Vegetation dynamics and human settlement across the conterminous United States

Abstract

Demography and ecology have long been intertwined in terms of understanding the relationships between population and the environment. Recent advances in data and technology, coupled with our increased understanding of social and ecological process, have greatly expanded the ability to link populations and ecosystems in order to understand their interrelationships. However, there remains a paucity of understanding of how climatic variability relates to the spatial patterning of people and how they may influence one another. Here we couple MODIS satellite estimates of interannual photosynthetic variability from 2000–2011 with housing density for the year 2000 to provide an estimate of the interaction between productivity dynamics and exurban influence at a 2 km resolution for the conterminous United States. The resultant map shows the convergence of population and climate influences on vegetation responses with broad patterns of interaction across the United States and notable extremes found throughout the Central Plains and localized regions of the Southwest US. These intersections of land use and vegetation dynamics have significant implications for ecological systems and ecosystem responses to climate dynamics.

Keywords:

• housing development
• MODIS
• settlement
• vegetation dynamics

1. Introduction

Understanding the linkages between population and the environment has been a focus of both demographers and natural scientists since at least the late 1700s, when Thomas Malthus published his seminal work An Essay on the Principle of Population (Malthus, 1798). In the intervening centuries, there has been marked advancement in understanding how demography shapes the environment and reciprocally how the environment shapes demography (e.g. An et al., 2011; Ehlrich & Holdren, 1971; Lepczyk, Hammer, Stewart, & Radeloff, 2007). One recent focus of integration has been on using housing units as a proxy for population in order to integrate with environmental patterns and processes. In particular, houses and housing units (hereafter ‘houses’) offer a different and perhaps more meaningful way to ascertain and investigate relationships between demography and the environment (e.g. Hammer, Stewart, Winkler, Radeloff, & Voss, 2004; Lepczyk, Linderman, Hammer, & Kulcsár, 2012; Liu, Daily, Ehrlich, & Luck, 2003; Radeloff et al., 2005).

The utility of houses is that they provide a key integrator of sociology, demography, and environmental impact. That is, houses contain residents, which make choices jointly that directly or indirectly effect the surrounding environment. In addition, the construction of houses directly transforms the biophysical characteristics of the surface of the earth, the land cover, as well as the nature of the activities, or land use, of individual parcels. As a result, the onset of housing development initially transforms the land followed by the residents who live in them continuing to have an influence on the land surrounding the house. Because these influences are governed in part by the social structure of the household, the nature of the occupants also influences the environment. Hence, aggregating houses across the landscape can provide unique insight into the intensity and nature of human modification of ecosystem processes and function (Carr, 2004).

One area of continuing research is focused on understanding the mutual interactions between human modification of ecosystems and expected vegetation response to climate dynamics. To begin addressing these interactions we coupled measures of the interannual variability of photosynthetic capacity as measured by the Normalized Difference Vegetation Index (NDVI) with the spatial distribution of housing density to provide a first analysis of the potentially complex spatial interactions of housing density and ecosystem responses. We demonstrate these relationships across the conterminous United States during the period 2000 to 2011. The resultant map provides a visualization and quantification of the interaction and potential feedbacks between climate dynamics, annual vegetation responses, and human modifications. The index provides an estimate of not only the mediation of climate drivers of vegetation dynamics, but also the underlying spatial demographics related to these changes.

2. Methods

Vegetation dynamics and productivity responses to climate trends and annual dynamics is of continuing research interest given productivity's role in carbon dynamics and the implications for society's reliance on these systems for food and fiber (Bradford, Lauenroth, Burke, & Paruelo, 2006; Linderman, Zeng, & Rowhani, 2010; Rowhani, Lobell, Linderman, & Ramankutty, 2011). In order to understand climatic and vegetation dynamics, we used Moderate Resolution Imaging Spectroradiometer (MODIS) satellite measures of NDVI from eight-day composite MOD43 data for 2000–2011. NDVI has been empirically and theoretically related to the fraction of absorbed Photosynthetically Active Radiation (faPAR) specific to individual plant functional groups (Myneni, Hall, Sellers, & Marshak, 1995; Yang et al., 2006). In turn, Gross Primary Productivity (GPP) in unstressed vegetation is linearly related to faPAR relative to its light use efficiency (Medlyn, 1998; Running et al., 2004). NDVI is calculated based on the MODIS MOD43 CMG 0.05 deg combined Terra and Aqua data. MOD43 atmospherically corrected reflectance estimates are further corrected for Bi-Directional Reflectance Function (BRDF) effects through inversion of 16 days of MODIS sensors aboard the Terra and Aqua platforms (Schaaf et al., 2002). The 0.05-degree resolution nadir corrected data, therefore, provide NDVI estimates for the globe at eight-day overlapping intervals. NDVI is calculated from the normalized difference of red and near infrared reflectance measurements based on the following equation:

where ρ _nir and ρ _red are the reflectances corresponding the near infrared (841–876 nm) and red (620–670 nm) portion of the electromagnetic spectrum, respectively (Tucker, 1979). Missing data due to extensive clouds, lack of accurate inversion, or sensor anomalies are filled with corresponding annual reference profile NDVI values. Reference profiles were calculated as the median value of all eight-day values for each of the individual 46 periods corresponding to a year across all 11 years. Where data were missing for all 11 years, data intervals were interpolated based on a spline interpolation across the full time series and substituted for missing data. Annual integrated NDVI values (iNDVI) were calculated based on a simple sum of all 46 measures per year. The Coefficient of Variation (CoV), σ/μ, was calculated on the 11 annual iNDVI values. CoV ranged from approximately −90 to 1000, while most ranged from approximately 0 to 1.5. Anomalously high and low values were typically related to low reflectance, highly variable systems such as edges of water bodies.

CoV data were reprojected in datum NAD83 and spheroid GRS 1980 to the Albers Equal Area Conic projection at a 5 km resolution (approximate CMG resolution) and combined with housing density data. The interaction between CoV(iNDVI) and housing density was used to examine the influence of urban development and exurban expansion on vegetation dynamics and the relative relationship between integrated climate factors as expressed through vegetation variability. The housing data are a spatiotemporally consistent vector coverage of housing units at the partial block group level for the conterminous United States from 1940 to 2000 as developed by Hammer et al. (2004) from US decennial census data. Partial block groups fall between blocks and block groups in the hierarchy of US Census Bureau geographies (see http://www.census.gov/geo/www/reference.html), and are roughly equivalent, in social terms, to subdivision-sized neighborhoods. We used the 2000 housing density data and rasterized it to a 2 km resolution. A CoV-housing density index was developed through the product of the CoV and ln(housing density +1) to provide an estimate of demographic related land use impacts on productivity dynamics. Values close to zero would represent either stable vegetation dynamics and/or low housing density, while high values (i.e. close to 0.65) represent those areas with high interannual vegetation dynamics coincident with high housing density.

3. Conclusions

We coupled measures of the interannual variability of vegetation with the spatial distribution of housing density to provide a first analysis of the potentially complex spatial interactions of housing density and ecosystem responses. Interannual variability of photosynthetic capacity, as measured by annual CoV(iNDVI), is expected to be the interaction between precipitation variability and meristem density. High productivity potential herbaceous biomes are able to respond most significantly to moderate changes in precipitation while the lowest relative variability is expected in forested ecosystems (Knapp & Smith, 2001). In turn, housing density through much of the Eastern US is characterized by large urban areas with moderate to high housing density existing throughout the region. Increasing abandonment of agriculture and associated vegetation response within mixed ex-urban housing development may lend itself to significant high interaction between human dominated areas and vegetation dynamics (Brown, Johnson, Loveland, & Theobald, 2005). Finally, high housing density within the Central Plains and Southwest US and corresponding climate and relative vegetation variability was expected to result in high localized interactions.

Regional trends in CoV of annual iNDVI associated with semi-arid systems show the highest levels of interannual variability where large relative annual differences in precipitation and relative herbaceous response capacity result in significant variability in annual iNDVI throughout the central plains and Southwest US Managed ecosystems (heavily agricultural Midwest) and forested biomes in the Eastern US resulted in relatively low CoV measures (i.e. little variation in iNDVI from year to year), with notable regional patterns of higher variability within the Ohio valley and metropolitan areas such as Atlanta, GA.

Concentrated pockets of extremely high housing density found in Southern California and Arizona and their interaction with high CoV resulted in the highest index values throughout the lower 48 states. Broad regional trends throughout Texas and Oklahoma resulted in localized and broadly distributed interactions between productivity dynamics and housing density yielding extensive distributions of high index values. Similar interactions are found in the Colorado Front Range urban corridor. Throughout the Eastern US, interactions between CoV and housing density were generally lower and largely coincided with regional patterns of higher index values throughout the Ohio Valley extending throughout northern Kentucky, Indiana, and Ohio. Relatively high values of CoV(iNDVI) in the Southeast are limited to pockets of protected areas with complex hydrology and coastal areas. The most notable exception is the Atlanta metropolitan area where localized variability occurred. Modification of ecological composition or decreases in annual productivity may result in more variable response relative to exurban expansion. Given the relatively localized patterns in extreme values, housing density, through historical drivers of population distributions and land use, is largely concentrated in relatively stable ecosystems. Regions of significant interactions, however, point to high potential modification of sensitive ecosystems and the soil, vegetation, and atmosphere interface and the associated surface energy balance, trace gas and hydrological processes.

Software

MODIS data were extracted, stacked and analyzed in ENVI/IDL 4.8/8.0. Additional modeling and index creation were conducted in Erdas Imagine 2011. Maps were produced in ESRI ArcMap v.10.0 software.

Acknowledgements

We would like to thank Roger Hammer for his development of the housing database and three anonymous reviewers who provided feedback on the draft manuscript.