Ecological homogenization of urban America

a research project funded by the U.S. National Science Foundation program on “MacroSystems Biology: Research on Biological Systems at Regional to Continental Scales.”

read more

Detailed Experimental Methods

Overview

 We are addressing our questions and hypotheses across six metropolitan statistical areas (MSA) that cover the major climatic regions of the U.S. (Figure 3).  At the household/parcel scale, we are coupling homeowner surveys with intensive biophysical measurements to determine how land-management practices influence ecological structure and function.  We are compiling extensive, high-resolution (≤1.0m pixels) remotely sensed and socio-demographic data to assess the extent, spatial distribution and quality of lawns and other cover types at the parcel and neighborhood levels.  These data will be used to link social decisions and preferences with ecological patterns and processes at broader (MSA) geographic scales.  Conducting these MSA-scale analyses across diverse regions of the U.S. will allow us to determine if scaling tools based on parcel level data can be used to produce continental-scale assessment of the effects of urban homogenization on ecosystem structure and function.

1. Experimental designs

in each city include intensive measurements of residential parcel plots through social surveys and ecological sampling, and extensive measurements by remote sensing and social data compilation at the neighborhood and MSA scale.  Between 25 and 30 residential parcels will be intensively sampled for soil and vegetation variables, and at least five neighborhoods with similar lifestyle classifications (Table 1) will be sampled extensively in each city.  Sampling parcels will be selected randomly from a set of possible parcels meeting experimental criteria generated from a telephone survey (see below).  The experimental designs will be fine-tuned for each MSA/city to account for variation in local factors related to residential development and native ecosystem gradients.  Key factors that will be incorporated into the design for each city include:

  • Density (e.g., urban, suburban, exurban) – to account for this major source of variation in all cities.
  • Neighborhood age (e.g, <10 versus >50 years old) – a critical controller of vegetation development, aboveground carbon stocks, microclimate.
  • Natural variations in soil or climate within the MSA region that influence ecosystem carbon and nitrogen pools and fluxes.
  • Landscaping style, e.g., xeriscaping versus traditional mesic design is a major contrast in arid regions.
  • Socio-economic characteristics/lifestyle classes/neighborhoods – at least three classes in each city.
  • Land use history – e.g., previous agricultural land use has a significant influence on contemporary ecosystem carbon and nitrogen pools and fluxes.
  • References – native and in some cases agricultural reference sites for comparison in each city.

PRIZM table

Table 1.  PRIZM lifestyle classes common to each of our study cities that will guide selection of intensive sampling parcels and neighborhoods.  PRIZM clusters (neighborhoods) are based on household education, income, occupation, race/ancestry, family composition and housing and are associated with spending on consumer goods and services, including yard care products and services.  Since every U.S. Census Block Group (neighborhood) is assigned a specific PRIZM category, these data provide the basis for scaling from neighborhood to regional (MSA) and continental levels.


It will not be possible to sample full factorial experimental designs accounting for all of these factors in all six cities.  However, a series of orthogonal contrasts can be used to isolate specific factors.  For example, in Baltimore, we will contrast three density, three age, two previous land use and three socio-economic/lifestyle classes with both agricultural and forest reference sites.  By restricting the age comparison to the suburban density class, the previous land use comparison to the exurban density class, and the socio-economic comparison to the urban density class, we can evaluate all these factors with 30 sampling plots (assuming three replicates for each site type).  Multivariate analysis of variance (MANOVA) will be used test for the main effects of age, density, land use history and socio-economic class treating each factor as distinct categorical variables.

study sites

Figure 1.  Study sites across the continental climate gradient (top) and urban land use in the six study cities (bottom). 

           

Central Arizona-Phoenix (CAP).  The CAP Long-Term Ecological Research (LTER) site encompasses the City of Phoenix and surrounding areas within the Sonoran Desert ecosystem, which averages 18 cm of rainfall annually.  Ongoing CAP research is examining the social and urban drivers of ecological structure and land-management practices and, by extension, associated ecosystem functions and services. The experimental design for Phoenix will contrast native and agricultural systems with three socioeconomic classes in the urban core (with an agricultural history) and in the urban fringe (no agricultural history). In the urban core sites, parcels will be stratified by land cover type, including grassy lawn and desert-like ‘xeriscapes’ which are common to cities of the arid southwest.

Miami.  The Florida Coastal Everglades (FCE) LTER program study site lies at the southern end of the greater Everglades watershed, a subtropical wetlands complex situated in a complex mosaic of agricultural and urban land uses. FCE scholars are combining geospatial and ethnographic analyses to uncover how zoning, sociodemographics, and climate events (hurricanes and sea-level rise) interact to impact land use and land cover at household to regional scales.  The experimental design for Miami, which is also the site of an ULTRA-ex project, will contrast three physiographic-geological formations, two economic classes, and two previous land uses, along with  agricultural and forest reference sites.

Boston.  The study area for the Plum Island Ecosystem (PIE) LTER project includes 26 Massachusetts towns in the Boston MSA and the adjacent Plum Island Sound estuary. The PIE-LTER project maintains an extensive database of monitoring and experimental data on land use, climate and watershed biogeochemistry and hydrology. Social processes have been the focus of the Human-Environment Regional Observatory (HERO; http://hero.clarku.edu) NSF REU Site program, which maintains an extensive database on land-use/land-cover, socio-economics, and demographics. The experimental design for Boston will include three forest reference sites, three reference pasture agricultural sites, nine suburban parcels (three young, three medium, three old), six exurban parcels (three former agriculture, three former forest), and nine urban parcels (three replicates each of three socio-economic classes). We will restrict sites to the dominant soil orders with level or gently sloping, well-drained to moderately well-drained soils formed in glacial till (Fuller and Francis 1984, Peragallo 2009).

Baltimore.  The Baltimore MSA developed on temperate forest landscape that was largely cleared for agriculture in the 1800’s.  The Baltimore urban LTER project (BES) includes long-term watershed and permanent plot studies as well as research to understand the social and ecological motivations, capacities, and pathways for increasing land management on private parcels.  The experimental design for Baltimore is described above.

Minneapolis-St.Paul. (Twin Cities). The Twin Cities lie at the intersection of three major biomes (boreal forest, temperate forest, tallgrass prairie) with a cold temperate climate. The Twin Cities Household Ecosystem Project (TCHEP) has surveyed 3000 households along an urban–exurban gradient to determine patterns and sociodemographic drivers of household biogeochemical fluxes (C, N, P), with intensive tree sampling at 360 of these households (Fissore et al. 2010), and quantification of total cultivated and uncultivated plant diversity in 150 of these households. The experimental design for the Twin Cities will contrast three density, two age, two soil type, two socioeconomic classes and forested and agricultural reference sites. The soil type and age class contrasts will be restricted to the exurban density class, while the socioeconomic classes will be restricted to the urban class, for a total of 30 plots.

Los Angeles.  The Los Angeles Basin is a coastal plain surrounded by the peninsular and transverse mountain ranges with an average annual temperature of 18.3°C and precipitation of 38 cm (downtown Los Angeles, Morris 2009).  In an NSF-USFS ULTRA-EX project, 13 neighborhoods of varying land cover, climates, and sociodemographics have been identified for intensive biological inventories in a study of tradeoffs between ecohydrology and other aspects of ecosystem services (Pataki et al. 2010a).  Other efforts have included intensive inventories of the urban forest and direct measurements of C sequestration and studies of the biogeochemistry of urban lawns (Bijoor et al. 2008, Townsend-Small and Czimczik 2010a, Townsend-Small et al. 2010) and unmanaged grasslands (Wang and Pataki 2010b, a). The experimental design for Los Angeles will include three density, three age, and three socioeconomic classes with reference sites in the native coastal sage shrub ecosystem. In addition, there will be three microclimate classes because of the large temperature gradient across the coastal to inland areas of the urban region.

2. Residential management surveys

We will tailor two surveys that our team has already used in four of our six MSAs for this project. The first survey is a household telephone survey that includes questions about land management practices, knowledge and sources of information for land management, watershed knowledge, willingness to participate in environmental activities, and perceptions of neighborhood environmental quality of life.  This survey was overseen by J. Morgan Grove, University of Vermont / Forest Service, in collaboration with Colin Polsky (Clark University) and was conducted by a private survey firm in fall 2012.  The survey included a question about the homeowner’s willingness to have measurements of soil, vegetation, and social variables taken on their property and to participate in a follow-up, face-to-face household survey.  This question was crucial for a) ensuring a sufficient number and distribution of field sites based upon our stratified sample and b) the ability to link the data collected through the telephone survey to the field and face-to-face surveys.   

The second survey will target the managers of the actual parcels in our experimental design and will address motivations, preferences, and attitudes for their lawn management practices, socioeconomic and demographic characteristics, and environmental attitudes.  In addition, the survey will quantify fertilizer and outdoor water use practices based upon on a survey field-tested by the PIE LTER project. The survey instrument will include both closed and open-ended questions for a nuanced understanding of the relationship between lawn care practices and environmental attitudes. The vegetation field survey team (see below) will administer the household survey at each research site. Laura Ogden, an anthropologist, and Rinku Roy Chowdhury, a geographer, will train the field crews to ensure optimal standardization. Survey data generated at this fine spatial scale (i.e. parcel) will allow us to associate lawncare practices and environmental attitudes with other socio-economic variables present in widely available data sources (U.S. Census, cadastral data, PRIZM).

3. Parcel structure remote sensing

  For each study site, we will develop an object-based image analysis (OBIA) system that extracts land cover information using the best available remotely sensed data and ancillary GIS data layers.  OBIA techniques have been found to be superior to pixel-based approaches, even in complex environments, producing highly accurate (>90% overall accuracy) classified land cover data in an efficient and consistent manner (Zhou et al. 2009a). The land cover classes will consist of: tree canopy, grass/shrub, bare soil, water, buildings, roads/railroads, and other paved surfaces.  The foundation of the OBIA system will be an expert system that incorporates image processing, segmentation, classification, morphology, and object fusion algorithms to extract meaningful land cover features.  Parallel processing will facilitate the analysis of multi-billion pixel datasets.  Following the automated analysis a detailed manual review of the data will be conducted at a scale of 1:5000, correcting any misclassified pixels.  Once the manual corrections have been incorporated an accuracy assessment will be conducted following Congalton and Green (1999).  The land cover information will be summarized for both property parcels and Census block groups.  Mapping for all sites will be done at the University of Vermont Spatial Analysis Laboratory (UVM SAL) with interpretation led by Jarlath O’Neill Dunne (UVM SAL) and Rinku Roy Chowdhury (Indiana University).

4. Soil carbon and nitrogen pools

  Undisturbed one-meter (or to maximum depth possible) soil cores will be taken from two randomly selected locations in the dominant cover type in each sampling parcel using a 3.3 cm diameter soil corer.  Cores will be enclosed in plastic sleeves with end-caps, put into coolers, and transported to the Cary Institute of Ecosystem Studies for processing.  Digital photos will be taken of each soil core followed by a visual inspection to determine horizon depths and Munsell color.  Soil cores will also be inspected for obvious signs of disturbance such as buried horizons, lithologic discontinuities, or human artifacts.  Cores will be divided into four soil depth intervals (0 to 10 cm, 10  to 30 cm, 30 to 70 cm, and 70 to 100cm) and sorted to remove coarse roots and rocks (> 2 mm). Subsamples from each depth interval will be analyzed for soil dry weight and percent moisture, allowing for calculation of bulk density.  Total C and N will be measured by flash-combustion.  The δ13C, and δ15N of plant and soil samples will be measured with an elemental analyzer coupled to an Isotope Ratio Mass Spectrometer (Delta Plus IRMS, Thermofinnigan, San Jose, CA) at the UC Irvine IRMS facility.

Exchangeable inorganic N (NO3-, NH4+) will be extracted with 2M KCl and analyzed colorimetrically with a Flow Injection Analyzer.  Rates of potential net N mineralization, nitrification, and microbial respiration will be measured in a 10-day laboratory incubation of soils at room temperature. Soils will be placed in 946-mL glass jars with lids fitted with septa for gas sampling. After 10 days, the headspace of the jars will be sampled by syringe, and the gas samples analyzed for carbon dioxide (CO2) by thermal conductivity gas chromatography.  Inorganic N will be extracted as described above. Mineralization will be calculated as the accumulation of total inorganic N, nitrification as the accumulation of NO3-, and respiration as the accumulation of CO2 over the course of the incubation.

5. Vegetation Characterization

To test hypotheses about plant community composition and vegetation similarity across sites and regions, the occurrence of all cultivated and spontaneously occurring vascular plant species will be recorded for each parcel and natural vegetation plot. A two-person team, with at least one expert in plant identification will survey the entire yard area. Spontaneously occurring species include all species likely not to have been planted, even if they are non-native (Knapp et al. 2008). Cultivated species are those that are obviously planted. Species will be designated as exotic, native, and invasive based on local floras and state-level designations, (e.g., state Departments of Natural Resources). Taxonomic information will be standardized for synonyms using the Integrated Taxonomic Information Systems (ITIS) database automated via Nix webtools (Kembel 2007).

In addition to species richness, the species identity, size (height and diameter), and crown characteristics (height, diameter, light exposure, condition) of all trees greater than 2.54 cm diameter at 1.37m will be recorded for each parcel or plot. These data will be entered into the USDA Forest Service’s i-Tree application (formerly UFORE, http://www.itreetools.org/eco/index.php) to estimate aboveground biomass and net primary production of trees (Nowak et al. 2008).

6. Denitrification

  At each site, we will sample sediments and water chemistry of six urban lakes/ponds and six reference water bodies.  In landscapes where lakes have historically been absent, we will choose other standing waters (e.g. wetlands) or exclude reference sites.  Sediments will be stored on ice and shipped to FIU for analysis.  Rates and limitation of denitrification in aquatic sediments will be determined by direct measurement of N2 production (by a membrane inlet mass spectrometer) in response to factorial manipulation of carbon and nitrogen.  Sediments will also be analyzed for sediment size distributions and organic matter and nitrogen content using standard methods.

7. Microclimate and soil moisture monitoring

Air temperature, relative humidity, and soil moisture will be monitored in a subset (18 plots) of land use classes within each city, including native reference and the dominant and/or extremes of urban and suburban ecosystems.  Within each plot, we will install a combined relative humidity/temperature probe at 1.5 m height to approximate the microclimate experienced by urban residents.  One soil moisture probe will be placed at 10 cm depth within the dominant land cover type.  Data will be collected hourly and downloaded to dataloggers in the field.  

8. Scaling

We will combine parcel-level data from (a) the vegetation and homeowner surveys described above with (b) other relevant measures of the social landscape (e.g., Census data, PRIZM lifestyle characteristics, zoning/land-use restrictions), (c) the biophysical measurements, and (d) the high-resolution remote sensing products to produce a combined multi-scale, multi-site dataset.  This data synthesis will draw on our substantial experiences producing similar integrated georeferenced ecological-social databases in the Baltimore, Boston and Twin Cities metropolitan areas.

We will first produce regional (MSA)-scale estimates of the effects of urban land use change on soil, vegetation and whole-ecosystem pools of carbon and nitrogen (total and reactive), plant diversity (native and exotic, functional and species) and landscape denitrification potential by scaling differences in these variables between each of our experimental classes and reference sites to the MSA scale.  Actual scaling algorithms for each city will depend on which factors (if any) are strong predictors of ecological effects and scaling exercises will be accomplished using GIS coverages compiled in our database.  A lack of predictive relationships would mean that our first scaling hypothesis is rejected, i.e. that remote sensing and socio-demographic data sets can not accurately depict the ecosystem structure and function of residential parcels.  This rejection would mean that alternative approaches are necessary to understand the effects of urban land use change at regional and continental scales.

Analysis of differences in scaling algorithms between cities will be the key test of our second scaling hypothesis, i.e. that relationships between broad-scale socio-demographic and remote sensing data and actual ecosystem structure and function are robust in all cities and can be used to evaluate continental scale effects of urban land use change on ecosystem structure and function.  For example, there may be predictive relationships between biodiversity and income that are constant across cities, or alternatively, each city may have a different relationship, or some cities may have no relationship. If this hypothesis is confirmed, high resolution scaling for MSAs across the continent can be accomplished in future research.

To produce preliminary continental scale estimates for this project, we will build on the national estimation procedure put forth by Milesi et al. (2005), with modifications developed by the Clark University/Boston team for similar purposes, under support from NSF (CNH, LTER, REU Site, ULTRA-ex) to extrapolate the fine-scale (≤1.0 m) resolution land-cover mapping and plot-scale biological sampling/analysis to the continental scale. Our remote sensing team will produce spatially explicit, quantitative estimates of correlations between fine-scale (<1.0 m) measures of “lawn” in each of our six metropolitan areas and coarse-scale (30m - 1km) measures of “lawn” from extant land-cover datasets for the 100 largest U.S. metropolitan areas, which comprise the vast majority of the current U.S. population of ~300 million people. The derived correlations will be used to compare cities of (broadly) similar biophysical characteristics, for example, L.A. will be compared to San Diego, Minneapolis with Chicago, Boston with Detroit, etc. The result will be a first-order estimate of the effects of urban land use change on our response variables for the vast majority of the continental U.S. occupied land area.