One of the most effective ways to map land use/land cover is through the use of remote sensing imagery collected from satellites and aircraft. Remote sensing satellites orbit at hundreds of miles above the earth continually imaging the surface and transmitting the images back to ground stations for use by the research community. This technology is an excellent medium for monitoring the condition of land throught the globe. Photography taken from airplanes is also useful, especially where greater detail of the land surface is needed. This combination 2000 land use and 1995 land cover data utilized SPOT and Landsat Thematic Mapper (TM) satellite image data. The satellite platforms' sensors have a 10 x 10 meter and 30 x 30 meter (approximately 100 x 100 foot) spatial resolution per pixel, respectively. The TM data were also utilized for the 1984 land cover data products in this series (1972 land cover's base image data was the Landsat Multispectral Scanner which has a spatial resolution of 80 meters and fewer image channels/bands). Advanced computer processing techniques allows the images to be combined with other environmental data sets to map land cover. Mapping land use requires visual interpretation by experienced image interpreters and is a time consuming labor intensive process.
This product is one in a series of four land cover grids processed for the NY / NJ Highlands Regional Assessment Update, the others being 1972, 1984, 1995.
New York/New Jersey Regional Assessment Update (2001-02):
The Center for Remote Sensing and Spatial Analysis (CRSSA), Rutgers University, participated in the NY-NJ Highlands Regional Assessment Update (Fiscal Year 2001), directed by the USDA Forest Service. The major question the New York-New Jersey Highlands Regional Assessment Update addressed is, what are the implications of continued land use change patterns on Highlands resources? The assessment focused on land use/land cover change, forest and watershed integrity, hydrologic systems, biodiversity, recreation and open space, population trends and projected future growth.
The study area covered the New York and New Jersey portions of the Highlands physiographic province. All municipalities that are wholly or partially included within the Highlands physiographic boundary, as delineated by the USDA Forest Service, were included in the study.
1. The Grant F. Walton Center for Remote Sensing and Spatial Analysis (CRSSA), Cook College, Rutgers University, disseminates this data layer as-is and makes no warranty, expressed or implied, nor does the fact of distribution consitute such a warranty. 2. The user should be aware that the data set is not intended for site specific analysis, however, its intention is to provide the end user with a regional-scale raster grid representation of a particular theme for which it was used in the Highlands Regional Study. 3. CRSSA, Rutgers University, will not be held responsible to further maintain the disseminated data, nor to provide the data in different formats other than its current availability. 4. Any maps, publications, reports or any other type of document produced as a result of an associated project utilizing this raster grid will credit the original author(s) as listed in the metadata (citation) as well as the Center for Remote Sensing and Spatial Analysis (CRSSA), Rutgers University.
A variety of efforts were undertaken to verify the accuracy of the land cover data sets. The 1995 NY-NJ Highlands land cover data set was chosen as the baseline for analyzing change both prior to (i.e., 1972 and 1984) and post 1995 (i.e., 2000). The NY-NJ Highlands land cover data set was a modification and extension of the 1994/1995 New Jersey Land Cover Mapping Project undertaken by Rutgers University Center for Remote Sensing & Spatial Analysis (CRSSA). This 1995 data set was the subject of an intensive "ground truthing" field campaign that was undertaken at two time periods: 1) pre-classification - during the fall of 1994 and winter/spring months of 1995; and 2) post-classification - during June 1997 and May 2000. As this field campaign was undertaken only in New Jersey, additional points were "ground truthed" via aerial photo interpretation to assess the accuracy of the New York portion of the 1995 land cover data set. An additional field campaign was undertaken in the Fall of 2002 to evaluate the 2000 land cover change data. More details on the accuracy assessment methods and the results are provided below.
While our accuracy assessment generally showed a high agreement between the land use/cover maps and the field and photo checked reference data, there are several caveats that must be noted.
1. In closer examination of the 1995 to 2000 change data, we noted that the rasterization process resulted in slight boundary shifts between the 1995 and 2000 land use maps. When the 1995 and 2000 maps were differenced to map change areas, there was a significant number of isolated pixels left along the edges of Developed land use areas. This resulted in a change of approximately 2,750 acres of additional Developed land or 10.6% of the mapped change between 1995 and 2000 (i.e., 2,750 out 25,800 acres). It is unclear what portion of this 10% change is real or an artifact of the GIS processing.
2. Due to the limited spatial resolution (i.e. a 30x30 meter grid cell size), the mapped land cover data are not adequate for site level specificity. The accuracy assessment procedure evaluates clusters of homogeneous land cover of approximately 2 acres in size (i.e., 3 x 3 pixel blocks); areas smaller than 2 acres were not explicitly evaluated. The land cover data were not "filtered" (i.e., individual isolated pixels eliminated and replaced by the surrounding majority land cover type) resulting in a number of single isolated pixels (this "salt and pepper" is typical of image based classification). While not explicitly evaluated, we would expect that the classification accuracy of these individual pixels would be significantly lower than the 92-93% accuracy reported above. However when looked at in sum across the entire region, these individual pixels do not add up to a significant acreage. For example, the single isolated pixels of the "Developed" category (i.e., combined 111-114) for the 2000 Level I land cover map represent approximately 950 acres or approximately less than 0.3% of the developed land area mapped. We did not "filter" the data as this process would remove small patches or discontinuous linear features (e.g., riparian wetlands) that may be quite real. Regardless, due to the 30 m grid cell, thin linear features such as roads and riparian wetlands are not well detected and mapped in our Landsat TM based land cover mapping effort.
In summary, our accuracy assessment shows high overall accuracy and kappa statistics for the 1995 Level I land cover map (i.e., 92-93% agreement in New York and New Jersey, respectively). Likewise, the 1995 to 2000 land use change data also showed a high level of agreement (89%) with field checked reference data. While, these maps are not completely error free, we conclude that the land cover data are of sufficient quality that they can be used in further regional to landscape scale analysis with confidence.
Pre-classification field checking for 1995 NJ land cover data
During the first field campaign, 314 field sites were visited to serve as accuracy assessment (240 points in the South Jersey study area and 73 points in the North Jersey study area). These sites were visited by RUCRSSA personnel; one person was primarily responsible for this effort to ensure consistency. These field reference sites were chosen using a stratified random sampling technique. The initial stratification was developed using the 1986 ITU data. A differential Global Positioning System (GPS) receiver was used to georeference the field reference sites locations. Field notes and slides were taken for each field reference point. High altitude color infrared aerial photography (acquired March 1991 and March 1995) was used to quality check the ground reference data in the accuracy assessment. In some cases, the original GPS-derived location was moved slightly to a new location to more closely correspond with the field note description and ground photo.
Post-classification field checking for 1995 NJ land cover data
Two additional field campaigns were undertaken subsequent to the classification process. The southern New Jersey study area was assessed during June 9-12, 1997 and the northern New Jersey study area during May 15-17, 2000. A validation team from the NOAA Coastal Services Center participated in each data verification exercise. The team was equipped with 2 portable color laptop computers linked to real-time differentially corrected GPS receivers. The field station runs software that supports the classified data as a raster background with the road network as a vector overlay with a simultaneous display of live GPS coordinates. Accuracy assessment points were generated by NOAA Coastal Services Center personnel with ERDAS IMAGINE software using a stratified random sample. To reduce problems in locating the field reference sites due to GPS positional inaccuracy or on-the-ground observer classification indecision due to spatial heterogeneity, an additional criteria was that field reference sites were located in areas that were homogeneous within a 3x3 pixel neighborhood. To make acquisition of the field reference data more practical, a sixteen pixel buffer area around roads (i.e., 8 pixels on each side of the road) was created. Several thousand random points were generated. 606 points were field checked in the south Jersey study area. 491 points were field checked in the North Jersey study area. Due to the absence of concurrent field or photo reference data, no attempt was made to independently assess the accuracy of the 1972 or 1984 time period image maps.
Post-classification photo checking for 1995 NY land cover data
Due to the several year lapse in time between the date of the 1995 imagery and the 2001-2002 time period of the Highlands study and the accelerated pace of the study itself, an intensive field campaign to validate the 1995 NY land cover data set was not deemed feasible. Instead, the 1994-1995 color infrared digital orthophotography (1 m ground resolution cell, acquired during the spring leaf-off month of March) was used as "ground truth" as this image data set was reasonably concurrent in time with the satellite imagery. Accuracy assessment points were generated by CRSSA personnel with ERDAS IMAGINE software using a stratified and equalized random sampling strategy. 500 homogenous 3x3 pixel neighborhoods were randomly selected, located on-screen on the digital orthophotography and then assigned to the appropriate land cover category (at Level II) based on the image interpreter's best judgement. Due to the absence of concurrent field or photo reference data, no attempt was made to independently assess the accuracy of the 1972 or 1984 time period land cover maps.
Post-classification field checking for 2000 NY-NJ land cover change data
An additional field campaign was undertaken in October 2002 (subsequent to the classification/mapping process) to validate the accuracy of the 2000 land cover change data. A sample of 341 polygons (223 in New Jersey and 118 in New York) delineated as having changed land use between 1995 and 2000 (based on the visual interpretation of the SPOT panchromatic Yr 2000 and Landsat Enhanced Thematic Mapper image data) was selected and field checked. The field team was equipped with a portable color laptop computer linked to a real-time GPS receiver to locate the candidate change polygons in the field. Each site was visited, observations made, a ground photo acquired, and the sites classified into the appropriate land use category.
1995 Land Cover Accuracy Assessment
The pre- and post-classification field checked reference data were pooled to give 847 points for the southern New Jersey study area and 564 points for the northern New Jersey study area, for a total of 1411 points for the 1995 New Jersey land cover data. 500 photo checked points were available for the 1995 New York land cover data. Only the accuracy of the Level II and I land cover maps were assessed. No separate validation was conducted of the Level III land cover map due to the often small areas for some of the land cover classes which it made it difficult to adequately sample these classes. An error matrix was determined and the Producer's (a measure of omission error), User's Accuracy (a measure of commission error) and the Kappa coefficient of agreement were calculated for each class and for the overall map. The Kappa coefficient measures the agreement between the classified and reference data corrected for chance agreement (Congalton and Green, 1999). A value greater than 0.80 represents strong agreement and a value between 0.40 and 0.80 represents moderate agreement. A minimum sample size of thirty points per class is generally recommended for a valid accuracy assessment for that particular class. Some of the more rare classes (i.e., those classes of land cover that represent a relatively small proportion of the state=s area) fell below this recommended minimum sample size.
See the accuracy assessment tables below for results of the data verification exercise. The overall accuracies for the Level I and Level II maps were quite high with the Level II maps between 80 and 85% correct and the Level I map greater than 90% correct. The accuracy of the New Jersey Level I classification was 93.0% (Kappa coefficient = 0.9129) (Table 1a, b). The New Jersey Level II map had a classification accuracy of 85.2% (Kappa coefficient = 0.8348) (Table 2a, b). The accuracy of the New York Level I classification was 92.0% (Kappa coefficient = 0.9005) (Table 3a,b). The New York Level II map had a slightly lower classification accuracy of 80.0% (Kappa coefficient = 0.7776) (Table 4a,b).
The Level II land cover classification shows that some categories had lower classification than others. The classification of grasslands, such as pastures, showed accuracies in the 55% to 80% range with frequent mis-classification as cultivated land or lightly developed un-wooded land and vice versa. The upland scrub/shrub (e.g., abandoned agricultural fields in mid-to-late stages of vegetation succession, power lines, open ridges and pine barrens) and wetland scrub/shrub categories were poorly sampled making a proper accuracy assessment difficult. Anecdotal evidence suggests that this upland scrub/shrub category is problematic and regularly mis-classified with cultivated land, grassland and forest land. A similar situation occurs with the wetland scrub/shrub category being frequently mis-classified with emergent and forested wetland. The unconsolidated shore (i.e., sand beaches, lakeshores and tidal mudflats) was also under-sampled due to its relative infrequent occurrence (this category wasn't sampled for the New York data set but lumped with water). This category also had a lower classification accuracy due to confusion with associated categories of water and emergent wetland.
2000 Land Cover Change Accuracy Assessment
A total of 367 polygons were field checked. Out of 249 polygons field checked in New Jersey, 223 or 89.6% were correctly classified as being in an altered land use category, either developed (e.g., commercial or residential), recreational or transitional. Out of 118 field checked in New York, 103 or 87.3% were correctly classified as being in a human altered land use category. Thus the combined results suggests that land use alteration and change (i.e., from a non-developed to a developed or transitional land use category) between 1995 and 2000 was correctly identified 88.8 % of the time.
Original ancillary GIS vector components inputs (e.g. NWI data) were rasterized and incorporated into Geographic Information System (GIS) models processed by CRSSA at a ground cell spatial resolution of 30 x 30 meters. Spatial accuracies are dependent on source vector data from various organizations.
Path/Row Date 14/31 November 4, 1994 (scene downshifted 50%) 14/32 November 4, 1994 (scene downshifted 50%)
Leaf-On Imagery: Path/Row Date 14/31 September 4, 1995 (scene downshifted 50%) 14/32 September 4, 1995 (scene downshifted 50%)
The land cover mapping was undertaken at two levels of generalization: Level I, the most generalized with 8 classes; and Level 2, with 15 classes (Table 30, Highlands Study Technical Report). The Level I and Level II classification schemes were designed to follow the NOAA Coastal Land Cover Classification System (Dobson et al., 1995). The more generalized Level I classification scheme was used for comparison of the 1972, 1984, 1995 and 2000 classifications. Table 30 (from Highlands Study Technical Report). Classification Scheme used in Land Use/Land Cover Mapping.
Table 30a (from Technical Report). Land Cover Level I (8 classes)
1 1.10 Developed (Classes 1.11-1.12) 2 1.20 Cultivated/Grassland (Classes 1.20 & 1.30) 3 1.40 Woody Land (Classes 1.41-1.44) 4 1.60 Bare Land (Class 1.60) 5 2.00 Unconsolidated Shore (2.00) *not present in 2000/1995 lc product 6 2.10 Estuarine Emergent Wetland (Class 2.10) 7 2.40 Palustrine Wetland (Classes 2.30 & 2.40) 8 2.50 Water (Class 2.50)
Note: Developed land cover category 1.10 includes impervious, bare or partially vegetated land surfaces due to commercial, industrial, residential and transportation land uses. Forest/wetland land covers include upland and wetland forests, scrub/shrub and emergent vegetation communities. Cultivated/Grassland includes agricultural lands (including cultivated land, pastures and hay fields), managed grasslands (e.g., large areas of mowed and irrigated/fertilized lawn and golf courses) and unmanaged grassland. Bare Land includes lands made barren by quarrying/mining activities.
Table 30b (Technical Report). Land Cover Level II (15 classes)
1 1.11 Highly Developed (> 75% impervious surface) 2 1.12 Moderately (25-75% impervious surface) 5 1.20 Cultivated 6 1.30 Grassland 7 1.41 Deciduous Forest Woody Land 8 1.42 Conferous Forest Woody Land 9 1.43 Mixed Deciduous/Coniferous Forest Woody Land 10 1.45 Scrub/Shrub Woody Land 11 1.60 Bare Land 12 2.00 Unconsolidated Shore *not present in 2000/1995 lc product 13 2.10 Estuarine Emergent Wetland 14 2.30 Palustrine Emergent Wetland 15 2.40 Palustrine Forest Woody Land 16 2.45 Palustrine Shrub/scrub Woody Land 17 2.50 Water
A combination of digital image analysis techniques were used to classify the Landsat Thematic Mapper (TM) and Multispectral Scanner (MSS) images into land cover maps. Additional digital data sets were incorporated in the context of a geographic information system (GIS) to provide further classification improvement. Two major types of GIS data were used to guide the classification process: wetlands data (e.g., U.S. Fish and Wildlife Service National Wetland Inventory (NWI), New Jersey Department of Environmental Protection (NJDEP) freshwater wetlands, New York Department of Environmental Conservation (NYDEC) wetlands, Natural Resources Conservation Service (NRCS) soils), and land use data (NJDEP and New York county-level land use, U.S. Geological Survey land use/land cover). These data sets were incorporated into the classification process for either pre-classification stratification or post-classification modification. For example, the various wetland data sets were combined to create a new data theme that showed the likelihood that any particular grid cell should be classified as a wetland and this information was used in the classification process. The land use data were used to stratify the study area into primary land use types (e.g., developed, agriculture, vacant). These were then used to constrain the land cover classification (i.e., developed land cover types could only be found in grid cells that were mapped as developed land use).
The land use maps were created independently for New Jersey and New York. The 1995 digital orthophotographs of New York (1994/1995) and New Jersey (1995/1997) were simultaneously available, and so served as the baseline for land cover mapping. Approximately 50% of the New York imagery was from 1994 and 50% from 1995. A vast majority of the New Jersey imagery was from 1995 with only a small portion of western hunterdon and Warren counties covered by 1997 imagery. A mosaic of color infrared digital orthophotographic quarter-quadrangles (DOQQs) at 1-meter grid cell resolution was created for the bi-state Highlands region. The 1995 LU/LC data were available for the portion of the Highlands study area that falls in New Jersey (NJDEP, 2000). Land use maps were available at the county level in New York State and generally dated from the early to mid 1990's. These land use maps were overlain on the 1995 DOQQs. The land cover was updated as necessary using on-screen interpretation and digitization. The land use categories mapped are listed in Table 31 (Technical Report).
Table 31 (from Technical Report). Land Use categories (11 classes).
Class # Description 1 Residential 2 Industrial/Commercial 3 Institutional/Recreational 4 Transportation 5 Utility 6 Quarry/landfill 7 Agricultural 8 Forest 9 Undeveloped/Vacant 10 Transitional 11 Wetlands/Water A similar process was used to overlay the original county level land use maps on the 1984 Landsat TM imagery and update accordingly. In this case, areas of new development (i.e. development occurring from 1984 to the 1990's) were removed. Unfortunately, the coarser spatial resolution of the Landsat TM imagery (30 meters as compared to the 1 meter of the DOQQ's) limited our ability to interpret land use and create highly accurate land use maps for the New York portion of the study area for 1984. LU/LC data was available for the New Jersey portion of the study area for the year 1986 (NJDEP, 1996).
A combination of SPOT panchromatic and Landsat TM imagery were used to provide updated land use for the Year 2000. The SPOT panchromatic imagery (10-meter spatial resolution) was available as a region-wide mosaic of individual SPOT scenes that spanned a range of dates from 1998 to 2000 (only a small subset of the study area had 1998 data, the majority was from 2000). Thus, the same SPOT scene did not cover the entire NY-NJ Highlands region. This panchromatic data was merged with the September 23, 1999 Landsat TM imagery using a Principal Components resolution merge algorithm. The 1995 land use maps were overlain on the composite image, and areas of new development (development subsequent to 1995) were interpreted and digitized, on-screen, to produce a Year 2000 land use map.
Landsat imagery served as the remotely sensed source data used to consistently map land cover across the entire NY-NJ Highlands study region throughout the period of interest. Landsat TM images were acquired for relatively cloud-free dates in 1994 and 1995 (November 4, 1994 and September 4, 1995) for the 1995 baseline. Cloud covered areas were replaced with December 22, 1994 imagery. The November "leaf-off" imagery was taken after normal deciduous plant leaf fall, allowing the clearer differentiation of evergreen and deciduous forests as well as developed areas. The September "leaf-on" imagery permits the further discrimination of cultivated, wetland and developed areas. Landsat TM images from April 5, 2001 and September 23, 1999 were used to provide more recent land cover information for the Year 2000. Corresponding images from November 8, 1984 and September 21, 1984 were acquired in order to perform change detection. Earlier generation Landsat Multi-spectral Scanner (MSS) imagery from October 10, 1972 was also acquired to extend comparisons further back in time, albeit at a coarser spatial resolution and more generalized level of categorization.
PLEASE READ 'DATA QUALITY' FOR ACCURACY ASSESSMENT INFORMATION (IMPORTANT).
1. The Grant F. Walton Center for Remote Sensing and Spatial Analysis (CRSSA), Cook College, Rutgers University, disseminates this data layer as-is and makes no warranty, expressed or implied, nor does the fact of distribution consitute such a warranty. 2. The user should be aware that the data set is not intended for site specific analysis, however, its intention is to provide the end user with a regional-scale raster grid representation of a particular theme for which it was used in the Highlands Regional Study. 3. CRSSA, Rutgers University, will not be held responsible to further maintain the disseminated data, nor to provide the data in different formats other than its current availability. 4. Any maps, publications, reports or any other type of document produced as a result of an associated project utilizing this raster grid will credit the original author(s) as listed in the metadata as well as the Center for Remote Sensing and Spatial Analysis (CRSSA), Rutgers University.