The NLCD 2001 is created by partitioning the U.S. into mapping zones. A total of 66 mapping zones were delineated within the conterminous U.S. based on ecoregion and geographical characteristics, edge matching features and the size requirement of Landsat mosaics. Mapping zone 51 encompasses whole or portions of several states, including the state of Michigan. Questions about the NLCD mapping zone 51 can be directed to the NLCD 2001 land cover mapping team at the USGS/EROS, Sioux Falls, SD (605) 594-6151 or mrlc@usgs.gov.
Conceptually, the descriptive tree is a classification tree generated by using the final minimum-map- unit land cover product (1 acre) as training data, and Landsat and other ancillary data as predictors. The goal of the descriptive tree is to summarize the effects of boosted trees (10 sequential classification trees) into a single condensed decision tree that can be used as a diagnostic tool for the classification process. This descriptive tree can be used to assess the relative importance of each of the input data sets on each land cover class. Such information may also be useful to customize the minimum-mapping-unit classification to meet a user's specific needs through raster modeling. Descriptive trees usually capture 60 to 80% of the information from the original land cover data.
The leaf or terminal nodes of the descriptive tree are assigned to sequential numbers (called node numbers) and mapped across the entire mapping zone on a pixel-by-pixel basis. These node numbers can then be matched with the various conditional statements associated with each respective terminal node. This spatial layer appears similar to a cluster map, but is the result of a supervised classification - not an unsupervised clustering. This node map can potentially be used as input by users to customize NLCD land cover, by linking the spatial extent of an individual node with the rules of the conditional statement.
The Land Cover spatial classification confidence data layer is provided to users to help determine the per-pixel spatial confidence of the NLCD 2001 land cover prediction from the descriptive tree. The C5 algorithm produces an estimate (a value between 0% and 100%) that indicates the confidence of rule predictions at each node based on the training data. This spatial confidence map should be considered as only one indicator of relative reliability of the land cover classification, rather than a precise estimate. Users should be aware that this estimate is made based on only training data, and is derived from a generalized descriptive decision tree that reproduces the final land cover data. However, this layer provides valuable insight for a user to determine the risk or confidence they choose to place in each pixel of land cover.
A logic statement from a descriptive tree classification describes each classification rule for each classified pixel. An example of the logic statement follows:
IF tasseled-cap wetness > 140 and imperviousness = 0 and canopy density < 4, then classify as Water
This logic file can be used in combination with the spatial node map to identify classification logic and allow modifications of the classification based on user's knowledge and/or additional data sets.
Additional information may be found at <http://www.mrlc.gov/mrlc2k_nlcd.asp>.
Field data were collected by MIDNR foresters and biologists in the summers of 1999-2001 drawn upon 1:15840 aerial photos. Quantitative information was collected with the photos relating to species composition in the understory, overstory, and ground cover. Training data collection was stratified over the entire state with regard to ecoregion, sub-ecoregion, and TM path/row. Within each "eco-scene", we attempted to collect examples of each of the land cover classes present. The information collected was sufficient to classify the sites, both natural and constructed, to an Anderson Level IV class, though the IFMAP classification was designed to classify land cover to level III.
During this time, the Landsat scenes selected for the classification were georeferenced to the MIRIS (Michigan Resource Inventory System) base roads and balanced and mosaicked. Three mosaics were produced: spring, summer, and fall, for the Upper and Lower Peninsulas. There were several areas of clouds, both in the Lower and Upper Peninsulas, that needed to be patched by imagery acquired much earlier, as early as 1992. These areas are small. Once the mosaics were completed, the training sites were digitized and checked for consistency on each of the dates. The signature data were then extracted to form the training site database for each date. The summer mosaics were classified to Anderson level 1 using statistical cluster analysis with ERDAS Imagine and S-plus statistical software. Once the result was satisfactory, the seven level 1 classes were separately classified to Anderson level 2 using the same techniques. Some modeling with ancillary data was necessary, primarily to isolate lowland types from upland types. The ancillary data used consisted of a weighted combination of NWI, a presettlement vegetation layer, and the MIRIS hydrology layer.
Acquisition dates of Landsat ETM+ (TM) scenes used for land cover classification in zone 51 are as follows:
SPRING-
Index 1 for Path 20/Row 29 on 05/04/02 = Scene_ID 7020029000212450
Index 1 for Path 20/Row 30 on 05/04/02 = Scene_ID 7020030000212450
Index 1 for Path 20/Row 31 on 05/04/02 = Scene_ID 7020031000212450
Index 2 for Path 21/Row 28 on 04/27/00 = Scene_ID 5021028000011810
Index 3 for Path 21/Row 29 on 04/30/01 = Scene_ID 5021029000112010
Index 3 for Path 21/Row 30 on 04/30/01 = Scene_ID 5021030000112010
Index 2 for Path 21/Row 31 on 04/27/00 = Scene_ID 5021031000011810
Index 4 for Path 22/Row 27 on 04/16/02 = Scene_ID 7022027000210650
Index 5 for Path 22/Row 28 on 04/26/00 = Scene_ID 7022028000011750
Index 5 for Path 22/Row 29 on 04/26/00 = Scene_ID 7022029000011750
Index 4 for Path 22/Row 30 on 04/16/02 = Scene_ID 7022030000210650
Index 6 for Path 22/Row 31 on 03/25/00 = Scene_ID 7022031000008550
Index 7 for Path 23/Row 28 on 04/26/03 = Scene_ID 7023028000311650
Index 8 for Path 23/Row 29 on 05/19/00 = Scene_ID 7023029000014050
Index 9 for Path 23/Row 30 on 04/25/00 = Scene_ID 5023030000011610
Index10 for Path 24/Row 27 on 04/24/00 = Scene_ID 7024027000011550
Index10 for Path 24/Row 28 on 04/24/00 = Scene_ID 7024028000011550
Index11 for Path 25/Row 27 on 04/23/00 = Scene_ID 5025027000011410
Index12 for Path 25/Row 28 on 05/02/03 = Scene_ID 5025028000312210
LEAF ON (Summer)-
Index 1 for Path 20/Row 29 on 07/18/03 = Scene_ID 5020029000319910
Index 1 for Path 20/Row 30 on 07/18/03 = Scene_ID 5020030000319910
Index 2 for Path 20/Row 31 on 07/15/02 = Scene_ID 5020031000219610
Index 3 for Path 21/Row 28 on 07/30/02 = Scene_ID 7021028000221150
Index 4 for Path 21/Row 29 on 06/25/01 = Scene_ID 7021029000117650
Index 5 for Path 21/Row 30 on 07/14/02 = Scene_ID 7021030000219550
Index 6 for Path 21/Row 31 on 06/09/01 = Scene_ID 7021031000116050
Index 7 for Path 22/Row 27 on 07/16/03 = Scene_ID 5022027000319710
Index 7 for Path 22/Row 28 on 07/16/03 = Scene_ID 5022028000319710
Index 7 for Path 22/Row 29 on 07/16/03 = Scene_ID 5022029000319710
Index 8 for Path 22/Row 30 on 07/02/01 = Scene_ID 7022030000118350
Index 8 for Path 22/Row 31 on 07/02/01 = Scene_ID 7022031000118350
Index 9 for Path 23/Row 28 on 06/23/01 = Scene_ID 7023028000117450
Index 9 for Path 23/Row 29 on 06/23/01 = Scene_ID 7023029000117450
Index10 for Path 23/Row 30 on 07/09/01 = Scene_ID 7023030000119050
Index11 for Path 24/Row 27 on 07/29/00 = Scene_ID 7024027000021150
Index12 for Path 24/Row 28 on 07/11/02 = Scene_ID 5024028000219210
Index13 for Path 25/Row 27 on 07/04/00 = Scene_ID 7025027000018650
Index14 for Path 25/Row 28 on 07/05/03 = Scene_ID 5025028000318610
Index15 for Path Derived/Row Various on Cloud Scene_ID Replacement
LEAF-OFF (Fall)-
Index 1 for Path 20/Row 29 on 11/04/99 = Scene_ID 7020029009930850
Index 2 for Path 20/Row 30 on 10/21/00 = Scene_ID 7020030000029550
Index 1 for Path 20/Row 31 on 11/04/99 = Scene_ID 7020031009930850
Index 3 for Path 21/Row 28 on 09/29/01 = Scene_ID 7021028000127250
Index 3 for Path 21/Row 29 on 09/29/01 = Scene_ID 7021029000127250
Index 4 for Path 21/Row 30 on 10/20/00 = Scene_ID 5021030000029410
Index 5 for Path 21/Row 31 on 10/28/00 = Scene_ID 7021031000030250
Index 6 for Path 22/Row 27 on 10/19/00 = Scene_ID 7022027000029350
Index 6 for Path 22/Row 28 on 10/19/00 = Scene_ID 7022028000029350
Index 6 for Path 22/Row 29 on 10/19/00 = Scene_ID 7022029000029350
Index 6 for Path 22/Row 30 on 10/19/00 = Scene_ID 7022030000029350
Index 6 for Path 22/Row 31 on 10/19/00 = Scene_ID 7022031000029350
Index 7 for Path 23/Row 28 on 10/10/00 = Scene_ID 7023028000028450
Index 8 for Path 23/Row 28 on 10/24/99 = Scene_ID 7023028009929750
Index 9 for Path 23/Row 29 on 10/24/99 = Scene_ID 7023029009929750
Index 9 for Path 23/Row 30 on 10/24/99 = Scene_ID 7023030009929750
Index10 for Path 24/Row 27 on 10/31/99 = Scene_ID 7024027009930450
Index10 for Path 24/Row 28 on 10/31/99 = Scene_ID 7024028009930450
Index11 for Path 25/Row 27 on 10/11/01 = Scene_ID 7025027000128450
Index11 for Path 25/Row 28 on 10/11/01 = Scene_ID 7025028000128450
Index12 for Path Derived/Row Various on Cloud Scene_ID Replacement
Once the level 2 map was complete, the map was divided by TM scene area and level 2 class. To achieve the level 3 classes different imagery dates were used depending upon the class. For instance, fall, or senescence, imagery was used to differentiate the level 3 broadleaf species, since the turning of the leaves aids in separation of broadleaf types. Differences between spring and summer images were used to classify some wetland types using differences in water levels. Differentiation of urban types (high-intensity/low-intensity) was often made using a thresholding technique in Landsat band 1 (blue).
Upon completion of the level 3 product, the map was disseminated among DNR foresters and biologists for field verification and checked for consistency against an older DNR forest database. The former step was very instructive in pointing out regions and sites in which the map accuracy was good and where additional work was needed. Using this data in conjunction with forest predictive models guided further supervised reclassification of natural areas.
From the IFMAP classification, a level 2 preliminary classification was made from a crosswalk developed between the IFMAP and C-CAP classification schemes. The IFMAP training sites were also recalculated under the C-CAP scheme. On average, 11% of the training sites did not directly crosswalk due to different land cover component thresholds. These training sites were traced to the pixels that they labeled, and those pixels were given the new C-CAP classification. This step produced a preliminary C-CAP basemap. Since the Southern Lower Peninsula of Michigan was completed for the IFMAP project prior to the Michigan C-CAP project coming up for bid, the required 3-mile buffer into Indiana and Ohio was not made. The buffer had to be classified, and, in anticipation of the edgematching award, the decision was made to extend the buffer to 10 miles.
To accomplish this, the triple-date TM imagery used to classify Southern Michigan originally was subset to cover an area at least 50 miles into Michigan and 10 miles into Indiana and Ohio. These subsets were then classified using an unsupervised routine (ISODATA), and, beginning with the summer date, were summarized with the C-CAP base map. The summaries were then used to label the clusters when there was a clear majority cover type. When a clear majority did not exist, another date was chosen to re-cluster unlabeled pixels. After the three dates were used few pixels remained unclassified, and those were clustered with Southern Michigan training sites, or manually edited. Manual editing was used as well to edgematch the extension to Southern Michigan. At this point, NOAA CSC staff visited the Ann Arbor office, and performed an accuracy assessment by gathering points throughout the state, as discussed in the accuracy portion of the metadata.
Once the first assessment of the draft map was complete, comments generated from drop points collected with AA points were incorporated into the map, and the map was sent for another assessment. The Upper Peninsula was assessed separately from the Lower Peninsula, and iterations of comments and modifications continued for several rounds, until the target accuracy of 85% overall and 80% minimum class accuracy was achieved. Several classes, such as Palustrine Aquatic Bed and Unconsolidated Shore, did not have enough points for statistical significance. Scrub Shrub in the Upper Peninsula had not enough points for significance. There was a fair amount of confusion between pasture grassland and cultivated land, especially in the Upper Peninsula. This required a number of iterations to create a satisfactory representation. Much of the confusion is believed to arise from phenological variation and crop rotation.
Large urban areas from the IFMAP classification had a different appearance than those in other C-CA products. The decision was made to filter the Michigan C-CAP urban areas to bring them into line with the products already completed. The model operated on contiguous areas of urban land cover greater than 2000 pixels After the classification was accepted by NOAA CSC, it was edgematched to classifications of the surrounding states. Edgematching involved using the overlaps between the state classifications and locating a path in that zone that would not produce a localized seam line. Subsequently, the overlap zone, including a zone up to 10 miles or more from the state border was edited as with the Ohio extension to equalize the classifications.
After the above classification, the C-CAP team at the NOAA Coastal Services Center performed a retrofit process to bring the data set into compliance with NLCD 2001 standards. This involved using the NLCD impervious surface for Developed Categories (includind adding a Medium Intensity and Open Space Developed) and the inclusion of a Pasture/Hay class). To develop the Pasture/Hay category, the original Grassland and Bare categories were subset for analysis. Training data for Pasture/Hay were derived from the 1992 NLCD and the Integrated Forest Monitoring Assessment and Prescription (IFMAP) data set, produced for the Michigan State Department of Natural Resources. A CART classification, based upon this IFMAP training data, was applied on the full zone. Hand editing was then performed to clean these newly classed areas. Similar classification procedures and editing was performed on all land cover categories in order to check and improve their consistency. These edits removed much of the speckle associated with the above produced classification while retaining the accuracy of that original product. The Developed classes were created through the incorporation of the masked NLCD Impervious surface. High, Medium, and Low Intensity Developed classes were included where the percent impervious value ranged from 80-100%, 50-79%, and 20-49% respectively. The Open Space Developed was included where the percent impervious value was from 1-19%, did not overwrite a forested class, and was in a spatially dense urban area (as determined through a focal filtered impervious mask image).
The completed single pixel product was then generalized to a 1 acre (approximately 5 ETM+ 30 m pixel patch) minimum mapping unit product using a "smart eliminate" algorithm. This aggregation program subsumes pixels from the single pixel level to a 5- pixel patch using a queens algorithm at doubling intervals. The algorithm consults a weighting matrix to guide merging of cover types by similarity, resulting in a product that preserves land cover logic as much as possible.