| Section 1 | Overview | |||
| Originator | Remote Sensing and Geospatial Analysis Laboratory, University of Minnesota - Version 2 | |||
| Title | Minnesota Land Cover Classification and Impervious Surface Area by Landsat and Lidar: 2013 update - Version 2 | |||
| Abstract | This is a 15-meter raster dataset of a land cover and impervious surface classification for 2013, level two classification. The classification was created using a combination of multitemporal Landsat 8 data and LiDAR data with Object-based image analysis. By using objects instead of pixels we were able to utilize multispectral data along with spatial and contextual information of objects such as shape, size, texture and LiDAR-derived metrics to distinguish different land cover types. While OBIA has become the standard procedure for classification of high resolution imagery we found that it works equally well with Landsat imagery. For the objects classified as urban or developed, a regression model relating the Landsat greenness variable to percent impervious was developed to estimate and map the percent impervious surface area at the pixel level.
This dataset was funded by the the Minnesota Environment and Natural Resources Trust Fund (ENRTF). | |||
| Purpose | Land cover information offers important inputs to local, regional, and state land use planning and natural resource monitoring.
| |||
| Time Period of Content Date | 2013 | |||
| Currentness Reference | A multitemporal composite of Landsat imagery from the summer of 2013 and 2014 and fall of 2013, lidar data of 2008, 2009, 2010, 2011, and 2012 were classified. | |||
| Progress | Complete | |||
| Maintenance and Update Frequency | As needed | |||
| Spatial Extent of Data | Minnesota | |||
| Bounding Coordinates | -97.23 -89.53 49.37 43.5 | |||
| Place Keywords | Minnesota | |||
| Theme Keywords | Landsat, Lidar,object based image classification, land cover, impervious surface area. | |||
| Theme Keyword Thesaurus | Landsat: an earth-orbiting satellite that acquires digital multispectral imagery at 30-meter spatial resolution. Lidar: Light detection and ranging. Lidar data provide information on height and elevation. | |||
| Access Constraints | The Remote Sensing and Geospatial and Analysis Laboratory, University of Minnesota, has attempted to produce accurate maps, statistics and information of land cover and impervious surface area. However, it makes no representation or warranties, either expressed or implied, for the data accuracy, currency, suitability or reliability for any particular purpose. Although every effort has been made to ensure the accuracy of information, errors and conditions originating from the source data and processing may be present in the data supplied. Users are reminded that all geospatial maps and data are subject to errors in positional and thematic accuracy. The user accepts the data -- as is -- and assumes all risks associated with its use. The University of Minnesota assumes no responsibility for actual or consequential damage incurred as a result of any user's reliance on the data. The data are the intellectual property of the University of Minnesota. | |||
| Use Constraints | This data may be used for educational and non-commercial purposes, provided proper attribution is given. Secondary distribution of the data is permitted, but not supported by the University of Minnesota. By accepting the data, the user agrees not to transmit this data or provide access to it or any part of it to another party unless the user includes with the data a copy of this disclaimer. | |||
| Contact Person Information | Joe Knight,
Associate Professor Remote Sensing and Geospatial Analysis Lab, University of Minnesota 1530 Cleveland Avenue North St. Paul, MN 55108 Phone: 612-625-5354 Fax: 612-625-5212 Email: jknight@umn.edu | |||
| Browse Graphic | None available | |||
| Associated Data Sets | ||||
| Section 2 | Data Quality | |||
| Attribute Accuracy | The data set has an overall classification accuracy of 97% for level 1 and 96% for level 2 land cover classifications. | |||
| Logical Consistency | ||||
| Completeness | Data provides complete coverage of the State of Minnesota. | |||
| Horizontal Positional Accuracy | RMS error of the Landsat data is less than 7.5 meters (0.25 Landsat pixel). The horizontal accuracy of the lidar data meets or exceeds 0.6 m RMSE. | |||
| Vertical Positional Accuracy | The vertical accuracy of the lidar data meets or exceeds 12.5 cm RMSE. | |||
| Lineage | Mosaic of Landsat 8 Images: The Landsat imagery consisted of several dates (June 24, 2013, July 3,7,21,23,28, 2013, August 22, 2013, September 16,23,25,30, 2013, October 7,9,11,16,18,27, 2013, May 26,28,30, 2014, June 4,6,8,13, 2014) of 7 multispectral bands, 30-meters and 2 bands (Thermal Infrared Sensor)10-11: collected at 100 meters but resampled to 30 meters to match the multispectral bands. Reprojected to UTM Zone 15N. Further information about Landsat is available at landsat.usgs.gov/. The following spectral indices were derived from the summer and fall landsat data: Normalized Difference Vegetation Index (NDVI, Transformed Normalized Difference Vegetation Index (TNDVI), Ratio Vegetation Index (RVI: Red/NIR), Infrared divided by Red (NIR/Red), and Modified Soil Adjusted Vegetation Index 2 (MSAVI2).
Lidar Data collected for Minnesota between 2006-2012 and made publicly available by the Minnesota Department of Natural Resources (DNR). For more information, the 2011 lidar metadata are available at: www.mngeo.state.mn.us/chouse/metadata/lidar_metro2011.html. The DNR-provided 1-meter bare earth DEM was also used to create additional lidar-derivative layers, such as Compound Topographic Index (CTI), slope, and normalized Digital Surface Model (nDSM). The U-Spatial - Support for Spatial Research, University of Minnesota provided a Digital surface model (1m resolution) for the state of Minnesota. Reference Data: Reference data used for classifier training and accuracy assessment were created using high-resolution aerial photos available through the MnGeo Geospatial Image Service (www.mngeo.state.mn.us/chouse/wms/geo_image_server.html), 3-band 1-meter, 4 bands meter natural color and color-infrared USDA National Agricultural Program Imagery (NAIP) and Spring Aerial Imagery 4-band leaf-off (2009-2014). For classes, such as extraction and agriculture, ancillary datasets were used to identify characteristic areas for training. Image Classification: We mapped 11 land cover classes including Deciduous Forest, Conifer Forest, Mixed Forest, Grassland, Hay and Pastures, Row Crops, Extraction, Urban/Developed, Emergent Wetland, Forested/Shrub Wetland, and open water (ponds, lakes and rivers). We employed an Object Based Image Analysis (OBIA) approach by creating rule sets for the seven ecoregions within the State of Minnesota. We used the Cognition Network Language (CNL) within the software package Definiens eCognition Developer version 9.1.0 was used to develop the rule sets. Each rule set was developed through a trial and error process using small subset areas. We used a divide and conquer approach, which is a multiscale iterative technique where objects vary in size, shape, and spectral attributes. Although the two major steps performed in the rule set development were the creation of objects and the classification of those objects, additional steps were required for the classification of each object to be assigned to the class of interest. Each rule set consisted of six key components: (a) image processing, (b) Masking out objects using a vector layer, (c) segmentation (c) classification, (d) export operations, and (e) cleanup operations. In the image processing phase, we executed the following tasks: calculation of the normalized Digital Surface Model (nDSM) = DSM-DEM, creation of nDSM filtered layer, and a computation of a nDSM slope layer.Given the availability of the recently available new high resolution and accurate National Wetlands Inventory (NWI) 2010-2015 data for the southern part of the state; we masked all the water bodies including wetland types, lakes, rivers and ponds. We used the Chessboard segmentation algorithm in eCognition for masking out these features. Also, for the entire state we masked the roads, railroads, and airports; vector layers were obtained from the Minnesota Department of Transportation (MnDOT). For more information about the updated NWI data, see: http://www.dnr.state.mn.us/eco/wetlands/nwi_proj.html. We classified preliminary objects (short versus tall) into temporary classes. The nDSM layer was used to separate short objects from tall objects. We executed the multi-threshold segmentation algorithm with the following parameters: a threshold value of equal or greater than 3 for tall object (potential tree canopy) and anything less than 3 for short objects. Previous parameters were determined after several trial-and-error experiments and a detailed visual assessment for separating short versus tall classes. The temporary tall objects were merged and a new segmentation step was applied to these temporary objects. For this step, a multiresolution segmentation algorithm was employed for segmentation with the following parameters: Image layer (Green, Red and NIR leaf-on only), scale parameter: 100, shape: 0.1 and compactness: 0.3. The refined tall objects were classified as deciduous, conifer or mixed forest classes using the Random Forest (RF) algorithm within eCognition. The RF is an ensemble learning method for classification that operates by constructing multiple decision trees. Each tree is grown from different random subsamples of the training data and during the split selection process uses a subsample of the available features. It allows for the use of a large number of features or variables and identifies the important predictors. We used all the features layers values from the spectral data to train the algorithm and classify the objects. Finally, the remaining objects that were not classified as a type of water body or forest were merged and a new segmentation was performed. The following parameters were used for this segmentation: spectral bands (Green leaf-off and leaf-on, NIR leaf-off and leaf-on, Red leaf-off and on, Short-wave Infrared (SWIR)1 leaf-off and leaf-on and SWIR2 leaf-off and leaf-on); Scale parameter: 150, shape: 0.1 and compactness: 0.3. These new objects were classified using the RF algorithm as one of the following classes: Grassland, Hay and Pastures, Row Crops, Extraction, Urban/Developed and water bodies (wetlands or open water) for the part of the state that didn’t have the updated NWI. These features characteristics were used to train the RF algorithm: spectral and lidar data values including min. and max. pixel values, means and standard deviations of individual bands; imagery brightness mean and differences; geometry including asymmetry, compactness, density, rectangular fit, roundness, area, length, number of pixels, and shape index; texture, including, homogeneity, contrast, dissimilarity, mean, and standard deviation; relations to neighbor objects including relative border to each class. In the export operation phase, we exported the final classes into raster and vector formats. Estimation and Mapping of Impervious Surface Area for the objects classified as urban or developed, was done by running a regression model relating to the Landsat greenness variable to percent of imperviousness. This was developed to estimate and map the percent impervious surface area at the pixel level. Greenness is sensitive to the amount of green healthy vegetation and inversely related to the amount of impervious surface area. Impervious was treated as a continuous variable from 1 to 100 percent, but can be grouped by the user into classes such as 1 - 10, 11- 25, 26 - 40, 41 - 60, 61 - 80, and 81 - 100 percent impervious. Accuracy Assessment was evaluated by comparing the classification results to an independent stratified random reference set of 19,532 points and reporting the error matrix and statistics derived from it. These included overall accuracy, user and producer accuracies for each class, and Kappa statistic. | |||
| Section 3 | Spatial Data Organization (not used in this metadata) | |||
| Section 4 | Coordinate System | |||
| Horizontal Coordinate Scheme | Universal Transverse Mercator | |||
| UTM Zone Number | 15 | |||
| Horizontal Datum | NAD83 | |||
| Horizontal Units | meters | |||
| Cell Width | 15 | |||
| Cell Height | 15 | |||
| Section 5 | Attributes | |||
| Overview | See table below for the Level 1 and Level 2 Land Cover class values and descriptions.
Descriptions of the lakes, streams and wetlands classes can be found at: www.npwrc.usgs.gov/resource/wetlands/classwet/index.htm. | |||
| Detailed Citation | ||||
| Table Detail: | ||||
| Classification Scheme for Level 1 & 2 Land Cover | |||
| Field Name | Valid Values | Definition | Definition Source |
|---|---|---|---|
| Urban/Developed | enumerated | ||
| 1-100 | Impervious (%) | Roads, parking lots, and building rooftops | |
| Wetlands | enumerated | ||
| 101 | Emergent Wetlands | Palustrine PEM | |
| 102 | Forested & Shrub Wetlands | Palustrine PFO and PSS | |
| Open Water | enumerated | ||
| 103 | Lakes, Ponds and rivers | Lacustrine L1, L2, Palustrine PAB, PUB, PUS, Riverine | |
| Extraction | enumerated | ||
| 104 | Extraction | Pits, Quarries and Mines | |
| Forest | enumerated | ||
| 105 | Conifer Forest | White Pine, Red Pine, Balsam Fir, Jack Pine, White & Black Spruce, White Cedar | |
| 106 | Deciduous Forest | Oak, Red Maple, Black Maple, Paper Birch, Black Ash, Aspen, Silver Maple | |
| 107 | Mixed Forest | Mixtures of Conifer and Deciduous | |
| Managed Grass/Natural Grass | enumerated | ||
| 108 | Managed Grass/Natural Grass | Golf courses, Parks, Natural grass and Herbaceous vegetation | |
| Agriculture | enumerated | ||
| 109 | Hay & Pasture | Alfalfa and other hay and pasture | |
| 110 | Row Crops | Annual crops such as corn, soybean, wheat, oats, barley and perennial crops | |
| Section 6 | Distribution | |||
| Publisher | ||||
| Publication Date | 01/29/2016 | |||
| Contact Person Information | Joe Knight,
Associate Professor Remote Sensing and Geospatial Analysis Lab, University of Minnesota 1530 Cleveland Avenue North St. Paul, MN 55108 Phone: 612-625-5354 Fax: 612-625-5212 Email: jknight@umn.edu | |||
| Distributor's Data Set Identifier | MN Land Cover Classification, 2013 | |||
| Distribution Liability | The data are provided as is and without any warranty to their performance, merchantability, or fitness for any particular purpose. The University of Minnesota does not represent or warrant that the data or documentation are error-free, complete, or current. The user should consult the data documentation to determine its limitations and accuracy. The user is responsible for any consequences resulting from use of the data or reliance on it. The entire risk of the results of the use of the data are assumed by the user. | |||
| Ordering Instructions | Contact Joe Knight at the University of Minnesota | |||
| Online Linkage | None available | |||
| Section 7 | Metadata Reference | |||
| Metadata Date | 01/29/2016 | |||
| Contact Person Information | Joe Knight,
Associate Professor Remote Sensing and Geospatial Analysis Lab, University of Minnesota 1530 Cleveland Avenue North St. Paul, MN 55108 Phone: 612-625-5354 Fax: 612-625-5212 Email: jknight@umn.edu | |||
| Metadata Standard Name | Minnesota Geographic Metadata Guidelines | |||
| Metadata Standard Version | 1.2 | |||
| Metadata Standard Online Linkage | http://www.mngeo.state.mn.us/committee/standards/mgmg/metadata.htm | |||