Data

The purpose of this work is to describe determinants and spatial patterns of atmospheric carbon dioxide (CO2) in Phoenix, Arizona. Specifically, we use geographic information systems (GIS) and regression-based analyses to identify the human and biological factors that contribute to spatial and temporal variations in near-surface atmospheric CO2 levels. We use these factors to create estimated surfaces of CO2 for the urban area. We validate our surfaces using independently collected records of CO2 from several monitoring stations and transects. To investigate the temporal patterns and variations of CO2, we were able to generate CO2 surfaces for the early mornings and the afternoons, and on weekdays when traffic is heavy and spatially focused and on weekends when it is lighter and more spatially dispersed. Our findings suggest there is a distinct relationship between the structure of Phoenix CO2 levels and spatial patterns of human activities and vegetation densities. Morning CO2 levels are higher than afternoon levels and correspond closely to the density of traffic, population, and employment. The spatial structure of human activity explains the pattern of CO2 better on weekdays than on weekends. CO2 surfaces reflect declining densities of human activity with distance from the city center, the pattern of irrigated agriculture in the Phoenix area, and riparian habitats on the urban fringe. Spatial and temporal patterns of CO2 are useful in understanding urban climate and ecosystem processes

Transect data collection

As part of the larger NSF project to study the characteristics of the CO2 dome in Phoenix, atmospheric concentrations of CO2 were measured with an InFrared gas analyzer (model LI-800; Li-Cor, Lincoln NE) along four transects on major roadways through metropolitan Phoenix on 10 14 consecutive days from 3 to 16 January, 2000. Transects 1 and 2 were chosen to bisect the urban area, north to south and east to west respectively. Transects 3 and 4 weave in and out of the urbanized area to capture CO2 processesat the urban/rural boundaries. Measurements were taken simultaneously along all four transects at 2 meters above the ground at points approximately one mile apart between the pre-dawn hours of 5:00 and 6:00 and again during the afternoon hours of 14:00 to 15:00. Idso et al. (in press) provides a more complete discussion of data collection methods. Inconsistent and erratic readings along Transect 4 led us to drop those observations from our analysis; thus we were left with morning and afternoon CO2 readings from 227 monitoring sites spread non-randomly, but covering a broad cross-section of environments, across the metropolitan area. For each site, we calculated morning and afternoon means across the 14 days (n = 14), and then on weekdays (n = 10) and weekends (n = 4) separately. Ideally, our monitoring sites would be randomly distributed across the metropolitan area. This, however, was operationally impossible given that CO2 needed to be collected at roughly the same time across all 227 sites. All sites along a given transect could be taken within an hour, and each transect was monitored simultaneously. Therefore, we assume there was a roadway effect on all transects (Idso 1998b), which we tested during our validation experiment. We separated weekday and weekend observations because intra-urban vehicular traffic is a major source of CO2 emissions and because the intensity and spatial organization of weekend and weekday traffic differs. The sheer amount of traffic along Phoenix-area thoroughfares is more than 40 percent higher on weekdays than on weekends, with Friday as the peak and Sunday as the low point. Moreover, because weekday travel is more work-oriented than weekend travel, it is more geographically concentrated in the early morning and late afternoon hours (Barber 1986). Weekend travel is less concentrated in both time and space. In a related paper characterizing urban CO2 in Phoenix, Idso et al. (in press) found that weekday CO2 levels exceed weekend levels by 30 percent in the commercial core of Phoenix but there are no weekday-weekend differences in surrounding residential areas. Independent variables for this study are population density (POP), employment density (EMP), average weekly traffic (AWT), and vegetation density (VEG). Population data were extracted from ESRI’s 1990 compiled Census data, which are based on the 1990 Census of Population (ESRI 1997). From this CD, we extracted total population for 1997 for all Census tracts in Maricopa County, Arizona. Data were converted from an ArcView© shapefile into an Arc/INFO© coverage. Employment and traffic data were obtained from the Maricopa Association of Governments (MAG). Employment data were converted from tabular form to an Arc/INFO point coverage with the total number of employees associated with each point. Average weekly traffic data were converted from MAG’s 1998 average weekly traffic map into an Arc/INFO coverage. Vegetation data were provided by the Central Arizona Phoenix - Long Term Ecological Research (CAP - LTER) project as normalized difference vegetation index (NDVI). NDVI is an indicator, with a range from 0 to 1, representing vegetation density (green leaf area, biomass, percent green cover, productivity, and photosynthetic activity). The Soil Adjusted Vegetation Index (SAVI) is often used as a measure of vegetation in a desert/urban environment to eliminate soil-induced variations in the index (Heute 1988). Since we wanted to account for differences in soil, we elected to use NDVI. Preliminary results of our analysis showed there were only minor differences between using NDVI and SAVI in our model. The NDVI were calculated in ERDAS© from a May 1998 Thematic Mapper image (cell resolution is approximately 30 meters). The ERDAS image was geographically referenced and converted into an Arc/INFO GRID format. The month of May was selected because the general consensus among the CAP - LTER scientists was that May coincided with maximum leafout and vegetation cover. May is also one of the best times of year for collection of remotely sensed data in this region because cloud cover tends to be minimal (the winter storms are gone, and the monsoon has not yet started). We do not view the differences between May and January as significant, because there is a strong spatial correlation in vegetation patterns from month to month.