See5, Rulequest 2001), which resulted in a regression model that was applied to the mosaicked tiles. A preliminary classification was the result of this first step. Several spatial models were then created in AML and applied to the output of the CART classification. The output was a final automated classification that was ready for manual editing. At this point the classification was visually assessed against the imagery and errors were manually edited. An accuracy assessment protocol was run on the image to assess total and per class accuracy. Matt Vernier Sanborn Senior Remote Sensing Analyst 734.213.1060 734.213.1085 Suite 80 320 Miller Ave. Ann Arbor MI 48103 US mvernier@sanborn.com M-F 9-5 Compilation: 1. Since the land cover and full canopy datasets originated in raster format, building footprints had to be converted to raster prior to incorporation into the model. This was accomplished in ArcMap through Vector to Raster Conversion with a raster resolution of 3.28084 feet (1 meter). 2. A raster model was constructed with the following rules: 1. If Full Canopy = “Canopy” a. If BFP = “Building” then Output = “Trees: Impervious Understory” b. If LC = “Impervious” then Output = “Trees: Impervious Understory” c. If LC = “Open Space” then Output = “Trees: Grass Understory, Grass 50- 75%” d. If LC = “Trees” then Output = “Trees: Grass Understory, Grass > 75%” e. If LC = “Urban: Bare” then Output = “Trees: Grass Understory, Grass 50- 75%” f. If LC = “Water” then Output = “Trees: Impervious Understory” 2. If BFP = “Building” then Output = “Impervious: Buildings/Structures” 3. If LC = “Trees” then Output = “Trees: Grass Understory, Grass 50- 75%” 4. If LC = “Impervious” then Output = “Impervious: Paved: Drain to sewer” 5. If LC = “Open Space” then Output = “Open Space: Grass/Scattered Trees” 6. If LC = “Urban: Bare” then Output = “Urban: Bare” 7. If LC = “Water” then Output = “Water Area” The result of the model was the 8-class raster in ESRI GRID format Building Footprints 1) Vector building polygons were digitized heads up in a 2D environment using the following rules: 2) Buildings for all counties were captured even if there were existing datasets with building footprints for counties. For example, building footprints derived from LiDAR data were available for Washtenaw County but these were not used for this dataset due to the difference in methodology for this project. 3) Small residential buildings were digitized at the eve of the building roof whereas larger buildings with substantial height and offset from ground were captured at the base or footprint of the building structure. 4) Buildings were captured as an envelope delineating the periphery of the buildings and did not show different polygons for different heights. This was not possible due to the 2D data compilation and out of the scope. 5) Estimated building height was calculated using LiDAR data for each county – because LiDAR data vintage and specifications differed by counties, the building height attribution was done by county. 6) In order to estimate building heights, building footprint vectors were used to summarize a Digital Surface Model, or DSM. The DSM was created from the 2009-2010 LiDAR provided by SEMCOG by differencing the ground and non-ground points from each laser pulse, thereby deriving the height of the feature encountered by the pulse. Since those features would very often be trees, rather than structures underneath trees, high vegetation points were classified and removed from the LiDAR data prior to height derivation. 7) The summary routine returned the median height for a building, which was found to be a more robust measure of height than mean – the mean is usually influenced by large outliers, such as misclassified tree pixels. 8) Some buildings were not visible at all in the LiDAR, meaning the height could not be directly obtained from the DSM. This was particularly the case for buildings that were completely covered by high density trees or structures that were small enough to get few LiDAR points. Those buildings were assigned the median height of all buildings within their census block. If a census block contained too few buildings that were free of canopy cover, buildings without heights were assigned a global median height for the county. 9) Unusual building heights were flagged and reviewed. 10) All buildings where heights were calculated through interpolation using averages of census blocks, etc. were flagged for greater level of uncertainly and such information is captured in the data. For example, if a height extracted from the LIDAR data was incorrect based on the its’ ratio against the footprint shape area or if building seemed abnormally tall or it has a limited number of non-canopy LiDAR returns, the FLAG field was assigned a value of 1. This allowed the flagged buildings to be easily singled out for future improvements in the data for those buildings using other datasets such as tax and assessment databases, etc. 11) Both Sanborn and SEMCOG QCed the data and any identified issues were addressed. 12) Once all revisions were complete, the numbering system requested by SEMCOG was incorporated into the “IDLong” field in the building footprints attribute table. This field was populated using a model that insured that every object received a unique value while maintaining the requested numbering format. 1,000,000 range – Wayne County – not including Detroit 2,000,000 range – Oakland County 3,000,000 range – Macomb County 4,000,000 range – Washtenaw County 5,000,000 range – Monroe County 6,000,000 range – St. Clair County 7,000,000 range – Livingston County 8,000,000 range – Detroit 2 51351 3.280840 50232 3.280840 1 0 13304893.861130 340130.169570 13304893.861130 508604.584412 13469697.016017 508604.584412 13469697.016017 340130.169570 13387295.438574 424367.376991 Oakland_cmp_final.vat Table 5 OBJECTID OBJECTID OID 4 10 0 Internal feature number. Esri Sequential unique whole numbers that are automatically generated. SUBTYPECD Land Cover Class Code SEMCOG Land Cover workplan 2011 7 Urban: Bare SEMCOG Land Cover workplan 2011 8 Water Area SEMCOG Land Cover workplan 2011 3 Open Space - Grass/Scattered Trees: Grass cover > 75% SEMCOG Land Cover workplan 2011 5 Trees: Grass/turf understory: Ground cover > 75% SEMCOG Land Cover workplan 2011 4 Trees: Grass/turf understory: Ground cover 50% - 75% SEMCOG Land Cover workplan 2011 6 Trees: Impervious understory SEMCOG Land Cover workplan 2011 1 Impervious Surfaces: Buildings/ structures SEMCOG Land Cover workplan 2011 2 Impervious Surfaces: Paved: Drain to sewer SEMCOG Land Cover workplan 2011 SUBTYPECD Integer 4 10 0 CLASSNAME CLASSNAME String 20 0 0 dataset EPSG 5.3(9.0.0) Band_1 244.000000 0.000000 8 Band_2 244.000000 0.000000 8 Band_3 219.000000 0.000000 8