Land Use Land Cover Change in the U.S. Great Lakes Basin 1992 to 2001

J. Great Lakes Res. 32:607–628 Internat. Assoc. Great Lakes Res., 2006

Land Use Land Cover Change in the U.S. Great Lakes Basin 1992 to 2001 Peter T. Wolter1,*, Carol A. Johnston2, and Gerald J. Niemi1 1Center

for Water and the Environment Natural Resources Research Institute University of Minnesota Duluth Duluth, Minnesota 55811 2Center

for Biocomplexity Studies P.O. Box 2202 South Dakota State University Brookings, South Dakota 57007

ABSTRACT. The pace of Land Use/Land Cover (LULC) change in the Great Lakes, particularly in urban and suburban areas, far exceeds that predicted by population growth alone. Thus, quantification of LULC and change through time may be a key factor in understanding the near-shore ecology of this system. The work described in this paper is part of a larger effort called the Great Lakes Environmental Indicators Project (GLEI), whose goal was to develop and refine environmental state indicators for the U.S. near-shore zone of the Great Lakes. Here we describe methodologies for using existing Landsatbased LULC maps to assemble consistent LULC data for the U.S. portion of the Great Lakes basin for 1992 and 2001, as well as summarizing salient LULC results. Between 1992 and 2001, 2.5% (798,755 ha) of the U.S. portion of the Great Lakes watershed experienced change. Transitions due to new construction included a 33.5% (158,858 ha) increase in low-intensity development and a 7.5% (140,240 ha) increase in road area. Agricultural and forest land each experienced ~2.3% (259,244 ha and 322,463 ha, respectively) decrease in area. Despite the large and enduring agricultural losses observed (2.23% of 1992 agricultural area), the rate of agricultural land decrease between 1992 and 2001 was less than that reported by the EPA (–9.8%) for the previous ~10-year period. Areas of new development were largely concentrated near coastal areas of the Great Lakes. Over 38% (6,014 ha) of wetland losses to development between 1992 and 2001 occurred within 10 km of a coastal area, and most of that area was within the nearest 1 kilometer. Clearly, these land use change data will be especially useful as quantifiable indicators of landscape change over time and aid in future land use planning decisions for protection of the integrity of the Great Lakes ecosystem. INDEX WORDS: Great Lakes, Landsat, land use, land cover, indicators.

land tenure systems. Different land uses impose different environmental stresses on natural plant and animal communities, with consequent implications to water quality, climate, ecosystem goods and services, economic welfare, and human health (Gutman et al. 2004). The 780,877 km 2 U.S. Great Lakes basin, like other regions of the U.S., is experiencing LULC changes that have potential ramifications to the condition of coastal and near-shore environments (Cummings 1978, Johnston 1992, Thorp et al. 1997). The pace of LULC change, particularly in urban and suburban areas (Fig. 1), far exceeds that

INTRODUCTION Land Use/Land Cover (LULC) change is one of the critical science issues of the 21st century, and quantifying LULC change is a major goal of government agencies in the U.S. (Climate Change Science Program 2003, McMahon et al. 2005) and the world (e.g., International Geosphere Biosphere Program 1999). LULC change is an indicator of changing human demographics, natural resource uses, agricultural technologies, economic priorities, and *Corresponding

author. E-mail: [email protected]

607

608

Wolter et al.

FIG. 1. False color change composite of Landsat sensor data showing LULC conversions between 1992 and 2001. Areas colored yellow in the change composite represent new developments that occurred during this time frame. predicted by population growth alone. Urbanization from 1970 to 1990 in the Chicago metropolitan area, for example, increased its developed area by 19%, with only a 2.2% increase in population (Auch et al. 2004). Land uses far removed from the coast can impact coastal environmental quality via pollutants released into tributary streams. For instance, stream quality often begins to degrade when impervious surface area within the contributing watershed exceeds 10% (Schueler 1994). Landscape and LULC changes such as increasing urbanization and impervious surface area (Kellog 1997); increased agricultural runoff of nutrients (Neilsen et al. 1982, Daniel et al. 1994); increased pesticide and/or herbicide applications (Frank et al. 1982, Gaynor et al. 1995, Jaynes et al. 1999); and increased erosion of sediment (Harris et al. 1993) have aggravated efforts to achieve water quality goals outlined in the Clean Water Act (U.S.

EPA 1988). For example, lands under development have been cited as contributing approximately five times the suspended solids per unit area in comparison with other LULC (Kellog 1997). In addition to water quality, the volume of storm water runoff has increased in concert with expanding development. Using aerial photographs from 1937 and 1995, Wegener (2001) studied changes in impervious surface area and found increases in urban area (formerly farmland) were responsible for a 69% increase in the volume of storm water runoff. He determined that a 7.62 cm (3 inch) rainfall event in 1995 resulted in a 65% greater rise in lake level than was the case in 1937 from the same volume of precipitation. These types of landscape change trends of declining storm water retention time and concomitant flashier stream flows, linked to increased urbanization, is a concern throughout the Great Lakes basin (Table 1).

Land Use Land Cover Change in the Great Lakes Basin TABLE 1. Hectares of land in farms within the U.S. Great Lakes watershed 1982–1992 (modified from U.S. EPA 1997). State Illinois Indiana Michigan Minnesota New York Ohio Pennsylvania Wisconsin Total

Farmland 1982 57,310 1,152,909 4,428,140 376,263 2,581,855 2,633,678 226,374 2,671,797 14,128,325

Farmland 1992 46,158 1,077,157 4,050,163 328,717 2,151,262 2,500,065 189,783 2,399,740 12,743,045

Change (%) –19.5 –6.6 –8.5 –12.6 –16.7 –5.1 –16.2 –10.2 –9.8

Increasingly, resource managers are using environmental indicators to evaluate the state of ecosystems (Jackson et al. 2000, Niemi et al. 2004). The State of the Lakes Ecosystem Conference (SOLEC) (Environment Canada and U.S. Environmental Protection Agency 2003) has already developed many environmental indicators for the Great Lakes basin including many LULC and LULC change categories. Ultimately, these indicators are designed to assess the status and trends of environmental conditions and serve as metrics of progress for current management policies. However, many indicators prescribed as part of the SOLEC process have not yet been fully developed. Regarding LULC data, sufficient classification standards for the entire basin do not yet exist. Work described in this paper is a component of a larger effort called the Great Lakes Environmental Indicators project (GLEI) (http://glei.nrri.umn.edu). GLEI is refining and developing indicators (e.g., Marks 2003, Johnston 2003, Niemi et al. 2004, Price et al. 2005) for the U.S. near-shore zone of the Great Lakes (~8,228 km of shoreline) based on extensive field sampling of near-shore areas (e.g., Sgro et al. 2006, Hanowski et al. 2006, Danz et al. 2005). In support of these activities, we produced a 30 m LULC dataset for the U.S. portion of the Great Lakes Watershed for 1992 and 2001. The objectives of this paper are to 1) describe methodologies used to produce the 1992 and 2001 LULC datasets most appropriate for change detection that cover the U.S. Great Lakes basin with existing thematic data sources, as well as raw Landsat sensor data, and 2) quantify LULC composition and changes in the watershed between 1992 and 2001.

609

BACKGROUND Multi-Resolution Land Characteristics Consortium The National Land Cover Data (NLCD) is a seamless LULC classification for the conterminous U.S. based on circa 1992 Landsat-5 Thematic Mapper (TM) sensor data, produced under the auspices of the Multi-Resolution Land Characteristics Consortium (MRLC), which initially represented the needs of the U.S. Environmental Protection Agency (EPA), U.S. Geological Survey (USGS), U.S. Forest Service (USFS), and the U.S. National Oceanic and Atmospheric Administration (NOAA) (Vogelmann et al. 2001). This initial dataset (NLCD 1992) consists of 21 hierarchical LULC classes at Anderson et al. (1976) level II detail (Table 2) and 30 m spatial resolution. A second stage MRLC effort is underway with additional cooperation from the Bureau of Land Management (BLM), National Aeronautics and Space Administration (NASA), Natural Resources Conservation Service (NRCS), and the National Park Service (NPS) to produce a 2001 update to the NLCD 1992 LULC classification. These NLCD 2001 data are being produced independently without direct reference to the NLCD 1992 layer or the raw circa 1992 TM data used to produce them (pers. comm., Loveland and Homer, USGS EROS Data Center, 2005). Rather, the NLCD 2001 effort is using an automated approach to select suitable circa 2001 Landsat imagery (primarily ETM+) for classification (see: Yang et al. 2001). Because of the differences in processing strategy between 1992 and 2001 (see: Vogelmann et al. 2001, Homer et al. 2004), the two datasets are expected to be reasonably compatible (Homer et al. 2004), but not particularly suitable for pixel-level change detection (pers. comm., Vogelmann and Homer, USGS EROS Data Center, 2005). This shift in image processing protocol was intentional and designed to 1) take advantage of new methodologies and technologies that were either not available or not practical at the time the 1992 data were being processed, 2) to meet increasing MRLC member needs, and 3) pave the way for automated updates (Homer et al. 2004). Therefore, it was understood, and accepted, that the NLCD 2001 dataset would not be entirely backward compatible with the NLCD 1992 dataset (Homer et al. 2004).

Wolter et al.

610

TABLE 2. National Land Cover Data (NLCD) classification schemes for 1992 and 2001. Some classes are mapped exclusively for coastal areas (*) or for Alaska (**). 1992 11 - Open water 12 - Perennial ice/snow

2001 11 – Open water 12 – Perennial ice/snow

21 - Low intensity residential 22 - High intensity residential 23 - Commercial/industrial/transportation

21 – Developed, open space 22 – Developed, low intensity 23 – Developed, medium intensity 24 – Developed, high intensity

31 - Bare rock/sand/clay 32 - Quarries/strip mines/gravel pits 33 – Transitional

31 – Barren land 32 – Unconsolidated shore

41 - Deciduous forest 42 - Evergreen forest 43 - Mixed forest

41 – Deciduous forest 42 – Evergreen forest 43 – Mixed forest

51 – Shrubland

51 – Dwarf shrub ** 52 – Scrub/shrub

61 – Orchards/vineyards/other 71 - Grasslands/herbaceous

72 – Grassland/herbaceous 73 – Lichens ** 74 – Moss **

81 - Pasture/hay 82 - Row crops 83 - Small grains 84 - Fallow 85 - Urban/recreational grasses

81 – Pasture/hay 82 – Cultivated crops

91 - Woody wetlands 92 - Emergent herbaceous wetlands

90 – Woody wetlands 91 – Palustrine forested wetlands * 92 – Palustrine scrub/shrub * 93 – Estuarine forested wetlands * 94 – Estuarine scrub/shrub * 95 – Emergent herbaceous wetland* 96 – Persistent palustrine emergent wetland * 97 – Palustrine emergent wetland* 98 – Palustrine aquatic bed * 99 – Estuarine aquatic bed *

Coastal Change and Analysis Program In addition to MRLC efforts, NOAA’s Coastal Services Center (CSC) contracted with Earth Satellite Corporation and Space Imaging to produce independent sets of classifications for most of the Great Lakes basin (Fig. 2) using Landsat-5 sensor data from circa 1996 and Landsat-7 sensor data from circa 2001 (U.S. NOAA 2003a, 2003b). The derived LULC data would then support the CSC’s Coastal Change and Analysis Program (C-CAP), whose goal was to produce land cover and change

information for the coastal areas of the U.S., including the Great Lakes (U.S. NOAA 2003a). The original C-CAP classification scheme consisted of 22 LULC categories (Table 3) and placed greater emphasis on wetland categories and less on class distinctions among developed and agricultural LULC than did the NLCD 1992 classification scheme (Table 2). Recently, NOAA and the U.S. Geological Survey formed a partnership to produce a more detailed classification scheme to better serve their respective

Land Use Land Cover Change in the Great Lakes Basin

FIG. 2.

611

Extent of C–CAP classified data within the Great Lakes basin.

needs. As a result, both the NLCD 2001 and new CCAP data layers have expanded LULC classification schemes in coastal areas (Tables 2 and 3). To date, the NLCD 2001 LULC classifications are only 50–60% complete for most areas within the Great Lakes basin (pers. comm., Homer, USGS EROS Data Center, 2005) and completion is not expected until late 2006 (Homer et al. 2004).

the watershed, and to apply the results toward development of environmental state indicators for the U.S. near-shore zone of the Great Lakes. In this paper, we present summaries of LULC within the U.S. portion of the Great Lakes watershed (hereafter simply referred to as the watershed) and LULC change between circa 1992 and 2001. METHODS

OBJECTIVES Our first objective was to determine a way to combine as much information as possible from extant data sources, including NLCD 1992, original C-CAP series (1996 and 2001), USGS Gap Analysis Program (GAP) data (www.gap.uidaho.edu), WISCLAND (Reese et al. 2002), and our northeast Minnesota (NEMN) classification (Wolter et al. 1995, Wolter and White 2002) (Table 4), to be able to compile a spatially and thematically consistent dataset for the U.S. portion of the Great Lakes basin from 1992 and 2001 without having to reprocess large volumes of raw Landsat sensor data. Additional objectives were to use these derived spatial data to summarize LULC and LULC change within

Datasets Used Generating spatially and thematically compatible LULC data for 1992 and 2001 required the use of various sources of raster and vector data (Table 4). National Land Cover Data from 1992 (Table 2), covering the whole watershed, was the base data layer used to produce an augmented NLCD 1992 data layer for the basin, hereafter referred to as Great Lakes 1992 (GL1992). Information from the original C-CAP classification from 2001 (Table 3), C-CAP’s 1996–2001 change layer (http://www.csc .noaa.gov/crs/lca/greatlakes.html), and NLCD 1992 data were the primary data used to create an NLCD 1992-compatible layer for 2001, hereafter referred to as Great Lakes 2001 (GL2001). Several data

Wolter et al.

612

TABLE 3. National Oceanic and Atmospheric Administration Coastal Change and Analysis Program (C-CAP) classification schemes. The Original C-CAP scheme (left column) was standard classification protocol prior to 2005. The new scheme (right column) was developed as a result of collaboration between NOAA and the U.S. Geological Survey (USGS) and has been in use since 2005. Original 1996 & 2001 C-CAP 2 - High intensity developed 3 - Low intensity developed

New 2001 C-CAP 2 - Developed, high intensity 3 - Developed, medium intensity 4 - Developed, low intensity 5 - Developed, open space

4 - Cultivated land

6 - Cultivated crops 7 - Pasture/hay

5 - Grassland

8 - Grassland/herbaceous 9 - Sedge/herbaceous

6 - Deciduous forest 7 - Evergreen forest 8 - Mixed forest

10 - Deciduous forest 11 - Evergreen forest 12 - Mixed forest

9 - Scrub/shrub

13 - Scrub/shrub 14 - Dwarf scrub 15 - Barren land

10 - Palustrine forested wetland 11 - Palustrine scrub/shrub wetland 12 - Palustrine emergent wetland (persistent) 13 - Estuarine forested wetland 14 - Estuarine scrub/shrub wetland 15 - Estuarine emergent wetland 16 - Unconsolidated shore

16 - Palustrine forested wetland 17 - Palustrine scrub/shrub wetland 18 - Palustrine emergent wetland (persistent) 19 - Estuarine forested wetland 20 - Estuarine scrub/shrub wetland 21 - Estuarine emergent wetland 22 - Unconsolidated shore

17 - Bare land 18 - Water 19 - Palustrine aquatic bed 20 - Estuarine aquatic bed

23 - Open water 24 - Palustrine aquatic bed 25 - Estuarine aquatic bed

21 - Tundra 22 - Snow/ice

26 - Tundra 27 - Perennial ice/snow 28 - Moss 29 - Lichens

sources were used to 1) augment the NLCD 1992 woody wetlands class, 2) reconcile errors in C-CAP wetlands data, 3) classify LULC change between 1992 and 2001 in areas devoid of thematic data, 4) fix 2001 wetland classes outside the map-extent of the C-CAP data (Fig. 2), and 5) resolve disparities between NLCD and C-CAP developed classes (Table 4). TIGER data from 1992 and 2001 were used because both original C-CAP datasets (1996 and 2001) had TIGER roads data incorporated into them (pers. comm., Raber and Herold, NOAA

Coastal Services Center, 2004), while NLCD 1992 data did not. NLCD1992 Woody Wetlands Wetland classes in the NLCD 1992 database (Table 2) are water, emergent herbaceous, and woody wetlands (WW). The WW class is a major component of the Great Lakes watershed making up 11.5% of the land area and 15.4% of Minnesota, Wisconsin, and Michigan’s combined watershed area. Unfortunately, the WW class definition was

Land Use Land Cover Change in the Great Lakes Basin

613

TABLE 4. Datasets used to produce spatially and thematically comparable, NLCD- like data for the U.S. Great Lakes watershed representing LULC as of 1992 and 2001. All data sources except TIGER were derived from 30-meter Landsat sensor data. Dataset NLCD

Source USGS

Date 1992

Primary Use 1992 base data

Reference Vogelmann et al. 2001

C-CAP

NOAA

1996, 2001

2001 base data

crs/lca/greatlakes.html

WISCLAND

Wisconsin DNR

1992

fix NLCD92 woody wetlands

Reese et al. 2002

NEMN

NRRI U of MN Duluth

1988, 1995, 2001

fix NLCD92 woody wetlands

Wolter et al. 1995, Wolter and White 2002

GAP

USGS

1992

fix NLCD92 woody wetlands

http://www.gap.uidaho.edu/

U.S. Census Bureau

1992, 2001

fix NLCD92 developed

http://www.census.gov/geo/

TIGER TM, ETM+

NASA, USGS

1992, 2001

fill voids in GAP, C-CAP

http://landsat.gsfc.nasa.gov/ http://edc.usgs.gov/

http://www.csc.noaa.gov/crs/

too ambiguous to be of use and thus was divided into lowland brush types, lowland conifer (LC), lowland hardwood (LH), and lowland mixed forest (LM) types by substituting ancillary classified data (Table 4) that contained these more detailed classes (Appendixes 1 and 2) for the WW class in NLCD1992 to produce GL1992. Upland forest, brush, and grass categories from the ancillary classifications that fell within the WW class were assumed wet and were overlaid as such. This procedure worked well because the WW class agreed visually with combined lowland woody types of the GAP, NEMN, and WISCLAND classifications. GAP data from 1992 were used to fix the NLCD WW class in Michigan. However, the lower 1/3 of the Lower Peninsula was not classified at the time this work was being conducted. For these areas, original C-CAP data from 1996 were used to extract more detailed woody wetlands class information. Because forested wetlands are a single class in C-CAP (Table 3), spring 1992 TM and summer 2001 ETM+ data were classified to extract LH, LC, and LM forest information that were substituted for WW in GL1992. The water class of the Michigan GAP classification available at the time of this project was found to include erroneous lowland conifer classes—

www/tiger

specifically black spruce in Landsat path 22/row 28. Similarly, the palustrine scrub/shrub wetland class of the C-CAP classification was ubiquitously confused with lowland conifer in Landsat paths 22–23/row 28. Black spruce has an unusually low reflectance in the visible and infrared wavelengths of the electromagnetic spectrum compared to other forest types, so it is understandable that these confusions occurred. Classification errors were fixed using raw Landsat data (TM and ETM+) and Michigan Department of Natural Resources aerial photographs (1992 and 1998) as ground truth (available: www.michigan.gov/dnr/). It should also be noted that 1992 and 2001 GAP classifications have since been completed for Michigan (http://www.msu.edu/user/skillen1/MIGAP.htm). NLCD1992 Developed Land and TIGER Road information was added to the original NLCD 1992 data to make it compatible with original C-CAP classifications, which included this information. To accomplish this, all major paved road vectors in the 1992 TIGER database were selected and converted to 30 m raster data, then overlaid on GL1992. This split the “Commercial/Industrial/ Transportation” class of the original NLCD database (Table 3) into two classes for GL1992: “Commercial/Industrial” and “TIGER92 Roads” (Table

Wolter et al.

614

TABLE 5. Classification scheme of the GL1992/GL2001 LULC data is in the left column. Numbers in parentheses indicate aggregated class membership (right column). The miscellaneous vegetation class (code 6) was generated to represent land that was vegetated, but not mature forest or annual row crop. (1) (1) (1) (1) (3) (1) (6) (2) (2) (2) (3, 6) (2, 6) (4) (4) (4) (3, 6) (3, 6) (5) (5) (5) (5) (5) (5) (5) (5)

Low intensity residential High intensity residential Commercial/industrial Roads (Tiger 1992) Bare rock/sand/clay Quarries/strip mines/gravel pits Urban/recreational grasses Pasture/hay Row crops Small grains Grasslands/herbaceous Orchards/vineyards/other Deciduous forest Evergreen forest Mixed forest Transitional Shrubland Open water Unconsolidated shore Emergent herbaceous wetlands Lowland grasses Lowland scrub/shrub Lowland conifers Lowland mixed forest Lowland hardwoods

5). Since it was not known how TIGER data were incorporated into the original C-CAP classifications, we repeated this procedure using 2001 TIGER road vectors and original 2001 C-CAP data then overlaid the results back on top of the 2001 CCAP data, producing a 2001 TIGER roads class. Producing Great Lakes 2001 Dataset Original NLCD and C-CAP 2001 classification schemes were not directly comparable (Tables 2, 3), so a series of masking, overlaying and recoding procedures were employed to generate a 2001 LULC map called GL2001 that could be compared with NLCD 1992. The first step was simply to copy GL1992 to GL2001. Areas mapped as developed on C-CAP2001 were compared with their corresponding areas on GL1992 and substituted into GL2001 according to several rules. If C-CAP 2001 was “High-Intensity Developed” (which included both commercial-industrial and high-intensity residential) and NLCD was “High-Intensity Residential” or “Commercial-Industrial,” then the original NLCD

1 Developed 2 Agriculture 3 Early successional vegetation 4 Forest 5 Wetland 6 Miscellaneous vegetation

classes were retained in GL2001 to preserve classification detail. If C-CAP 2001 was “High-Intensity Developed and NLCD was “Low-Intensity Residential,” then the C-CAP 2001 class was inserted into GL2001. If C-CAP 2001 was “High-Intensity Developed” or “Low-Intensity Developed” and NLCD was any undeveloped class, then the C-CAP 2001 class was inserted into GL2001. To generate GL2001 LULC data in areas mapped as non-developed by C-CAP 2001, we used C-CAP’s 1996–2001 change layer (http://www.csc .noaa.gov/crs/lca/change_analysis.html) to identify areas where a LULC change had occurred by 2001. If a change had been mapped by C-CAP in a nondeveloped area, then the land use from C-CAP 2001 was inserted into GL2001; if no change was indicated, then NLCD 1992 classes were retained in GL2001. The matrix operation (ERDAS 1999) was then used to combine 1992 (NLCD, time-1) and 2001 (C-CAP, time-2) LULC into one file containing 333 unique LULC transition cases between the two classifications and time periods.

Land Use Land Cover Change in the Great Lakes Basin

615

TABLE 6. Overall change and transitions of non-developed land to developed land between 1992 and 2001 within near-shore zones of the Great Lakes and for the watershed. Attribute Measured Total area (ha) Area unchanged (ha) Area changed (ha) Percent of area changed Percent of area unchanged Non-developed to developed (ha) % of buffer area % of basin area % of all basin transitions % of basin non-dev. to dev.

0–1 km 647,440 616,447 30,994 4.8% 95.2% 151,889 2.3% 0.1% 1.9% 3.9%

Because original NLCD and C-CAP classification schemes were not directly comparable, transition cases were examined for logical consistency and edited where needed. Conversions to forest were checked to ensure that inferred succession rates were reasonable. For example, a time-1 to time-2 change from “transitional” to “hardwood forest” was deemed feasible, but a change from “grassland” to “hardwood forest” was not, because establishment of a mature forest is impossible over such a short time span within the climatic region of the Great Lakes. In the latter case, the time-2 class was edited from “hardwood forest” to “transitional,” a category employed by NLCD 1992, but not by C-CAP 2001. Transitions among classes were also checked to eliminate those due merely to differences in classification conventions between the two inventories. For example, if the time-1 class was mixed forest and the time-2 class was coniferous forest, the time-1 type was left unchanged. If the time-1 type was “row crop” and the time-2 type was “cultivated,” the time-2 type was changed to the more specific “row crop.” This strategy was adopted because the original C-CAP classification scheme did not distinguish agricultural types beyond “cultivated.” Analyzing Resulting LULC Change (GL1992–GL2001) We examined LULC and change basin-wide and within 1, 5 and 10 km buffer zones from Great Lakes shorelines using Imagine’s search, mask, and

Shoreline Buffer Zones 0–5 km 0–10 km 2,686,163 4,936,957 2,592,019 4,777,057 94,144 160,120 3.5% 3.2% 96.5% 96.8% 50,145 83,592 1.9% 1.7% 0.2% 0.3% 6.3% 10.5% 12.7% 21.2%

Whole Basin 31,525,961 30,727,206 798,755 2.5% 97.5% 393,719 — 1.2% 49.3% 100.0%

matrix routines (ERDAS 1999). First, the matrix routine was used to generate a GL1992-GL2001 change layer that resulted in 576 potential LULC change categories. Then, shoreline buffers were used to clip the change layer producing 0–1, 1–5, and 5–10 km concentric change layers. Finally, to condense the presentation of LULC transition results into salient change trends, the original 24 types were aggregated into six general LULC categories (Table 5) within change matrices. RESULTS Basin-wide LULC Change Overall, 798,755 hectares (2.5%) of the U.S. portion of the Great Lakes watershed experienced some sort of LULC change from 1992 to 2001 (Table 6, Fig. 3). The two dominant LULC types in 1992 were forest and agriculture, covering 44.8% and 36.8% of the watershed, respectively. By 2001, each had decreased in area by ~2.3% (Fig. 4). Of the changes that occurred in the basin, 49.3 % were transitions from undeveloped to developed land (Table 6, Fig. 5). Development (high-intensity, lowintensity, and roads) and most early successional vegetation (ESV) classes (e.g., upland grasses and brush) increased with concomitant decreases in forest and agricultural classes (Fig. 3). For instance, low-intensity development increased by 33.5%, high-intensity development by 19.6%, roads by 7.5%, upland brush by 137.4%, upland grasses by 14.7%, and lowland brush by 3.8%, while upland

616

Wolter et al.

FIG. 3. LULC type changes for the U.S. Great Lake basin by area and percent change since 1992 [numbers above and below bars].

and lowland forest classes all decreased between 1.1% and 2.6% (Fig. 3). Although forest loss between 1992 and 2001 was small on a percentage basis, the area changed was large—specifically the decrease in upland hardwoods by ~215,000 ha (–2.6%) (Fig. 3). Forest types combined occupy the largest LULC type in the watershed for both 1992 and 2001 (Fig. 4) with an area of 14,108,375 ha or ~44.8% in 1992 that decreased by 322, 463 ha or ~2.3% by 2001. Increases in low-intensity development and road area were similar in magnitude to the loss of forest (Fig. 3), but the percentage increase was much greater due to the smaller proportion of developed land in the Great Lakes basin. For example, road area increased 7.5% between 1992 and 2001 (Fig. 3), and was the fourth most dominant LULC type in the watershed, covering ~2 million hectares or ~5.9% in 1992 (Fig. 4). Watershed and Near-shore LULC Transitions The original 24 LULC classes were aggregated into six classes for transition analyses (Table 5).

The ten most common transition categories can be summarized into three general groups: agriculture to developed, forest to early successional vegetation (ESV), and forest to developed (Fig. 5). Agricultural to developed conversions represented the category of greatest change (210,068 ha), forestland to ESV was the second largest transition (180,690 ha), and forest to developed land was the third (154,681 ha). Of the 2.5% (798,755 ha) of watershed area that experienced change between 1992 and 2001, 4.8% (30,994 ha) of this total occurred within one kilometer of the Great Lakes shoreline (Table 6). The 1–5 and 5–10 km buffer zones each contained ~3% of the total watershed change (Table 6). Thus, the greatest percentage of the overall, watershed change observed within the three near-shore buffer zones studied occurred within the narrowest and closest zone to the shore (Table 6). Land use/land cover transitions between 1992 and 2001 within these near-shore zones of the Great Lakes largely parallel those of the overall watershed (Fig. 5). While the same three transition categories dominated, their proportions varied by

Land Use Land Cover Change in the Great Lakes Basin

617

FIG. 4. LULC type area and change within U.S. Great Lakes basin 1992–2001. Numbers above the bar pairs indicate proportion of the watershed that each type represents (based on GL1992 dataset). See Figure 3 for legend of LULC types. buffered distance from the lakes. Within the 0–1 km zone from the Great Lakes shoreline, conversions of forest to both ESV (9,087 ha, 5.0% of total category change (TCC)) and developed land (8,657 ha, 5.6% of TCC) were the largest transitions, followed by conversion of 3,935 ha (1.9% of TCC) of agricultural land to developed. For the 1–5 km zone inland from the shore, forest to developed conversion was the largest of the three transitions (17,049 ha, 11.0% of TCC), followed by agricultural to developed (14,279 ha, 6.8% of TCC) and forest to ESV (13,116 ha, 7.3% of TCC). Within the 5–10 km zone from shoreline, transition category dominance was most similar to the trend for the whole watershed (Fig. 5), with 16,113 ha (7.7% of TCC) of agriculture converted to developed, 14,516 ha (8.0% of TCC) of forest converted to ESV, and 14,390 ha (9.3% of TCC) of forestland being devel-

oped by 2001 (Fig. 6). When all buffers form shoreline out to 10 km are combined, the forest to developed transition category was the largest (40,099 ha, 25.9% of TCC), followed by forest to ESV (36,726 ha, 20.3% of TCC), and agricultural to developed (34,328 ha, 16.3% of TCC). The remaining seven transitions categories were potentially minor in importance with regard to the three change categories discussed above; however, a few are noteworthy. For instance, of the 15,685 ha of wetland converted to developed land between 1992 and 2001 within the watershed (Fig. 5), 12.8% occurred within 1 km of the Great Lakes shoreline, 14.9% within the 1–5 km range, and 10.7% within the 5–10 km zone (Fig. 6). When combined, 38.3% of wetland conversion to developed land between 1992 and 2001 occurred within 10 km of the Great Lakes shoreline (Fig. 6). Also within this 10 km,

Wolter et al.

618

FIG. 5.

LULC transitions for the whole U.S. portion of the Great Lakes basin.

near-shore, zone, 25.9% of the conversion of upland forest to developed land occurred, as well as 33.7% of all conversion of ESV to developed land (Fig. 6). Lastly, although it appears that conversion of developed land to miscellaneous vegetation was dramatic within the 10 km near-shore zone over the time period studied (Fig. 6), these conversions only involved 2,205 ha of land for the whole watershed (Fig. 5), 54.6% of which occurred within 0–5 km from shore (Fig. 6). DISCUSSION Data Processing Comparing LULC data for the Great Lakes watershed for 1992 and 2001 using the methods described above was efficient, repeatable, and necessary to be able to analyze LULC change trends within the Great Lakes basin without having to reprocess great quantities of raw Landsat sensor data. In the future, NLCD products dating from 2001 and beyond will be spatially and thematically

compatible allowing pixel-wise change analysis. Thus, efforts such as these should not be necessary. Because the nominal time period of these change analyses was 1992 to 2001, one may question the use of the C-CAP 1996–2001 change layer to determine GL2001 LULC type because C-CAP 1996 data represents an approximate four-year difference from NLCD 1992. It is worth noting that within the Great Lakes basin the NLCD 1992 dataset was generated using Landsat TM data from as early as 1988 and late as 1994, and C-CAP 1996 with TM data from 1993 through 1997. Upon closer review of the years of satellite imagery used to derive these LULC datasets, we found a ~26% overlap in time between NLCD 1992 and C-CAP 1996. For instance, of the 47 Landsat footprints covering our study area, 56% of the TM data used to produce NLCD 1992 were from 1988 to 1991, 25% from 1992, and 19% from 1993 to 1994. As for C-CAP 1996, 54% of the TM data used were from 1993–1995, 37% from 1996, and 9% from 1997 (see: http://landcover.usgs.gov/ natllandcover.asp and http://www.csc.noaa.gov/crs/

Land Use Land Cover Change in the Great Lakes Basin

619

FIG. 6. LULC change by buffer class as a percentage of watershed-level changes quantified (area figures above each category) for each LULC change category. lca/greatlakes.html). Therefore, NLCD1992 and CCAP 1996 are more comparable in time than their labels would imply. Lastly, it should also be noted that of the 798,755 ha of change observed within the watershed between 1992 and 2001, 49.4% represented transitions to developed land (Fig. 5), which was quantified without using the C-CAP 1996–2001 change dataset. Methodologies certainly exist for generating NLCD 1992-compatible LULC for 2001 without having to use ancillary thematic data layers such as C-CAP, GAP, NLCD, NEMN, or WISCLAND (Appendixes 1, 2). However, because considerable time was spent assessing the accuracy of these ancillary LULC data layers from across the watershed (Wickham et al. 2004, Reese et al. 2002, U.S. NOAA 2003a, U.S. NOAA 2003b, Wolter and White 2002), we believe that careful use of these data layers provided an efficient and defensible methodology for producing quasi-NLCD 2001 LULC data for the Great Lakes watershed. Both the

original NLCD 1992 and C-CAP efforts strove for individual class accuracies of at least 85%. However, the NLCD 1992 dataset, using Anderson et al. (1976) level II, fell short due largely to confusion of agricultural types with each other. Accuracy of NLCD 1992 LULC improved considerably when agricultural classes were merged into one class (Wickham et al. 2004). C-CAP’s overall class accuracies for the Great Lakes (excluding Michigan) were all above the acceptable minimums with the exception of the scrub/shrub and mixed forest classes, which were largely being confused with each other (U.S. NOAA 2003a). In Michigan, all accuracy standards were met except for high intensity developed, grassland and palustrine emergent, which were being confused with low intensity developed, cultivated land and palustrine scrub/shrub, respectively (U.S. NOAA 2003b). Finally, the WISCLAND classification had an accuracy of ~94% at Anderson Level I and 77% at level II (Reese et al. 2002), while the NEMN classification, within the

Wolter et al.

620

FIG. 7.

Percent of LULC category change with respect to area analyzed.

watershed, was 75% accurate at Anderson level III (Wolter and White 2002). Confusions that were documented within the NLCD 1992 and C-CAP classifications were deemed not serious enough to preclude our efforts. LULC and Change Forest and agricultural classes make up the majority of LULC in the watershed for each time period. Transitions of these classes to agriculture, ESV, forest, or other non-developed land represent changes with relatively low landscape retention times (Pastor and Wolter 2002). For example, forest clear-cuts within managed forestlands will stay in the new ESV state for a relatively short time before the patch succeeds back into forest cover. Conversely, 49.3% of the change that occurred within the watershed between 1992 and 2001 involved transitions of non-developed land to developed land (Table 6, Fig. 5), which has an extremely high landscape retention time (Pastor and Wolter 2002)—

meaning that once land is developed (e.g., into residential or commercial land) there is an extremely low probability that these newly developed lands will revert back into the initial, non-developed state, such as agriculture or forest. This steady increase in hardened surface area is of particular concern for near-shore areas of the Great Lakes and the watershed as a whole. Over 21% (~84,000 ha) of all newly developed land within the basin between 1992 and 2001 occurred within 10 km of a shoreline, even though this only represents 0.27% of the whole watershed. Moreover, the highest concentrated change occurred within the nearest one kilometer (Table 6, Fig. 7). The increased hardening of the Great Lakes coastal areas is a major concern, as increased storm-water runoff often carries with it excessive sediment and chemical loads (Neilsen et al. 1982, Frank et al. 1982, Daniel et al. 1994, Gaynor et al. 1995, Jaynes et al. 1999). This ultimately becomes an environmental and economic burden to local communities and society as a whole

Land Use Land Cover Change in the Great Lakes Basin (Harris et al. 1993). Of equal concern, especially given stringent wetland protection measures during the 1990s (NRC 2001), is the large wetland area converted to development between 1992 and 2001 within 10 km of the coast (6,014 ha), a change that constituted 0.3% of land area in the nearest 1 kilometer from the shoreline. Much of the new development observed within the watershed for this time period occurred in the form of urban or suburban sprawl. Radial growth of new housing developments and associated infrastructure outward from urban centers was an ubiquitous phenomenon observed throughout the watershed between 1992 and 2001 (Fig. 8). The 33% increase in low-intensity developments since 1992 represents the greatest percentage and area increase of the three developed classes, most of which originated as agricultural land. Overall, agricultural land decreased by 2.24% between 1992 and 2001, which represents the largest transition area of all categories analyzed (Fig. 5). Albeit large in area, this represents a considerably slower rate of change than the 9.8% that was reported for the preceding 10-year period within the U.S. Great Lakes watershed (Table 1). Of the transitions away from agricultural land by 2001, approximately 210,068 ha (81%) converted to development (Fig. 5), and 16.3% of that occurred within 10 km of the Great Lakes shoreline (Fig. 6). Interestingly, some researchers have suggested that transitions from agriculture into development (impervious surfaces) result in reductions of sediment delivery to streams, although storm water runoff volume and rate increase along with other negative ecosystem effects (Brim Box and Mossa 1999). Finally, transition of forested land to ESV, developed, and agricultural land together amount to 1.2% of the watershed area and 46.1% of all land area that experienced change (Fig. 5). Near-shore conversions of forest to development and ESV comprise a greater percentage of land area (0–1 km) when compared to the watershed figures (Fig. 7). Conversion of near-shore wetlands and ESV to developed land follows this trend as well. Because the effects of forest cutting on stream water quality have been well documented (Coats and Miller 1981, Martin et al. 2000, Ensign and Mallin 2001), summaries of these transitions by sub-watershed may serve as a valuable indicator of ecosystem stress.

621

Forest Change The forest change results reported here appear to contradict the work of others that found general increases in forest cover throughout the upper Great Lakes in the decade preceding this study (Miles and Chen 1992, Leatherberry and Spencer 1996, Schmidt 1997), and that increases were linked to agricultural abandonment and increased housing in forested areas (Brown et al. 2000, Michigan Economic and Environmental Roundtable 2001). For example, Brown et al. (2000) used a Markov transition probability-based modeling approach on LULC data they derived from 60 m Landsat Multispectral Scanner (MSS) data from three time periods (1972–1975, 1985–1987, and 1990–1992) to determine drivers of LULC change in the Minnesota, Wisconsin, and Michigan. They found average forest cover on 136 sample sites taken from 17 counties increased ~4.5% from the early 1970s to the early 1990s. Similar findings have been reported by the United States Department of Agriculture (USDA) Forest Service for these states (Miles and Chen 1992, Leatherberry and Spencer 1996, Schmidt 1997). Using a subsample of the 136 sites, Brown et al. (2000) also found a positive relationship between forest regeneration and agricultural abandonment and a negative relationship between forest regeneration and decadal rate of increasing development. Brown et al. (2000) note that their measure of agricultural abandonment may have been exaggerated because 1) it did not account for new land brought into production, and 2) fallow agricultural land was mislabeled as abandoned. In any event, they suggest that there is a considerable lag-time between agricultural abandonment and afforestation of these sites. Whether or not overall forest area in the Great Lakes basin has changed relative to previous decadal point-estimates using Landsat MSS data (Brown et al. 2000) or other point-based estimates (Miles and Chen 1992, Leatherberry and Spencer 1996, Schmidt 1997) is unclear. It is noteworthy, however, that a recent change detection analysis using Landsat TM data in northeast Minnesota between 1990 and 1995 revealed greater forest harvesting than afforestation due to an apparent pulse in demand for fiber over this period (Wolter and White 2002). In this case harvesting occurred primarily on private lands (Wolter and White 2002), where ~0.7 million cords of wood sold in 1990 increased to 2.0 million cords sold in 1994—a harvest level maintained until 1999, thereafter steadily de-

622

Wolter et al.

FIG. 8. LULC change in the lower Green Bay basin of Lake Michigan (A) and the area surrounding Detroit, MI (B).

Land Use Land Cover Change in the Great Lakes Basin creasing to 1.5 million cords sold in 2001 (Minnesota Department of Natural Resources 2003). Currently, the authors are unaware of similar increases in fiber demand occurring elsewhere across the Great Lakes basin that may have contributed to the observed reductions in forest cover between 1992 and 2001. Although determination of specific, economic, causality of LULC changes observed in the Great Lakes basin was beyond the scope of this research, the rate of forest clearing for new developments (154,681 ha) between 1992 and 2001 far exceeded forest gains from agricultural sources (18,331 ha). Thus, we suspect that when coupled with temporary forest losses due to changing forest management practices, previous afforestation figures based on probabilistic estimates are likely being overshadowed by larger trends during this period. CONCLUSIONS We quantified LULC and change for the U.S. portion of the Great Lakes watershed using 1) an array of existing Landsat-derived LULC maps from various state and federal sources throughout the region, and 2) raw Landsat sensor data where existing LULC maps were unavailable. Careful use of the former facilitated assemblage of spatially and thematically compatible LULC data for 1992 and 2001 using a repeatable, efficient approach with minimal reliance on raw satellite sensor data. Between these two nominal time periods (1992 and 2001), the U.S. portion of the Great Lakes watershed has undergone substantial change in many key LULC categories. Of the total change that occurred (2.5% of watershed area), salient transition categories included a 33.5% increase in area of low-intensity development, a 7.5% increase in road area, and a decrease of forest area by over 2.3%— the largest LULC category and area of change within the watershed. More than half of the forest losses involved transitions into ESV, and hence will likely remain in forest production of some sort. However, nearly as much forest area was, for all practical purposes, permanently converted to developed land. Likewise, agriculture lost over 50,000 more hectares of land to development than forestland, much of which involved transitions into urban/suburban sprawl. Contrary to previous decadal estimates showing an increasing forest area trend from the early 1980s to the early 1990s, due to agricultural abandonment and transitions of forest land away from active management, we ob-

623

served an overall decrease (~2.3%) in forest area between 1992 and 2001. Explanation of this trend is largely unclear; however, we suspect both increased forest harvesting practices in parts of the region coupled with forest clearing for new developments are overshadowing gains from the agricultural sources observed in previous decades. Interesting, but not surprising, was the concentration of new developments near coastal areas of the Great Lakes. For instance, over one third of wetland losses to development between 1992 and 2001 occurred within 10 km of a coastal area, and most of that area was within the nearest 1 kilometer. This is a concern because Great Lakes coastal wetlands provide habitat for a wide variety of fauna, support plant communities adapted to water level extremes, and buffer land-lake exchanges of nutrients and other materials. Clearly, LULC data such as these will be powerful raw materials for developing environmental indicators for near-shore areas of the Great Lakes. The concomitant gathering of these watershed-level LULC trends with field biological measurements made in these near-shore zones will increase understanding of the effects of human activities on the health of the Great Lakes and their near-shore ecology (Niemi et al. 2004, Danz et al. 2005). ACKNOWLEDGMENTS This research has been supported by a grant from the National Aeronautics and Space Administration (NAG5-11262-Sup 5) and through a cooperative agreement with U.S. Environmental Protection Agency’s Science to Achieve Results (STAR) Estuarine and Great Lakes (EaGLe) program through funding to Great Lakes Environmental Indicators (GLEI), U.S. EPA Agreement R828675-00. Although the research described in this article has been funded in part by the United States Environmental Protection Agency, it has not been subjected to the agency’s required peer and policy review and therefore does not necessarily reflect the views of the agency and no official endorsement should be inferred. The authors would like to thank Tom Hollenhorst and two anonymous reviewers for improving the quality of this manuscript. This is publication number 413 for the Center for Water and the Environment and number 52 for the NRRI NRGIS Laboratory.

Wolter et al.

624 REFERENCES

Anderson, J.R., Hardy, E.E, Roach, J.T., and Witmer, R.E. 1976. A land use and land cover classification system for use with remote sensor data. U.S. Geological Survey Professional Paper 964. Auch, R., Taylor, J., and Acevado, W. 2004. Urban growth in American cities: Glimpses of U.S. urbanization. U.S. Geological Survey Circular 1252. U.S. Geological Survey, Sioux Falls, SD. Brim Box, J., and Mossa, J. 1999. Sediment, land use, and freshwater mussels: prospects and problems. J. North Am. Benthol. Soc. 18(1):99–117. Brown, D.G., Pijanowski, B.C., and Duh, J.D. 2000. Modeling the relationships between land-use and land-cover on private lands in the Upper Midwest, USA. J. Environ. Manage. 59:247–263. Climate Change Science Program. 2003. Strategic plan for the U.S. Climate Change Science Program. Climate Change Science Program and the Subcommittee for Global Change Research, Washington DC. Coats, R.N., and Miller, T.O. 1981. Cumulative silvicultural impacts on watersheds: A hydrologic and regulatory dilemma, Environ. Manage. 5(2):147–160. Cummings, T.R., 1978. Agricultural Land Use and Water Quality in the Upper St. Joseph River Basin, Michigan. U.S. Geological Survey Open-File Report 78-950. Daniel, T.C., Sharpley, A.N., Edwards, D.R., Wedepohl, R., and Lemunyon, J.L. 1994. Minimizing surface water eutrophication from agriculture by phosphorus management. J. Soil Water Cons. 49:30–38. Danz N.P., Regal, R.R., Niemi, G.J., Brady, V.J., Hollenhorst, T., Johnson, L.B.J, Host, G.E, Hanowski, J.M., Johnston, C.A., Brown, T., Kingston, J., and Kelly, J.R. 2005. Environmentally stratified sampling design for the development of Great Lakes environmental indicators. Environ. Monit. Assess. 102:41–65. Ensign, S.H., and Mallin, M.A. 2001. Stream water quality changes following timber harvest in a coastal plain swamp forest. Wat. Res. 35(14):3381–3390. Environment Canada and U.S. Environmental Protection Agency, 2003. The State of the Great Lakes 2003. U.S EPA 905-R-03-004, Governments of Canada and the U.S. ERDAS. 1999. ERDAS Field Guide: Fifth Edition, version 8.5. ERDAS, Inc. Atlanta, Georgia. Frank, R., Braun, H.E., van Hove Holdrinet, M., Sirons, G.J., and Ripley, B.D. 1982. Agriculture and water quality in the Canadian Great Lakes Basin: V. Pesticide use in 11 agricultural watersheds and presence in stream water, 1975–77. J. Environ. Qual. 11:497–505. Gaynor, J.D., MacTavish, D.C., and Findlay, W.I. 1995. Atrazine and metolachlor loss in surface and subsurface runoff from three tillage treatments in corn. J. Environ. Qual. 24:246–256. Gutman, G., Janetos, A., Justice, C., Moran, E., Mustard, J., Rindfuss, R., Skole, D., and Turner II, B.J. (eds.)

2004. Land Change Science: Observing, Monitoring, and Understanding Trajectories of Change on the Earth’s Surface. New York, NY: Kluwer Academic Publishers. Hanowski, J., Danz, N., Howe, R., Niemi, G., and Smith, C. 2006. Considerations of scale and space in the development of Great Lakes coastal wetland bird indicators. EcoHealth. In press. Harris, H.J., Kraft, C., and Wiesman, J. 1993. The State of the Bay—A Watershed Perspective, Institute for Land and Water Studies, University of WisconsinGreen Bay. Green Bay, WI. Homer, C.G., Huang, C., Yang, L., Wylie, B., and Coan, M. 2004. Development of a 2001 National Land-cover Database for the United States. Photogramm. Engin. Remote Sens. 70 (7):829–840. International Geosphere Biosphere Program. 1999. Land-use and land-cover change implementation strategy. In IGBP Report 48, C. Nunes and J.J. Augé, (eds.). International Geosphere Biosphere Programme Secretariat, Stockholm, Sweden. Jackson, L.E., Kurtz, J.C., and Fisher, W.C. 2000. Evaluation Guidelines for Ecological Indicators. U.S. Environ. Protection Agency, Office of Research and Development, Research Triangle Park, North Carolina, USA. Jaynes, D.B., Hatfield, J.L., and Meek, D.W. 1999. Water quality in Walnut Creek watershed: herbicides and nitrate in surface waters. J. Environ. Qual. 28:45–59. Johnston, C.A. 1992. Land use activities and western Lake Superior water quality. In: Making a Great Lake Superior. In Proc. 1990 Int. Conf. on Remedial Action Plans in the Lake Superior Basin, J. Vander Wal and P.D. Watts, (eds.), pp. 145–162. Lakehead University Centre for Northern Studies Occasional Paper #9, Thunder Bay, Canada. ———. 2003. Shrub species as indicators of wetland sedimentation. Wetlands 23:911–920. Kellog, W.A. 1997. Metropolitan growth and the local role in surface water resource protection in the Lake Erie basin. J. Great Lakes Res. 23:270–285. Leatherberry, E.C., and Spencer, J.S. 1996. Michigan Forest Statistics, 1993. Resource Bulletin NC-170, St. Paul, MN: USDA Forest Service, North Central Forest Experiment Station. Marks, D.R. 2003. Habitat and landscape associations of breeding birds in Great Lakes coastal wetlands. M.Sc. thesis, University of Wisconsin-Green Bay, Green Bay, Wisconsin. Martin, C.W., Hornbeck, J.W., Likens, G.E., and Buso, D.C. 2000. Impacts of intensive harvesting on hydrology and nutrient dynamics of northern hardwood forests. Can. J. Fish. Aquat. Sci. 57(2):19–29. McMahon, G., Benjamin, S.P., Clarke, K., Findley, J.E., Fisher, R.N., Graf, W.L., Gundersen, L.C., Jones, J.W., Loveland, T.R., Roth, K.S, Usery, E.L., and

Land Use Land Cover Change in the Great Lakes Basin Wood, N.J. 2005. Geography for a changing world— A science strategy for the geographic research of the U.S. Geological Survey, 2005–2015. U.S. Geological Survey Circular 1281. U.S. Geological Survey, Sioux Falls, SD. Michigan Economic and Environmental Roundtable. 2001. Michigan land resources project. Frey Foundation and the W.K. Kellogg Foundation (Levy, D., editor), (available: http://www.pscinc.com/Documents/ lbilu/index.htm). Miles, P.D., and Chen, C.M. 1992. Minnesota Forest Statistics, 1990. Resource Bulletin NC-141. St. Paul, MN. USDA Forest Service, North Central Forest Experiment Station. Minnesota DNR. 2003. Minnesota’s Forest Resources. Department of Natural Resources, Division of Forestry publication, 500 Lafayette Road, St. Paul, MN, (available: http://files.dnr.state.mn.us/forestry /um/mnforestresources.pdf). National Research Council (NRC). 2001. Compensating for Wetland Losses Under the Clean Water Act. Washington, DC: National Academy Press. Neilsen, G.H., Culley, J.L.B., and Cameron, D.R. 1982. Agriculture and water quality in the Canadian Great Lakes Basin: IV. Nitrogen. J. Environ. Qual. 11:493–496. Niemi, G.J., Wardrop, D., Brooks, R., Anderson, S., Brady, V., Paerl, H., Rakocinski, C., Brouwer, M., Levinson, B., and McDonald, M. 2004. Rationale for a new generation of ecological indicators for coastal waters. Environ. Health Persp. 112(9):979–986. Pastor, J., and Wolter, P.T. 2002. Mapping and Modeling Forest Change in a Boreal Landscape. In Proc. NASA LCLUC Science Team Meeting, Washington D.C., 19–21 November. Price, S.J., Marks, D.R., Howe, R.W., Hanowski, J., and Niemi, G.J. 2005. The importance of spatial scale for conservation and assessment of anuran populations in coastal wetlands of the western Great Lakes. Landscape Ecology 20:441–454. Reese, H.M., Lillesand, T.M., Nagel, D.E., Stewart, J.S., Goldmann, R.A., Simmons, T.E., Chipman, J.W., and Tessar, P.A. 2002. Statewide land cover derived from multiseasonal Landsat TM data—A retrospective of the WISCLAND project. Remote Sensing of Environment 82(2–3):224–237. Sgro, G.V., Ketterer, M.E., and Johansen, J.R. 2006. Ecology and assessment of the diatom communities of four Lake Erie estuaries using Lange-Bertalot Tolerence Values. Hydrobiologia 561(1):239–249. Schmidt, T.L. 1997. Wisconsin Forest Statistics, 1996. Resource Bulletin NC-183, St. Paul, MN: USDA Forest Service, North Central Forest Experiment Station.

625

Schueler, T.R. 1994. The importance of imperviousness. Watershed Protection Techniques 1(3):100–111. Thorp, S., Rivers, R., and Pebbles, V. 1997. Impacts of Changing Land Use. State of the Lakes Ecosystem Conference (SOLEC). Ottawa and Washington: Environment Canada and U.S. EPA. U.S. Environmental Protection Agency [EPA]. 1988. Nonpoint Source Pollution in the US: Report to Congress. Office of Water, Criteria and Standards Division. USEPA, Washington, D.C. ———. 1997. United States Great Lakes Program Report on the Great Lakes Water Quality Agreement (December). EPA-160-R-97-005. U.S. National Oceanic and Atmospheric Administration [NOAA]. Coastal Services Center. 2003a. Great Lakes Accuracy Assessment Report. Internal report prepared by Perot Systems Government Services. (Project 12, Event 9.01). Charleston, SC: Coastal Services Center. ———. 2003b. Michigan Accuracy Assessment Report. Internal report prepared by Perot Systems Government Services. (Project 12, Event 5.02). Charleston, SC: Coastal Services Center. Vogelmann, J.E., Howard, S.M., Yang, L., Larson, C.R., Wylie, B.K., and Van Driel, N. 2001. Completion of the 1990s National Land Cover Data Set for the Conterminous United States from Landsat Thematic Mapper Data and Ancillary Data Sources. Photogramm. Engin. Remote Sens. 67 (6):650–662. Wegener, M.W. 2001. Long-term land use/cover change patterns in the Yahara Lakes region and their impacts on runoff volume to Lake Mendota. M.Sc. thesis, University of Wisconsin-Madison, Madison, Wisconsin. Wickham, J.D, Stehman, S.V., Smith, J.H., and Yang, L. 2004. Thematic accuracy of MRLC-NLCD land cover for the western United States. Remote Sens. Environ. 91:452–468. Wolter, P.T., and White, M.A. 2002. Recent forest cover type transitions and landscape structural changes in northeast Minnesota. Landscape Ecology 17:133–155. ——— , Mladenoff, D.J., Host, G.E., and Crow, T.R. 1995. Improved forest classification in the northern Lake State using multi-temporal Landsat imagery. Photogramm. Engin. Remote Sens. 61(9):1129–1143. Yang, L., Homer, C.G, Hegge, K., Huang, C., Wylie, B., and Reed, B. 2001. A Landsat 7 scene selection strategy for a National Land Cover Database. In Proc. IEEE 2001 International Geoscience and Remote Sensing Symposium. Sydney, Australia, unpaginated CD-ROM. Submitted: 10 October 2005 Accepted: 8 June 2006 Editorial handling: David J. Schwab

Golf course Grassland

Grass, domestic Grass, native Grass, cool season

Coniferous Jack pine Red pine White spruce Mixed/other coniferous

Broad-leaved deciduous Aspen Oak Northern pin oak Red oak Maple Sugar maple Mixed/other broad-leaved deciduous

Jack pine Red pine Spruce-fir Conifer, regeneration Conifer, low density

Aspen-birch Northern pin oak Red oak Northern hardwoods hardwoods, misc. (lowland)

Agriculture Herbaceous/field crops Row crops Corn Other row crops Forage crops Cranberry bog

WISCLAND High intensity Low intensity

NE Minnesota

Aspen/birch Northern hardwood Oak

Mixed conifer < 70% crown closure Mixed conifer > 70% crown closure Mixed pine Jack pine Red pine White pine White spruce < 70% crown closure White spruce > 70% crown closure Balsam fir < 70% crown closure Balsam fir > 70% crown closure Hemlock < 70% crown closure Hemlock > 70% crown closure

Agricultural-cropland

Herbaceous open land

Urban

Upper Michigan GAP

Aspen/birch Oak Northern hardwood Other broad-leaved deciduous forest

Upland jack pine Red pine White pine Other coniferous forest

Agricultural crops Orchard/vineyard

Urban grassland Herbaceous open land

High intensity urban Low intensity urban Extractive

Lower Michigan GAP

LULC classifications used to produce NLCD1992-compatible LULC data for 2001 Minnesota, Wisconsin, and Michi-

Developed Roads

APPENDIX 1. gan. 626 Wolter et al.

Mixed deciduous/coniferous

Open water Emergent/wet meadow Floating aquatic herbaceous vegetation Lowland shrub Broad-leaved deciduous shrub Broad-leaved evergreen shrub Needle-leaved shrub

Forested wetland Broad-leaved deciduous wet forest Mixed deciduous/coniferous wet forest Coniferous wet forest

Barren Shrubland

Jack pine - hardwood Jack pine - oak Red pine - hardwood Oak - pine Spruce-fir - hardwood Aspen-birch—conifer mix Aspen-birch—conifers in understory Northern hardwoods - conifer mix Northern hwd, conifers in understory

Open water Emergent Grass, native (lowland) Floating aquatic Sphagnum spp. Brush, alder (lowland) Brush, willow (lowland) Brush, misc. (lowland) Brush, ericaceous

Black ash Black ash - conifer mix Black ash - conifers in understory

White cedar White cedar - hardwood Tamarack Black spruce Acid bog conifer, stagnant

Bare ground Brush, alder (upland) Brush, willow (upland) Brush, misc. (upland)

Non-vegetative Shrubland

White cedar < 70% crown closure White cedar > 70% crown closure Tamarack Black spruce < 70% crown closure Black spruce > 70% crown closure

Lowland hardwoods Wet hardwood/conifer mix

Open water Wetlands

Dry hardwood/conifer mix

Barren land Shrubland

White cedar Cedar/spruce/fir Black spruce Lowland jack pine Mixed lowland conifer/hardwood

Other forested wetland Mixed lowland hardwood

Water Emergent wetland/wet meadow Other lowland shrub Lowland broad-leaved deciduous shrub Lowland broad-leaved evergreen shrub Lowland needle-leaved evergreen shrub

Oak/jack pine Northern hardwood/conifer

Land Use Land Cover Change in the Great Lakes Basin 627

Barren/hard-surface/rubble/gravel

Perennial herbaceous

Agriculture: row crop Agriculture: pasture and grasslands

Deciduous forest (upland) Mixed forest (upland) Evergreen forest (upland) Deciduous woodland (50-75) Deciduous successional shrubland

Swamp woodland (deciduous) Floodplain forest (deciduous)

Open water

Sparsely vegetated/unvegetated Herbaceous Deciduous shrubland

Agricultural, corn Agricultural, soybeans Agricultural, winter wheat Agricultural, other small grains and hay Agricultural, winter wheat/soybeans Agricultural, rural grassland Agricultural, other

Forest, upland Forest, upland (dry) Forest, upland (dry-mesic) Forest, upland (mesic) Forest, coniferous Forest, partial canopy/savanna upland

Wetland, shallow marsh/wet meadow Wetland, deep marsh Wetland, seasonally/temporarily flooded Wetland, floodplain forest Wetland, floodplain forest: mesic Wetland, floodplain forest: wet-mesic Wetland, floodplain forest: wet Wetland, swamp Wetland, shallow water

Other, surface water Other, barren and exposed land Other

Water

Hardwood Conifer Mixed wood Transitional

Urban: high density Urban: low density Developed: other (non-vegetated)

Developed, high density Developed, medium density Developed, low density Developed, urban open space

(classes split into: Urban, Suburban, Rural)

Indiana GAP

Pennsylvania GAP

Barren Alpine meadow/rock/heath summit Sand flats/slope Successional shrub

Deciduous wetland Evergreen wetland Mixed wetland Emergent marsh/open fen/wet meadow Salt marsh Dwarf shrub bog Shrub swamp Salt shrub/maritime shrubland Open water

Sugar maple mesic Oak Successional hardwoods Spruce-fir Evergreen plantation Alpine krummholtz Evergreen-northern hardwood Pitch pine-oak Appalachian oak-pine

Cropland Orchard/vineyard Old field/pasture

Urban Sub-urban residential Roads Golf course/park/lawn

New York GAP

LULC classifications used to produce NLCD1992-compatible LULC data for 2001 in Illinois, Indiana, Pennsylvania,

Illinois GAP

APPENDIX 2. and New York.

628 Wolter et al.