Environ. Sci. Technol. 2010, 44, 7890–7896
Human-Impacted Water Resources: Domain Stratification and Mapping To Determine Hydrologically Similar Units K A S E Y J . H U T C H I N S O N , * ,† DAVID A. HAYNES,‡ AND JERALD L. SCHNOOR† Department of Civil and Environmental Engineering, 4105 Seamans Center, The University of Iowa, Iowa City, Iowa 52242, and Department of Geography, 316 Jessup Hall, The University of Iowa, Iowa City, Iowa 52242
Received May 27, 2010. Revised manuscript received August 28, 2010. Accepted September 2, 2010.
Geographic Information Systems (GIS) and multivariate statistical analyses were used to partition the United States into Human Influenced Water Environmental Classes (HIWECs) for the purpose of determining hydrologically similar units within the conterminous United States. Such a framework could be used to investigate various categories of watersheds throughout the country or to establish observatory sites in variant hydrologies. These HIWECs represent areas that are relatively homogeneous with respect to the human influence variables of land cover, population density, and water use; the climate variables of temperature and precipitation; and the physical variables of slope, bedrock permeability, and soil permeability. These variables combine to characterize hydrologic condition and collectively represent the major hydrologic variability that exists across the U.S. GIS was first used to break each of the variables into low to high ranges. Multivariate statistics were then used to identify and cluster areas that share similar characteristics in multivariate data space. HIWECS can serve as a framework for development of water research and management; observatory sites can be selected to capture variant hydrologies that incorporate the human influence element with the criteria that each type of human-influenced hydrologic condition, or class, be represented.
Introduction Beginning with the industrial era, population growth, and economic development have caused global water use to grow exponentially (1). Natural processes are extensively impacted by humans, linking anthropogenic alteration to the hydrologic system. It thus becomes necessary to consider human influence in addition to land-surface form, geologic texture, and climate when characterizing hydrologic setting as well as predicting future system responses (2-4). To successfully manage future water issues, it is necessary to assess water quality and quantity as well as to make links to the factors affecting each (1). A Water and Environmental * Corresponding author phone: (319)358-3617; fax: (319)358-3606; e-mail:
[email protected]. Corresponding author address: USGS Iowa Water Science Center, Iowa City, IA 52244. † Department of Civil and Environmental Engineering. ‡ Department of Geography. 7890
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 20, 2010
Research Systems Network (WATERS Network) was originally proposed to make these links by constructing a conceptual design of a national water quality monitoring network, the impetus of this study. The goal of the network was to analyze and forecast water quality and quantity on a national level by gaining a better understanding of the dynamics and spatial variability of environmental processes as well as depict the impacts of anthropogenic and climate change on the hydrologic cycle (5). While the development of WATERS Network is no longer likely the need for such a national network still exists. To meet the goals of such a network, experimental observatories and field facilities would have to be placed in a way to sufficiently capture the variation of major hydrologic conditions that exist across the nation. Therefore, the placement of facilities becomes critical in the design and development of an observatory network. The National WaterQuality Assessment (NAWQA) program, launched in 1991, distributed 60 different study areas in such a manner to capture 60-70% of the nation’s publicly supplied water use and population (6). Budget constraints in 2001 required the number of study sites to be reduced. Wolock (2) created Hydrologic Landscape Regions (HLRs) to provide a means to reduce the number of required sites while maintaining a representative sample of hydrologic settings by partitioning the United States into zones of similar hydrology based on the hydrologic-landscapes concept of Winter (7). Underlying this HLR concept is the idea that the same characteristics control the movement of water through a landscape regardless of its geographic location, and thus diverse landscapes can share many common features with respect to water movement (2). Land surface form, geologic framework, and climatic characteristics were used to create these regions of similar hydrologic condition. TheNationalEnvironmentalObservatoryNetwork(NEON), a continental-scale research platform, focuses on monitoring and assessing ecology in response to climate change, landuse change, invasive species, and the interaction with the atmosphere, hydrosphere, and geosphere (8). Multivariate geographic clustering was used to subdivide the United States into subdivisions for purposes of systematically sampling the ecological variability of the country (8). In both NAWQA and NEON cases, the placement of study sites is based on an objective delineation of regions of relative homogeneity with respect to the selected factors. Geographic frameworks, such as the HLRs and NEON divisions, have been used in natural resource management for decades. Baily initiated extensive use of macroscale patterns of interdependent factors when he subdivided the Earth’s surface and created Ecoregions of the United States in 1976 (9). Omernik’s 1987 ecoregions, widely used today, were produced by combining individual map layers representing various characteristics to create a pattern of combined “causal and integrative factors” that together define regions of homogeneity (10). The factors used to define homogeneous regions will vary depending on the objective at hand. The resulting geographic framework can provide a means to visualize similarities across multidimensional environmental characteristics (11). This study provides such a framework but incorporates a necessary factor not previously considered: human influence.
Methods ArcMap 9.2 was used to create a national-scale map with physical, climatic, and human influence variables stratified 10.1021/es101582c
2010 American Chemical Society
Published on Web 09/27/2010
into categories of low to high values resulting in varying numbers of categories, listed in Table S1. All eight map layers representing each of the variables were then combined, thereby characterizing a location by its combination of values. The combination of stratified variables then served as input to ENVI 4.4 (12), geospatial imagery software, for subsequent cluster analysis to delineate the United States into HIWECs that represent areas of relative homogeneity with respect to the variables. Physical characteristics consist of slope, soil permeability, and bedrock permeability. Climate characteristics consist of temperature and precipitation. Land cover, water withdrawals, and population density were chosen as additional variables to represent the third component of human influence, not incorporated in the HLR development. Readily available, publicly accessible data were used in this study and was provided by the United States Geological Survey (USGS) (13, 14), the USGS Multi-Resolution Land Characteristics Consortium (MRLC) (15), the USGS National Atlas (16, 17), and PRISM Group at Oregon State University (18-21). All data sets were projected to the same Albers Equal Area Conic, NAD 83, and all analyses were conducted at 1 km resolution. This method is meant to provide a means to select a minimum number of sites that effectively capture the hydrologic variability that exists across the nation. HIWEC Development. Each of the data sets were first divided into ranges of values with the number of resulting ranges varying among the data sets, depending on how to best represent each. For example, population density was divided into three different ranges representing low, medium, and high population density categories, while bedrock was broken into seven categories representing seven different lithologic groups. Table S1 lists the resulting categories for each variable. Break values were set so that the data were distributed in a manner to represent low-to-high conditions of the country for that specific variable. This was done essentially to aid in the interpretation of final results by setting known bounds in the data. Various methods were used to determine category breaks; subjective judgment was necessary for some of these decisions, but sensitivity analyses were used to ensure that small changes in the data do not result in large changes in the results. Because land cover type and bedrock permeability are categorical data, values were assigned to each category representing the relative infiltration ability and bedrock permeability for land cover type and lithologic group, respectively. Next, all variables were nondimensionalized by scaling the data, assigning each category for each variable a consecutive number up to seven; seven is the greatest number of categories out of all variables (land use). Categories of higher numerical values were assigned higher scaled values, and categories of lower numerical values were assigned smaller scaled values i.e. the high population category was assigned a value of seven. Table S1 lists the assigned classified values for each category. Results of Variable Stratification. Human Influence Variables. Population Density. The population density data set obtained from the United States Geological Survey (USGS) represents the 2000 population density by block group (13). The United States Census Bureau defines urbanized area as 1000 persons per square mile or 386 persons per square kilometer. This number was rounded to 400 and set as the upper category break to roughly represent an urban category as defined by the Census Bureau. Various classification schemes were used and thematic maps created to determine where to set other category breaks. In order to achieve a satisfactory distinction between the population densities of the mountain West versus the East, and the population density of the rural Midwest versus the rural East Coast, category breaks of four and 40 people per
square kilometer were chosen, breaking population density into a total of four classes, as shown in Table S1. Water Use. The digital 2000 water use file was created by the USGS National Atlas of the United States (17). It includes both ground and surface, fresh and saline withdrawals by all sectors. The field of total gallons per year per county was converted to a field of millimeters per year per county for this data set. This data set is aggregated to county level; therefore, the same total water use value is assigned to an entire county. In order to better reflect water use trends this data set was modified to reflect the categorization of satellite imagery. Pixels of land cover data sets designated as urban were given the urban water use value for the county for which the pixel belonged to and pixels that were labeled rural were given the rural water use value. The data set was broken into five categories as listed in Table S1. Land Cover. Land cover data sets were obtained from the National Land Cover Database 2001 (NLCD 2001) at http:// www.mrlc.gov/, which has been compiled for the contiguous United States by the Multi-Resolution Land Characteristic (MRLC) Consortium (15). The NLCD 2001 digital data set breaks land use into 16 categories. For this study, similar land cover classes were combined into single classes to reduce the total cover types to five, which include water, forest, agriculture, urban, and other. Class assignment was based on the land cover descriptions provided by the MRLC for each land cover class. Table S2 shows the land cover type class assignments. Figure 1 shows the map classifications for population density, water use, and land cover class. Climate Variables. Precipitation. The digital data set for average annual precipitation for 1971-2000 was obtained from the PRISM (Parameter-elevation Regressions on Independent Slopes Model) Group at Oregon State University at http://www.prism.oregonstate.edu/ (19). The PRISM Group produced this data set through development and use of the PRISM model. Three category breaks of less than 380, between 380 and 1140, and greater than 1140 mm per year were set to represent arid, semihumid, and humid regions, as shown in Table S1. The category breaks were set to reflect precipitation trends as seen in the Natural Resources Conservation Service, National Weather and Climate Service annual precipitation maps and NOAA regional climate maps. Temperature. The digital data set for the United States average monthly temperature for 1971-2000 was obtained from the PRISM Group at Oregon State University. Two data sets are provided; one for average daily maximum temperature and one for average daily minimum temperature (20, 21). The average of these two data sets was calculated via Spatial Analyst Extension and was broken into five categories. The category breaks are shown in Table S1. Figure 2 shows the classification of United States average annual precipitation and average temperature for 1971-2000. Physical Variables. Soil Permeability. Soil permeability was estimated by averaging the high and low permeability limits (inches per hour) from the STATSGO soil characteristics for the conterminous United States digital data set, obtained at http://water.usgs.gov/lookup/getgislist (14). The United States Department of Agriculture (USDA) provides a designation of permeability classes based on inches per hour. The six defined classes were reduced to three classes of slow, moderate, and rapid permeability by combining the classes and are shown in Table S1. Bedrock Permeability. Wolock (2) estimated bedrock permeability based on the lithologic groups of the principal aquifers of the United States, assigning each lithologic group a number representing its relative permeability. The classes developed by Wolock have been used in this study. The digital map layer is a shapefile that contains the shallowest principal aquifers of the conterminous United States and was provided by the USGS National Atlas at http://www.nationalatlas.gov/ VOL. 44, NO. 20, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
7891
FIGURE 1. Human influence variables as represented by (A) 2000 population density by block group, (B) 2000 water withdrawals, and (C) land cover classes.
FIGURE 2. Classification of (A) average annual precipitation for 1971-2000 and (B) average annual temperature for 1971-2000. (16). Each group was assigned scaled values in order of low to high permeability based on the lithologic groups of each aquifer. Table S1 shows the lithologic classes in order of high to low permeability. Slope. Slope was derived from a national Digital Elevation Model (DEM) produced and provided by the PRISM Group at Oregon State University (18). ArcMap was used to derive the slope from the DEM. Percent slope was broken into the four categories listed in Table S1. Figure 3 shows the resulting map classifications for soil permeability, bedrock permeability, and slope. All of these variables combine to define hydrologic condition, controlling the spatial distribution and quality of water resources. A final map, shown in Figure S1, shows the spatial distribution of water stress, calculated by taking the ratio of water use intensity to precipitation rate and then dividing by the area from which the withdrawals were taken. The map illustrates that moderate to high stress extends to 7892
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 20, 2010
all regions of the country; water management issues are not restricted to arid or heavily populated regions but rather are a function of how all the variables combine to create hydrologic environments.
Results and Discussion After classification of the variables, a multidimensional data set was created by combining all layers into a single layer using ERDAS Imagine 9.0. Each pixel represents a single value, recording the combination of all eight individual scores on each of the eight variables. The resulting image file was imported into ENVI 4.4 as well as converted to ASCII for import into SPSS 15 for subsequent statistical analysis. Statistical Analyses: Principal Components and ISODATA Cluster Analyses. PCA is a multivariate statistical method used to reduce the dimensionality of data by extracting a smaller set of factors from overlapping variables
FIGURE 3. Classification of (A) average soil permeability, (B) bedrock permeability, and (C) slope.
FIGURE 4. Resulting 12 HIWECs. and grouping them into a set of uncorrelated “components” (22, 23). Each component represents a different “dimension” of the data, listed in order from largest to smallest eigenvalues, representing the importance with regard to explaining the variance of the data (24). If the original variables are uncorrelated, the smaller number of transformed variables will fail to explain the variance much better than the individual variables do alone. It is convention to include only those variables that show a correlation of the order of 0.3 with at least one other variable in the analysis (25). The data were imported into SPSS 15 for PCA; in this case, the correlation among variables was very weak with only precipitation and land cover meeting the 0.3 criteria. This resulted in ambiguous, or unexplainable, PCA results and suggested that the dimensionality of the data could not be reduced. The lack of underlying structure within the data could be partly attributed to the complexity and unpredictability of the human element. For example, high water use values correlate with areas of high population density. However, there are significant areas in the west where irrigation farming dominates. In this case, high water use is
correlated with very small population densities, and the resulting overall correlation is thus much lower than expected. Because the number of variables is small, the PCA results ambiguous, and the correlations between variables weak, the cluster analysis was based on the raw data rather than the PCA results. The cluster analysis is used to create groups of relative homogeneity with regard to the variables used. This reduces the data set into a smaller number of groups, which in this case are the resulting HIWECs. Criteria can be selected to control the number of resulting groups, in this case into a more manageable number of HIWECs. ENVI 4.4 was used for the cluster analysis on the original data sets. This program was selected due to its computational efficiency in conjunction with its mapping capabilities. The ISODATA (Iterative Self-Organizing Data Analysis Technique) clustering algorithm was chosen as the unsupervised classification method for this study. This algorithm first calculates class means evenly distributed in data space. It then uses minimum distance techniques to iteratively cluster the remaining pixels, recalculating the mean and reclassifying pixels with respect to the new means for each consecutive iteration (26). When performing the cluster analysis within ENVI, a range of classes must be selected prior to analysis. The main objectives when deciding on site number and locations is to minimize the cost of the effort while still adequately representing the various hydrologic conditions across the country. For this analysis we selected a range of number of classes that was considered feasible for the WATERS Network and would be suitable to meet the network’s goals. The low range value was set to five and the upper range was capped at 20 because the initial number of classes (arbitrarily) assigned is equal to the highest number in the selected range, and a higher initial value will result in a higher number of classes in the final clustering results. The cluster analysis, therefore, is very sensitive to initial starting values for class range. VOL. 44, NO. 20, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
7893
TABLE 1. Characteristic Descriptions for Each Individual HIWEC in Ascending Rank Order from the Least to the Most Humid Classes for 12 HIWECS Shown in Figure 4 class one (blue) class two (light green)
class three (red)
class four (maroon)
class five (coral)
class six (yellow)
class seven (turquoise)
class eight (magenta)
class nine (green) class ten (purple) class eleven (black) class twelve (mauve)
Characterized by very low to mid bedrock permeability (mid high to high excluded), mostly other land cover type, mostly low water use, low precipitation, mid high to high slope (low excluded), low soil permeability, and mid to high temperature. Characterized by very low to mid bedrock permeability, mostly other and agricultural land cover type, mostly low water use, low to mid population density (high excluded), low precipitation (mid to high excluded), low to mid low slope, and mid low to mid temperature. Characterized by low bedrock permeability, mostly other and agricultural land cover type, varied water use, low to mid (mostly low) population density (high excluded), low to mid precipitation, mostly low slope, mid to high soil permeability (low excluded), and mostly high temperature. Characterized by mid high to very high bedrock permeability, mostly other and agriculture land cover type (agriculture dominant), varied water use, low to mid low population density (mid high to high excluded), low to mid precipitation (mid high to high excluded), low to mid low slope (high excluded), low to mid soil permeability (high excluded), and mid to high temperature. Characterized by mid high to very high bedrock permeability, other and agriculture land cover type, varied water use, mostly low to mid low population density, low to mid precipitation, low to mid low slope, high soil permeability, and mid low to high temperature. Characterized by very low to low bedrock permeability, mostly agriculture land cover type, varied water use, low to mid low population density (high excluded), mid precipitation (low and high excluded), low to mid low slope, low soil permeability (mid to high excluded), and mostly low to mid temperature. Characterized by very low to mid bedrock permeability, mostly forest land cover type, low water use, low to mid low population density, mid precipitation (low excluded), mid high slope (low excluded), low to mid soil permeability, and low to mid low temperature. Characterized by very low to mid bedrock permeability, mostly forest land cover type, mostly low to mid water use, low to mid low population density (high excluded), mid to high precipitation (low excluded), mid high to high slope, mid to high soil permeability (low excluded), and low to mid low temperature. Characterized by very low to mid low bedrock permeability, varied land cover type, mostly low water use, varied population density, mid to high precipitation, low slope, low soil permeability (mid to high excluded), and mid high to high temperature. Characterized by very low to mid low bedrock permeability, varied land cover type, varied water use, varied population density, mostly high precipitation, low to mid high slope, mid soil permeability, and mid to high temperature. Characterized by mid to very high bedrock permeability, varied but mostly urban land cover type, varied water use, varied but predominantly high population density, high precipitation, low to mid low slope, low soil permeability (high excluded), and mid to high temperature. Characterized by high to very high bedrock permeability, varied land cover type mostly mid to high water use, varied population density, high precipitation, low slope, mid to high soil permeability, and high temperature.
The multiple parameters that must be selected in addition to the range of number of classes are listed in Table S3. Because selection of the threshold parameters is so subjective, multiple analyses were run to determine how sensitive the results are to modifications in the parameters. In addition, multiple runs aid in establishing a compromise between the number of resulting classes and thus sites for an observatory network and what is considered acceptable for the various threshold parameters. The results, shown in Table S3, are most sensitive to changes in the maximum standard deviation and minimum class distance parameters, with the exception of the range of the number of classes. A minimum class distance of five was selected as a reasonably low value, helping maximize the between class difference. Setting this as a constant, the maximum standard deviation was examined. Using the values determined acceptable for all other parameters, the maximum standard deviation was varied, and the results are presented separately in Table S4 to facilitate comparison. In most cases, the resulting number of classes increases with increasing maximum class standard deviation. However, as seen in the table, once the maximum standard deviation reaches a value of 16 the resulting number of classes remains at 12 even with subsequent increases in this parameter. With consecutive one unit incremental decreases from a value of 16 the resulting number of classes increases by one until a maximum standard deviation of 12 is reached. At this value, 7894
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 20, 2010
the resulting number of classes increases to 19. The class range cap of 20 classes is reached with a maximum standard deviation of 11. In order to minimize necessary costs and resources it is desirable for the delineation of HIWECs to sufficiently capture hydrologic variability with a minimum number of classes chosen. Ultimately, the final determination of classes is subjective. The results suggest that a minimum number of 12 classes is a reasonable compromise and is adequate for capturing the major hydrologic variability. Figure 4 shows a map of the resulting 12 HIWECs (27); a description of each class is provided in Table 1. Each HIWEC region represents a set of human-impacted hydrologic conditions, with groupings such that within class variance is minimized and between-class variance is maximized. Map cells that fall within the same region share similar combinations with respect to the eight variables and thus hydrologic condition. Each region has been assigned a different color, and it can be seen that regions are not necessarily contiguous; spatially separated regions of the country can share the same hydrologic characteristics, thus grouping into the same HIWEC. For comparison purposes, the resulting 20 HIWECs, 20 HLRs (2), US EPA Level III Ecoregions (28), and NEON domains (8) are shown in Figure 5. In both the NEON and Ecoregion classifications, the U.S. is broken up into nonrepeating classes or regions. The
FIGURE 5. Visual comparison of the resulting 20 HIWECs, USGS Hydrologic Landscape Regions, USEPA Level III Ecoregions, and NEON domains. HIWECs and HLRs, based on many of the same variables, show the greatest similarities and both seem to be driven by bedrock permeability. A notable difference between the two is the appearance of a human influence group in the HIWECs represented by a small gold class, most visible in the Southeast region. This class does not appear in all heavily populated or high water use areas, stressing that it is the combination of variables that controls the water environment, yet incorporation of human influence variables does create a redefinition of classes relative to the HLRs. For example, western Iowa, southwestern Minnesota, and the Dakotas compose a homogeneous HIWEC class (bright yellow) characterized by wheat/corn agriculture and loamy soils, while this same area is represented by many different classifications in the USGS HLR map. In addition, the Sand Hills of Nebraska and most of Florida make up a single HLR class (reddish brown) but belong to different HIWEC classes. Sandy, permeable soils, and limestone bedrock characterize these locations; it is possible that the separation of the two in the HIWEC map is due to intense water withdrawals for agriculture in the Sand Hills. Another possible reflection of human influence in the form of water withdrawals is represented by the homogeneous blue HLR in the desert Southwest; this region is represented by several different HIWEC classes. The resulting classes can be made as distinctively different as desired. Finer resolution of such delineations will require a greater investment of resources if all classes are to be represented. Such a process can be modified to suit the purpose and feasibility of the objectives at hand. This methodology identifies areas of similarity, incorporating necessary variables, so that an understanding of processes obtained by direct observation in one location can be transferred to an unmonitored location. HIWECs can serve as a regional hydrologic framework for broad applications as well as for specific objectives such as providing a means for placement of observatory network
sites. Sites can be placed such that human-impacted hydrologic variability is adequately represented by locating sites in each of the HIWECs. Locating multiple observatories within the same classification from different parts of the country would be a check on the similarity within a HIWEC class. Such a framework can easily be reproduced at varying resolutions, and additional variables can be incorporated to produce new regions reflecting human and hydrologic characteristics as needed. HIWECs incorporate direct representations of human influence; anthropogenic factors continue to exert greater impact on water resources, and thus HIWECs can provide a more representative classification of the varying hydrologic environments.
Acknowledgments The authors thank the National Science Foundation WATERS Network Project Office, through subcontracts with the University of Illinois Urbana-Champagne (UIUC) National Center for Supercomputer Applications (NCSA), and the University of California Santa Barbara for funding for this project. We also thank Jeremie Moen and Marc Linderman for providing necessary resources and assistance and Danny Reible, David Wolock, Tom Winter, and David Maidment for helpful discussions.
Supporting Information Available Additional information including 4 tables and 1 figure. This material is available free of charge via the Internet at http:// pubs.acs.org.
Literature Cited (1) Vorosmarty, C. J.; Sahagian, D. Anthropogenic Disturbance of the Terrestrial Water Cycle. BioScience 2000, 50, 753–765. (2) Wolock, D. M.; Winter, T. C.; McMahon, G. Delineation and Evaluation of Hydrologic-Landscape Regions in the United States Using Geographic Information System Tools and MulVOL. 44, NO. 20, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
7895
(3) (4) (5)
(6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
7896
tivariate Statistical Analyses. Environ. Manage. 2004, 34, S71– S88. Oki, T.; Kanae, S. Global Hydrological Cycles and World Water Resources. Science 2006, 313, 1068–1072. Alley, W. M.; Leake, S. A. The Journey from Safe Yield to Sustainability. Ground Water 2004, 42, 12–16. Montgomery, J. M.; Harmon, T.; Haas, C. N.; Hooper, R.; Clesceri, N. L.; Graham, W.; Kaiser, W.; Sanderson, A.; Minsker, B.; Schnoor, J.; Brezonik, P. The WATERS Network: An Integrated Environmental Observatory Network for Water Research. Environ. Sci. Technol. 2007, 41 (19), 6642–6647. Leahy, P. P. National Water-Quality Assessment Program. USGS Numbered Series; 1991; pp 91-54. Winter, T. C. The Concept of Hydrologic Landscapes. J. Am. Water Resour. Assoc. 2001, 37 (2), 335–349. National Ecological Observatory Network, NEON. In Overview. Retrieved March 12, 2008, from http://www.neoninc.org/aboutneon/overview.html. Bailey, R. G. Identifying Ecoregion Boundaries. Environ. Manage. 2004, 34, S14–S26. Omernik, J. M. Ecoregions of the Conterminous United States. Ann. Assoc. Am. Geographers 1987, 77, 118–125. Hargrove, W. W.; Hoffman, F. M. Potential of Multivariate Quantitative Methods for Delineation and Visualization of Ecoregions. Environ. Manage. 2004, 34, S39–S60. ITT Visual Information Solutions. ENVI Software. 2010. Available at http://www.ittvis.com/ (accessed month day, year). Hitt, K. J. 2000 population density by block group for the conterminous United States, 200311; U.S. Geological Survey, Reston, VA. Wolock, D. M. STATSGO soil characteristics for the conterminous United States, 1997; U.S. Geological Survey, Lawrence, KS. U.S. Geological Survey, 2001 National Land Cover Database (NLCD), 200309; Multi-Resolution Land Characteristics (MRLC) Consortium, Sioux Falls, SD. U.S. Geological Survey, Principal Aquifers of the 48 Conterminous United States, Hawaii, Puerto Rico, and the U.S. Virgin Islands, 200310; U.S. Geological Survey, Madison, WI, USA.
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 20, 2010
(17) U.S. Geological Survey, Water Resources of the United States, Estimated Use of Water in the United States, 2000, 200509; National Atlas of the United States, Reston, VA. (18) The PRISM Group at Oregon State University, DEM Terrain Elevation, 30 Arcseconds, 200706; The PRISM Group at Oregon State University, Corvallis, OR. (19) The PRISM Group at Oregon State University, United States Average Monthly or Annual Precipitation, 1971-2000, 200606; The PRISM Group at Oregon State University, Corvallis, OR. (20) The PRISM Group at Oregon State University, United States Average Monthly or Annual Maximum Temperature, 19712000, 200606; The PRISM Group at Oregon State University, Corvallis, OR. (21) The PRISM Group at Oregon State University, United States Average Monthly or Annual Minimum Temperature, 1971-2000, 200606; The PRISM Group at Oregon State University, Corvallis, OR. (22) Jackson, J. E. A User’s Guide to Principal Components; John Wiley & Sons, Inc.: Hoboken, 1991. (23) Green, S. B.; Salkind, N. J. Using SPSS for Windows and Macintosh, Analyzing and Understanding Data; Pearson Education, Inc.: Upper Saddle River, 2005. (24) Manly, B. F. J. Multivariate Statistical Methods, A Primer; Chapman & Hall/CRC: Boca Raton, 2005. (25) Kinnear, P. R.; Gray, C. D. SPSS 15 Made Simple; Psychology Press: New York, 2008. (26) ITT Visual Information Solutions. ENVI 4.4. Help Guide; 2007. (27) Hutchinson, K. J. Human-impacted water resources: Domain Stratification and Mapping for Determination of a Potential Observing Network. M.S. Thesis, University of Iowa, Iowa City, 2008. (28) U.S. Environmental Protection Agency, Western Ecology Division. Level III Ecoregions. Retrieved December, 2009, from http://www.epa.gov/wed/pages/ecoregions/level_iii.htm.
ES101582C
