3.1. Top Ten State-Level Emitters
 Figure 1 shows the spatial distribution of CO2 emissions and its sectoral breakdown at the state level. The results not only demonstrate the spatial variation of total CO2 emissions, but also the sectoral variation in each state. Electricity production in the middle of the U.S. is the largest proportion of the total CO2 emissions, while transportation dominates in the coastal states. The largest emitters overall are a combination of those with large populations and/or significant industrial activity.
 Tables 1 and 2 present the top ten fossil fuel CO2 emitting states sorted by magnitude within each economic sector in absolute and per capita units, respectively. In absolute terms, the states of Texas and California are consistently in the top positions across the economic sectors. The sector for which Texas does not occupy a position within the top five is the residential sector. This is likely due to the limited winter demand for space heating, the dominant source of onsite residential CO2 emissions in colder locales. Similarly, California does not occupy a position in the top five for the electricity production sector. This is likely due to the fact that a significant share (>50%) of California electricity consumption is generated by nonfossil fuel sources (Energy Information Administration, 2009).
Table 1. The Top Ten Absolute Vulcan 2002 Fossil Fuel CO2 Emitting States by Magnitude Within Each Economic Sectora
Table 2. The Top Ten per Capita Vulcan 2002 Fossil Fuel CO2 Emitting States by Magnitude Within Each Economic Sectora
 Examination of the top emitters on a per capita basis shows a greater mix of states occupying the top positions. States with low populations combined with more northern latitudes, energy-intensive industry or electricity production determine the top positions.
3.2. County-Level Spatial Patterns
 The spatial patterns of the absolute and per capita total CO2 emissions at the county spatial scale are shown in Figure 2. Large total (Figure 2a) absolute CO2 emissions occur in Florida, the upper Midwest population centers, the Southwest, west coast population centers, the southern Rocky Mountain region, and the “BosNYWash” corridor of the east coast. On a per capita basis, emissions have a nearly inverse relationship to the absolute spatial distribution in which larger emissions occur in the Western Plains and Rocky Mountain regions coincident with lower population density. The influence of county size is notable in Figure 2. County sizes tend to be larger in the western half of the United States and hence must be considered when interpreting results at the county scale.
Figure 2. The (left) absolute (units: MtC/year) and (right) per capita (units: tonne C/year/person) CO2 emissions at the county spatial scale from (a) all sources; (b) the residential sector; (c) the commercial sector; (d) the industrial sector; (e) the electricity production sector; and (f) the transportation sector.
Download figure to PowerPoint
 Absolute residential CO2 emissions (Figure 2b) are distinct from the total emissions pattern due to relatively greater emissions in the northeast, upper Midwest, west coast and southwest. When normalized by population, residential emissions are concentrated north of roughly 36°N latitude and east of the Rocky Mountains. These gradients reflect space heating needs driven by regions with more continental climate and, hence, longer, colder winters [Energy Information Administration, 2001]. The residential sector is strongly tied to population as absolute emissions are dominated by nonelectrical space heating in the Vulcan system [Gurney et al., 2009].
 Absolute and per capita commercial CO2 emissions (Figure 2c) show a pattern similar to that found in the residential sector but with less latitudinal dependence and a more scattered distribution of the larger per capita values. Because space heating constitutes a somewhat lower overall proportion of the total commercial fossil fuel energy use when compared to the residential sector, the spatial pattern appears less dependent upon climate conditions [United States Department of Energy, 2008].
 The per capita calculation in both the residential and commercial sectors reveals the weakness of aggregation at the county spatial scale. For example, some states show distinct boundary outlines when normalized by population (Utah, Illinois, and New York) and these are primarily due to the fact that building density varies significantly at scales below the county level. In the Vulcan data product produced at the 10 k × 10 km scale, these state outlines are eliminated due to the fact that residential and commercial emissions are distributed via census tract density of building area statistics [Gurney et al., 2009].
 Large absolute industrial CO2 emissions are distributed heterogeneously across the U.S. while centers of high per capita industrial CO2 emissions show a slight concentration in particularly intense industrial regions with somewhat lower population density such as the Gulf Coast, the oil-producing/refining regions of Texas and Oklahoma, the upper Midwest and the Front Range of the Rocky Mountains (Figure 2d). Dominated by large power facilities, absolute and per capita electricity production CO2 emissions show a similarly heterogeneous pattern across the U.S. due to the presence of large fossil fuel-based power production facilities in most regions and the presence of some of these facilities in counties with low population.
 High absolute transportation CO2 emissions (Figure 2f) show a pattern similar to that found in the residential and commercial sectors while per capita transportation CO2 emissions are clustered in the Western U.S. due to a combination of lower population density and longer average trip distance [Peng and Lu, 2007].
3.3. Probability Distributions of Per Capita Emissions
 The CPD of the sectoral per capita CO2 emissions at the county level is shown in Figure 3. The distribution of the sectoral per capita CO2 emissions can be described by three distinct groupings. In the case of the electricity production sector, the distribution shows per capita emissions which are spread over a wide distribution of values with the presence of both very small and very large per capita values. However, unlike the other economic sectors, emissions are present in a minority of the 3,141 counties in the U.S.; only 1,215 counties reported emissions from the electricity production sector. About 25 percent of those counties contain emissions less than 0.01 tonne C/person. In absolute terms, 75% of all electricity production CO2 emissions are located in 233 counties, and 95% of the emissions are achieved after including 515 counties.
Figure 3. The cumulative probability distribution (CPD) of sectoral per capita CO2 emissions at the county level. Per capita CO2 emissions are on log scale.
Download figure to PowerPoint
 Per capita transportation CO2 emissions exhibit a relatively compressed distribution with values spanning the 1 to 4 tonne C/person range. This demonstrates the relatively homogeneous need for transportation on a per capita basis. The CPD of residential, commercial, and industrial CO2 emissions can be considered a third distributional group with a range less compressed than the transportation sector but not as widely scattered as the electricity production sector. Of the three, industrial per capita CO2 emissions span both the greatest range of values with a more even distribution, from very small to slightly greater than 1 tonne C/person. The commercial and residential per capita CO2 emissions exhibit similar distributions but centered around different mean values of 0.15 and 0.4 tonne C/person, respectively.
3.4. Explanatory Variables
 In order to better understand the geographic and environmental influences on CO2 emissions, we have binned per capita CO2 emissions in each of the economic sectors according to three geographic and two climate variables: latitude, longitude, elevation, HDD, and CDD. We normalize the sector-specific, binned, per capita CO2 emissions by subtracting the mean value and dividing by the standard deviation. Figure 4 shows the distribution of these sector-specific per capita CO2 emissions as a function of the five variables.
Figure 4. Spatial distribution of normalized per capita CO2 emissions due to the total and the residential, commercial, industrial, electricity production, and transportation sectors as a function of (a) longitude; (b) latitude; (c) elevation; (d) heating degree day; and (e) cooling degree day.
Download figure to PowerPoint
 The per capita total CO2 emissions are dominated by electricity production and hence, exhibit patterns similar to the per capita electricity production CO2 emissions, as was noted in section 3.2 at the county spatial scale. The dependence of per capita electricity production CO2 emissions is complicated by the fact that the geographic pattern is not expected to follow variables that drive electricity demand because of the potentially long distance that can separate production from demand, and that about 30% of electric power was generated from nonfossil fuel energy in 2002 [Energy Information Administration, 2003].
 The longitudinal distribution of per capita electricity production emissions is larger in the middle of the country with lesser amounts toward both coasts. Per capita electricity production emissions also decrease somewhat from south to north but increase with elevation. The increase in per capita emissions across the middle and southern portion of the country is in large part due to the location of large electricity production facilities in areas with low population density. This is further evidenced by states, such as Wyoming, Montana and North Dakota which are among the highest in-state coal producers [Energy Information Administration, 2009]. This results in high ratios of electric generation to in-state retail sales, implying that these states are net exporters of electricity [Energy Information Administration, 2008]. Finally, electricity production in the Midwestern and intermountain West is predominantly fueled by coal as opposed to less carbon-intensive energy sources such as hydro, prevalent on the west coast [Energy Information Administration, 2007]. Increasing per capita emissions with elevation is due to the fact that population density declines with elevation more dramatically than does electricity production. CO2 emissions from electricity production show no consistent relationship to HDD, consistent with the fact that only about 8% of the space heating is directly driven by electricity (Energy Information Administration, Residential energy consumption survey, 2001, available at http://www.eia.doe.gov/emeu/recs/recs2001/ce_pdf/spaceheat/ce2-2c_construction2001.pdf). By contrast, there is a slight rise in per capita electricity production CO2 emissions with increasing CDD. This is consistent with the expectation that electrical air-conditioning demand is higher in locales with a greater number of hot days [Pétron et al., 2008]. Furthermore, we speculate that this may also be a collinear effect with the preponderance of coal-burning facilities in the Midwest which spans 1000 to 3000 cooling degree day range.
 The per capita residential and commercial CO2 emissions share similar patterns for each of the independent variables and can be explained primarily by climate influences. Higher elevations and northern latitudes exhibit a greater number of HDD and hence greater space heating needs. The longitudinal dependence, with slightly higher values in the East and declining toward the West, may be partially due to the colder, more continental climate going from West to East and the fact that average house/building square footage also increases from West to East (Energy Information Administration, Residential energy consumption survey, 2005, available at http://www.eia.doe.gov/emeu/recs/recs2005/c&e/summary/pdf/tableus1part1.pdf). For example, average residential floor space per household based on 2005 sampling is smaller in the Pacific and Mountain census regions (1,708 and 1,951 ft2, respectively) and larger in the East North Central and New England census regions (2,483 and 2,472 ft2, respectively). Both residential and commercial per capita CO2 emissions show a decline as the CDD increases. This is due to the fact that air conditioning needs are supplied through electricity, and hence, evident in the relationship between electricity production and CDD.
 The per capita industrial CO2 emissions are larger in the interior versus coastal areas in the longitudinal direction while exhibiting a minimum in the 37°N to 41°N latitudinal bin. The larger values in the southern latitudes are due to the high-emitting oil production and refining of the Gulf coast region which are less labor-intensive as evidenced by their higher ratio of receipts per paid employee (http://www.census.gov) than other industrial sectors. The per capita industrial CO2 emissions have little dependence upon elevation with a shift in values at approximately 600 m, above which are mainly mountain states. The per capita industrial CO2 emissions exhibit a complicated relationship to HDD and CDD with a minimum in the center of the HDD and CDD numerical spans and this may be collinear with the underlying geographic distribution. Furthermore, the use of a per capita normalization in the industrial sector (like the electricity sector to a somewhat lesser degree) is complicated by a number of factors. The amount of labor required to support industrial activities varies and that variation depends upon broad industrial classifications which, in turn, have geographic relationships. For example, the top coal producers, West Virginia, Kentucky and Wyoming (Energy Information Administration, 2009), are among the states with the highest GDP in the mining industrial category (http://www.bea.gov/regional/gsp).
 The longitudinal dependence of per capita transportation emissions exhibit a maximum in the continental interior corresponding to the ridge of large values running west to east along the Mountain and intermountain West. This is driven, in large measure, by the presence of large coast population centers with high population density and small trip distance values [Puentes and Tomer, 2008]. By contrast, the latitudinal distribution of per capita transportation emissions has a minimum value in the middle of the country. The relationship with elevation correlates with the region of sparse population and high trip distance and further corresponds to lower road densities noted by the National Highway Planning Network data (http://www.fhwa.dot.gov/planning/nhpn/).
 Increases in per capita transportation CO2 emissions with elevation are due to greater trip distances in predominantly rural, high-elevation locales. The relationship between per capita transportation emissions and HDD exhibits maxima at values ranging from 2000 to 4000 and at values greater than 7000. The relationship is likely collinear with geography, particularly the increasing per capita transportation emissions at the higher HDD values, which corresponds to the rural, high trip distance mountain and intermountain West. Similarly, the relationship between per capita transportation emissions and CDD exhibits some collinearity with geography (the lowest CDD values correspond to the cold mountain/inter mountain west) though research supports lessening vehicle efficiency at higher temperatures due to increased air conditioner use. Studies indicate that running the air conditioning in a passenger car reduces fuel efficiency by approximately 12% at highway speeds [Parker, 2005; Climate Change Science Program, 2007].
 The relationship between the sectoral per capita emissions and the CDD/HDD values has implications for how energy demand and emissions will respond to climate change. Though spatial gradients are not a perfect substitute for temporal behavior, the spatial relationships are informative. For the HDD metric, both the residential and commercial per capita emissions show a reasonably linear response. In the residential and commercial sectors, binned HDD values explain 88% and 86% of the variation in per capita carbon emissions. Furthermore, the relationship suggests a decline of 0.07 and 0.03 kg C/person per unit of HDD decline in the residential and commercial sectors, respectively, a reflection of the lessened need for space heating as HDD values decline over space.
 For the CDD metric, the relationship is most pronounced for emissions in the electricity production sector (explained variance of 68%). This relationship suggests that an increase in one unit of CDD would be accompanied by 0.57 kg of carbon per person. This exceeds the incremental residential and commercial space heating emissions decline due to warmer temperatures by over a factor of five. Some of this is explained by the carbon intensiveness of electricity production versus space heating. However, even if one assumed that all electricity production was based on coal and all space heating was based solely on natural gas, the ratio of carbon intensity would suggest a factor of two. Hence, it would appear that per capita electricity production CO2 emissions are far more sensitive to external temperature than residential and commercial per capita CO2 emissions. This stands in stark contrast to studies that have suggested future warming would be accompanied by savings in space heating needs that nearly offset the requirements of increased cooling [Hadley et al., 2006].
3.5. Spatial Clustering
 To objectively explore the spatial patterns of fossil fuel CO2 emissions produced by the Vulcan inventory, a null hypothesis, implying random spatial distribution, was tested via spatial autocorrelation using a Global Moran's I. The results for absolute and per capita CO2 emissions in each sector are summarized in Table 3. The statistical significance of the spatial clustering was computed using a permutation approach with 9999 permutations [Anselin et al., 2004]. The results indicate statistically significant positive spatial autocorrelation for the absolute and per capita CO2 emissions in all sectors except for the per capita emissions in the commercial sector. A large positive value indicates that similarly valued emissions are highly clustered in space.
Table 3. The Global Moran's I for Absolute and per Capita Fossil Fuel CO2 Emissions in Each Economic Sector and Total Sourcea
|CO2 Emissions||Moran's I (Absolute)||Moran's I (per Capita)|
|Total||0.23 (0.001)||0.13 (0.001)|
|Residential||0.43 (0.001)||0.82 (0.001)|
|Commercial||0.25 (0.001)||0.00 (0.1)|
|Industrial||0.10 (0.001)||0.03 (0.02)|
|Electricity Prod||0.08 (0.001)||0.13 (0.001)|
|Transportation||0.34 (0.001)||0.15 (0.001)|
 The positive autocorrelation of per capita CO2 emissions in each sector is lower than the absolute value except for the residential and electricity production sectors. This is due to the clustering effect of population and the associated emissions in large population centers. For example, the global Moran's I coefficient for transportation CO2 emissions decreases from 0.34 to 0.15 when normalized by population.
 The largest spatial autocorrelation value is present in the residential sector with a global Moran's I coefficient of 0.43 and 0.82 for the absolute and per capita values, respectively. This is consistent with the evidence that residential emissions are dominated by space heating and space heating is driven by local climate, itself a positively autocorrelated variable [Tan et al., 2005]. Normalization by population heightens this effect by focusing on colder areas with lower population density. The result is high clustering in the upper Midwest, New England, and the Rocky Mountains. The low Moran's I coefficients for the industrial sector and electricity production indicates a more random distribution of emissions. This is not entirely surprising as the emissions in these two sectors are dominated by point sources which are often isolated and in low population density locales [Gurney et al., 2009].
 Figure 5 presents the sector-specific LISA values (denoted as “high-high” and “low-low”) for the absolute and per capita CO2 emissions.
Figure 5. LISA cluster maps for the (left) absolute and (right) per capita CO2 emissions at the county spatial scale from (a) total sources; (b) the residential sector; (c) the commercial sector; (d) the industrial sector; (e) the electricity production sector; and (f) the transportation sector.
Download figure to PowerPoint
 The clusters of low-low absolute CO2 emissions show similar patterns across all the sectors except for electricity production in which the low-low clustering throughout the western Plains region is absent. This owes in part to the presence of large electricity production facilities in relatively remote portions of the West and Southwest United States [Gurney et al., 2009]. The extent of these low-low clusters varies in each economic sector, however. The low-low absolute CO2 emissions in transportation sector have the greatest spatial extent compared to other sectors. The clusters of low-low absolute industrial CO2 emissions are distributed somewhat more heterogeneously than other economic sectors. The high-high clusters of absolute CO2 emissions are less extensive than the low-low clusters across all sectors. Aside from electricity production and the industrial sector, the clusters of high-high absolute CO2 emissions are mainly distributed throughout the high population urban corridors. As with the low-low clustering, the high-high clustering for electricity production and the industrial emissions are scattered and limited in spatial extent. This, once again, highlights the disaggregated nature of these point source facilities.
 Normalization by population causes a shift in which the low-low cluster moves from predominantly inland locations to more coastal regions across all sectors. The exception to this pattern is electricity production which shows only a minor change when normalized by population. The spatial extent of these clusters tends to decrease except for the residential sector. The residential per capita CO2 emissions, by contrast, show large low-low clusters throughout the coastal U.S., occurring along the west coast, the Southwest, the Gulf coast, and coastal Southeast. The areas of low-low per capita commercial CO2 emissions are next in magnitude, and share much of the pattern of the residential per capita emissions except that the western and southwest maxima do not occur. These low-low clusters are coincident with milder marine-influence climates, requiring less extreme wintertime interior heating.
 The high-high clusters of per capita CO2 emissions are most pronounced for the residential and transportation sectors where they tend to occupy inland areas, especially in the case of the transportation sector. The high-high clusters of per capita CO2 emissions in the residential sector are the largest and most spatially coherent, and they mainly occur in New England, the Middle West, Utah, and Kansas. The low-low and high-high spatial clustering of per capita electricity CO2 emissions is small compared to other sectors. Owing to the fact that they dominate the total emissions, this pattern tends to drive the spatial clustering in Figure 5a.