View
1
Download
0
Category
Preview:
Citation preview
Sensitivity of residential water consumption to variations in climate: An intra-urban analysis of
Phoenix, Arizona. An edited version of this paper was published by AGU. Copyright 2007 American Geophysical Union.
Balling Jr., R. C., Gober, P., and N. Jones. (2008), Sensitivity of residential water consumption
to variations in climate: An intra-urban analysis of Phoenix, Arizona, Water Resources Research,
44, W10401, DOI:10.1029/2007WR006722. To view the published open abstract, go to
http://dx.doi.org and enter the DOI.
1
Sensitivity of residential water consumption to variations in climate: An intra-urban analysis of Phoenix, Arizona
R.C. Balling, Jr., P. Gober, and N. Jones
School of Geographical Sciences, Arizona State University, Tempe, Arizona, USA
[1] Water remains an essential ingredient for the rapid population growth taking place in
metropolitan Phoenix. Depending upon the municipality, between 60 and 75% of
residential water is used outdoors to maintain non-native, water-intensive landscapes and
swimming pools [Mayer and DeOreo, 1999]. Residential water use in Phoenix should be
especially sensitive to meteorological and climatic variations because of the strong
emphasis on outdoor water use. This study explores the intra-urban spatial variations in
the sensitivity of residential water consumption to atmospheric conditions. For 230
census tracts in the city, we developed times series of monthly water-use anomalies and
compared them to monthly anomalies of temperature, precipitation, and the Palmer
Drought Hydrological Index. We found that one third of census tracts have little-to-no
sensitivity to climate, while one tract had over 70% of its monthly variance in water use
explained by atmospheric conditions. Greater sensitivity to atmospheric conditions
occurred in census tracts with large lots, many pools, a high proportion of irrigated mesic
landscaping, and a high proportion of high-income residents. Low climatic sensitivity
occurred in neighborhoods with large families and many Hispanics. Results suggest that
more affluent, non-Hispanic neighborhoods will be disproportionately affected by
increasing temperatures due to urban-heat-island effects and the buildup of greenhouse
gases.
2
Citation: Balling, R.C., Jr., P. Gober, and N. Jones (2008). Sensitivity of residential
water consumption to variations in climate: An intra-urban analysis of Phoenix, Arizona,
Water Resour. Res.,
1. Introduction
[2] Water is the key resource for growth in a desert city such as Phoenix. The
ability to capture and use large quantities of water from faraway watersheds was the basis
for prehistoric settlement until 1450 A.D., large-scale agricultural development during
the 19th and early 20th centuries, and modern urbanization [Gammage, 1999; Gober,
2006]. The availability of large supplies of surface water from upstream watersheds of
the Salt, Verde, and Colorado Rivers, augmented by groundwater from large sedimentary
aquifers, fostered the development of an oasis culture. This culture initially was based on
irrigated agriculture and later on urban development with heavy water use for urban
lakes, golf courses, outdoor landscaping featuring turf grass, palm trees and other humid-
region vegetation, and backyard pools.
Phoenicians use far more water than the typical urban resident, especially for
outdoor purposes; the average residential household uses approximately 650,000 liters
per year. Approximately three-quarters of that water is used outside for swimming pools
and landscape irrigation [Mayer and DeOreo, 1999; Phoenix Water Resources Plan,
2005]. A more judicious use of outdoor water will be needed to accommodate the
projected urban growth from today’s 4 million residents to 8 million in 2040 [Arizona
Department of Economic Security, 2007], and as a hedge against a potentially warmer,
drier climate future. In a recent study based upon 19 climate models from the Fourth
3
Assessment of the Intergovernmental Panel on Climate Change, Seager et al. [2007]
found a broad consensus among models that the western United States will dry
significantly during the 21st century, and indeed that the transition to a more arid climate
is already underway. Local water planning is shifting away from short-term fixes
designed to deal with drought conditions and toward long-term adaptation strategies that
respond to a range of uncertain conditions, including climate change, environmental
regulations, water quality concerns, and increasing competition for supplies.
Conservation is now focused on programs and levels of efficiency that can become part
of residents’ lifestyles. Structural efficiencies associated with new homes and higher
density urban development reduce the impact of immediate shortages, but also bring
long-term benefits by reducing the costs of infrastructure and supply augmentation
[Phoenix Water Resources Plan, 2005].
[3] Understanding the sensitivity of local water use to variations in climate is a
first step toward designing water conservation policies focused on reducing outdoor
water use and targeting neighborhoods where outdoor use is high. In a previous study,
Balling and Gober [2007] collected annual residential water use data from the City of
Phoenix from 1980 to 2004, and, using a series of multivariate statistical procedures, they
showed that annual water consumption is significantly related to the overall state of
drought, autumn temperatures, and summer monsoonal precipitation levels. While these
climate variables had a statistically significant impact on water consumption, the
relationships were weaker than expected given Phoenix’s hot, desert climate and
substantial outdoor water use. They concluded that outdoor water use is strongly
influenced by structural conditions, such as durable irrigation systems that deliver water
4
irrespective of temperature and precipitation conditions, and by urban lifestyles that are
heavily oriented toward single-family homes, heavily-watered landscape treatments
(including popular citrus trees), backyard swimming pools, and water features such as
waterfalls and fountains. Prevailing water landscape practices (e.g., failing to reset
automated timers in response to changes in the atmosphere’s demand for water), low
water prices, and water-dependent lifestyles probably account for the relative
insensitivity of urban water use to variations in climate conditions.
[4] Phoenix is not, however, monolithic. In a study of residential water use at the
census tract level, Wentz and Gober [2007] found that intra-urban differences in water
use across the city were substantial and explained by average household size, a factor
important in indoor water use, and by the presence of swimming pools, average lot size,
and mesic landscaping—factors related to outdoor use. Levels of consumption increased
with larger households, more swimming pools, larger lots, and more of the average lot
covered with mesic landscaping. They also showed, using geographically-weighted
regression [Fotheringham, et al., 2002], that there were spatial effects associated with
household size and the presence of pools. In other words, adding an additional resident or
pool increased water use more in some parts of the city than in others. This finding is
consistent with our thesis that some neighborhoods are more climate sensitive than
others.
Further evidence of a geographic pattern in climate sensitivity is the ratio of
summer versus winter water use (Figure 1). Assuming that indoor water use is constant
throughout the year, outdoor water use tends to increase in the summer months. Across
the city as a whole, single-family usage averages about twice the levels in the high-use
5
summer months of June, July, August, and September than in the low-use months of
December, January, February, and March. The difference between summer and winter
usage does not capture all outdoor use, because there is a substantial amount of outdoor
use in winter months to maintain trees, shrubs, and winter lawns. Nonetheless,
neighborhoods where summer use is dramatically higher than winter use are concentrated
near the city center and the inner core of the metropolitan area, while low ratios (summer
water use is about equal to winter) along the city’s northern and southern peripheries.
Areas where summer use greatly exceeds winter use are not necessarily the areas with the
highest overall use. Rather, they are older neighbohoods with dense mesic vegetation
and are possibly influenced by urban-heat-island effects that heighten the need for
summer watering [Guhathakurta and Gober, 2007].
[5] Previous efforts to relate water use to climate conditions have been
concentrated at the citywide scale and focused on variations in time, not space. Based on
work throughout the southwestern United States, some studies have found significant
relationships between temporal variations in water consumption and variations in
climate [e.g., Billings and Agthe, 1980, 1998; Maidment and Parzen, 1984; Woodard
and Horn, 1988; Billings and Day, 1989; Wilson, 1989; Martin and Kulakowski, 1991;
Rhoades and Walski, 1991; Agthe and Billings, 1997; Gutzler and Nims, 2005], while
others found no link whatsoever [e.g., Berry and Bonem, 1974; Cochran and Cotton,
1985; Gegax et al., 1998; Michelsen et al., 1999]. Gutzler and Nims [2005, p. 1778]
concluded that studies “ in the southwestern United States have reached surprisingly
diverse and apparently contradictory conclusions about the impact of climatic variability
on water demand.” This may result from differences in variables and methods used in
6
the analyses, but also from differences in the price of water, urban lifestyles, and the
relative importance of outdoor versus indoor water use.
[6] Parts of Phoenix are characterized by low-income housing with little mesic
landscaping and few if any outdoor swimming pools. Water consumption there is
relatively low and largely for indoor purposes; theoretically these areas should be less
sensitive to prevailing climate conditions than other parts of the city where high-income
residents enjoy lush landscaping, large lots, private swimming pools, fountains, and other
water features. Water is required to keep lawns alive, bushes and trees green, and pools
filled. Many of the region’s estimated 300,000 swimming pools have automatic refilling
devices, and thus, many homeowners are unaware of the high rates of summertime
evaporation (Figure 2). A typical uncovered swimming pool with around 60,000 liters
loses approximately two meters of water a year to evaporation. We anticipate that
neighborhoods with private pools, large lots, and heavy use of mesic landscaping will be
more climate sensitive than those without grass and pools and where homes are built at
higher densities.
2. Water Use Data
[7] We obtained 1995 to 2004 residential water records from the City of
Phoenix’s Water Services Department (Figure 3). Records are organized by census tracts
so as to protect the identity of individual users. Although these data are based on
imperfect billing records, contain substantial variations due to leakages and meter
problems, and require aggregate rather than individual analysis of water consumption
trends, they provide information about overall spatial and temporal trends in water use.
7
Intra-urban variations at the census tract level have been used to make inferences about
the determinants of water consumption [Wentz and Gober, 2007], the effects of the urban
heat island on residential water consumption [Guhathakurta and Gober, 2007], and the
effects of conservation policies on water demand [Campbell et al., 2004].
[8] From the original data set with 303 census tracts, we eliminated cases in
which there were fewer than 50 records upon which to base the monthly water
consumption value for a particular census tract. This left 282 tracts for further
investigation based on 2,783,405 individual single-family monthly records over the 1995
to 2004 period. We plotted and visually inspected the time series of average monthly
water consumption values from 1995 to 2004 for all tracts. We identified anomalous
values in calendar year 2004 for dozens of tracts and therefore eliminated 2004 from
further analysis. We also identified other time series that contained what were clearly
erroneous water consumption values. Rather than simply eliminate tracts with the
obvious spurious values, we introduced a more rigorous statistical evaluation of each
time series.
[9] We converted all of the monthly water consumption values for each tract into
standardized z-scores based on the 1995-2003 mean and standard deviation. If the
absolute value of any z-score was greater than 3.0 (which should happen by chance only
once in approximately 370 cases in a normal distribution), the tract was eliminated from
our investigation. This left 230 tracts, and visual inspection of their monthly water
consumption data showed no spurious values. If we chose a z-score cut-off of 4.0
(expected only once in approximately 16,666 cases), we would have been left with 274
tracts. The use of the 3.0 cut-off undoubtedly eliminated some tracts in which all data
8
were valid, but the 3.0 cut-off left us with a large spatial sample of 230 tracts with what
appeared to be all valid monthly water consumption values.
[10] Annual water consumption for single-family residential units averaged
652,788 liters but ranged from 409,461 liters to 2,491,429 liters, indicating large spatial
variability in single-family water consumption (Figure 4). The distribution was severely
right skewed (the mean of 652,788 liters exceeds the median of 599,411 liters) with a
standardized coefficient of skewness of 28.06 (p<0.01). The map shows a patchwork
pattern of consumption with the highest values in the wealthier northeastern sections and
several high-income, gentrified, inner-city neighborhoods.
[11] Finally, we converted all monthly water consumption values in each tract to
percentages of normal based on monthly means for individual tracts determined for the
1995-2003 time period. The conversion to monthly percentages effectively eliminated the
large annual cycle (Figure 5) that would otherwise dominate temporal variance in the
water consumption values. This left us with a matrix of 108 rows, one for each month
from January, 1995 to December, 2003 and 230 columns, one for each census tract. Each
cell in the matrix contained the percent of normal monthly water consumption for a given
census tract.
3. Climate Data
[12] To represent monthly variations in climate, we selected temperature,
precipitation, and drought data for the Phoenix area from the updated United States
Historical Climatology Network [Karl et al., 1990]. The USHCN data are calculated
from many weather stations within relatively homogeneous climate divisions. The
9
records are adjusted by others for time-of-observation biasing [Karl et al., 1986],
instrument adjustments [Karl and Williams, 1987; Quayle et al., 1991], and missing data
from stations within a division. We assembled the monthly records from 1995 to 2003 for
the “South Central” division that covers 12.8% of Arizona, including all of the Phoenix
metropolitan area. These are the same climate data used by Balling and Gober [2007] in
their study of temporal variations in Phoenix-wide water consumption levels. While the
use of the USHCN data may mask spatial variability in climate related to the extensive
Phoenix urban heat island, the USHCN data does effectively capture the temporal
variations in climate of interest in this investigation.
[13] The monthly mean temperature record shows a range from 33.81°C in July
of 2003 to 9.72°C in December of 1997. We found an upward linear trend in the monthly
data of 0.09°C year-1, although the trend is not statistically significant (p=0.76) over the
relatively short 1995 through 2003 study period. To eliminate the annual cycle in the
monthly temperature records, and to make the data comparable to the transformed
monthly water consumption time series, the temperature records were converted into
anomalies (°C) from the mean monthly temperatures determined for the 1995 to 2003
period. The anomalies ranged from +3.03°C in October, 2003 to -4.44°C in November,
2000.
[14] The monthly precipitation data also showed no significant trend (-0.57 mm
year-1; p=0.48) and the monthly values were converted to anomalies (in mm) based on the
mean monthly precipitation totals from 1995 to 2003. The monthly precipitation
anomalies ranged from +85.88 mm in February, 1998 to -35.28 mm in February, 2002.
10
[15] We selected the Palmer Hydrological Drought Index (PHDI) to represent
monthly drought conditions in the Phoenix metropolitan area. Palmer [1965] developed
the PHDI, along with other drought measures, and these indices have been used in many
research studies as well as in operational drought monitoring during the past 40 years [see
Newman and Oliver, 2005]. The PHDI accounts not only for precipitation totals, but also
for temperature, evapotranspiration, soil runoff, and soil recharge. Values near zero
indicate normal conditions for a region, values less than –2 indicate moderate drought,
values less than –3 indicate severe drought, and values less than –4 indicate extreme
drought. Values greater than +2 indicate moderately wet conditions, those above +3
represent very wet conditions, and PHDI values above +4 are for extremely wet
conditions. Alley [1984] identified three positive characteristics of the index that
contribute to its popularity: (a) it provides decision makers with a measurement of the
abnormality of recent weather for a region; (b) it provides an opportunity to place current
conditions in an historical perspective; and (c) it provides spatial and temporal
representations of historical droughts. There are certainly limitations when using the
PHDI (or any other index), and these are described in detail by Alley [1984], Karl and
Knight [1985], and Guttman [1991]. The values in the Phoenix area over our 1995-2003
study period ranged from 5.82 in June, 1998 to -4.66 in August, 2002 and showed a
significant trend (p<0.01) toward drought of -0.27 PDSI units year-1 (Figure 6).
4. Land-Use and Socio-Demographic Data
[16] We chose six land-use and socio-demographic variables to account for the
spatial variation in climate sensitivity at the census tract level. The first is the percent of
land covered in “mesic:” irrigated turf, shrubs, and trees. This variable should increase
11
the sensitivity to climate because non-native, heavy-water-using landscape treatments
require more irrigation water to survive during hot, dry, drought-prone periods. The
percentage of mesic coverage was calculated for a previous study by Wentz and Gober
[2007] based on a land cover classification system developed by Stefanov et al. [2001]
and acquired from a 1998 Landsat Thematic Mapper image. They found a significant
positive relationship between the presence of mesic landscape treatments and residential
water use. With each additional 1% of mesic vegetation, water consumption increased by
almost 3,797 liters, which accounts for 0.6% of the typical household’s water
consumption. Moving from zero to 50% mesic would increase the typical household’s
water consumption by 30%. We assume that landscaping practices remained relatively
stable during the timeframe represented.
[17] “Mean household income” is based on data acquired from the 2000 Census
and represents average incomes for residents of single-family units. Previous studies have
shown a significant relationship between income and residential water use, although the
relationship is complex and amenable to a variety of interpretations. Dalhuisen et al.
[2003] and Kallis [1999] showed that water use increases with income, but the
relationship is not significant when water bills constitute a small proportion of disposable
income [Martínez-Espiñeira and Nauges, 2004]. Indoor water use is relatively stable for
people of different incomes [Loh and Coghlan, 2003], but large variations occur with
more discretionary outdoor use. In Phoenix, water is a heavily regulated commodity and
extremely cheap; it does not constitute a large proportion of a household’s income.
Residential customers in the City of Phoenix pay only $1.65 for 2,832 liters during the
12
low-use months of December, January, and February, $1.97 in April, May, October, and
November, and $2.65 in the high-use months of June, July, August, and September.
[18] “Household size” is the average number of persons who occupy a single-
family residence as reported in the 2000 Census. The number of household members has
been shown in a host of studies to influence water use [Mayer and DeOreo, 1999;
Domene and Saurí, 2006; Wentz and Gober, 2007]. More household members result in
more water used to flush toilets, take showers, wash dishes, and launder clothes, although
the per capita use of water tends to decline with household size as economies of scale are
achieved in larger households. It is generally assumed that household size influences
climate-insensitive indoor use rather than outdoor use, although it is also possible that
larger households with more children prefer swimming pools and grass for children to
play and thus influence outdoor use as well.
[19] “Lot size,” in square meters of single-family plots, addresses the amount of
land area potentially exposed to climatic conditions. As in many cities in the United
States, the density of housing units in Phoenix is related to the vintage of those units and
the mode of transportation in vogue when the units were built [Muller, 1995]. Unlike
most cities in the United States, however, newer units in Phoenix tend to be built on
relatively small lots. Lot size grew continuously during the 20th century with increasing
reliance on the automobile and its door-to-door accessibility (Table 1). After a peak in
1970 however, lot sizes declined as land on the fringe became increasingly scarce and
expensive. The typical new development offers densely-packed, often two-story homes
on lots that average only 608 m2 (compared to 763 in 1970). This means that a smaller
portion of the lot is now available for gardens and pools, and thus is climate sensitive.
13
Data for the lot size were acquired from the Maricopa County Assessor’s Office for 2003,
the most recent year available at the time of this study.
[20] The percent of single-family residential lots containing “swimming pools”
was included to incorporate the high water-demand requirements of open water in the
desert environment. Pools require constant refilling as water is lost due to evaporation,
general use, possible leakage, and filtration maintenance. Pan evaporation rates are over a
cm day-1 in June and July (Figure 2). Wentz and Gober [2007] found a significant
positive relationship between the presence of pools and residential water use. A
coefficient of 4,631 liters indicates that increasing the number of households with pools
will increase water use of the typical household by 4,631 liters. While this may not seem
large given an average use of 271,000 liters, a tract where 50% of households have pools
will use 232,000 more liters than one with no pools. A high incidence of pools will
drastically increase residential water use. Using geographically-weighted regression,
Wentz and Gober [2007] documented that the effects of pools on water use varied
spatially, with residential water demand being more responsive to the presence of pools
in the inner urban core than in outlying neighborhoods. They speculated that strong
urban-heat-island effects accelerate evaporation in warmer parts of the central city. The
swimming pool data were acquired from the Maricopa County Assessor’s Office for
2003.
[21] The sixth variable was percent “Hispanic.” Although much has been written
about the differing environmental values and attitudes between Hispanics and other
ethnic groups, relatively little is known about the effects of ethnicity on residential water
use. Campbell et al. [2004] included this variable in a study of water use and
14
conservation behavior in Phoenix, controlling for other relevant characteristics such as
household income, poverty, and age of household members. The purpose of the study
was to assess the relative effectiveness of policy instruments in water conservation.
Hispanic status was a control variable to increase confidence in the independent effects of
the policy-relevant variables. The authors did find a positive relationship between water
use and Hispanic status and interpreted this to mean that Hispanics used more water than
would be expected after the effects of the policy interventions were taken into account.
They speculated that language barriers might preclude Spanish-speaking residents from
consuming the City’s water-conservation messages, but because they did not control for
lot size or vegetation type, we do not know whether cultural practices or the built
environment of Hispanic neighborhoods account for the positive relationships. We have
few expectations regarding outdoor water use among Hispanics except for the fact that
Hispanic households live in older, central city neighborhoods with small lots and might
therefore be expected to have lower water use. Although this variable is exploratory at
this point, it represents a vitally important demographic process, as Hispanics represent a
substantial and increasing percentage of the city’s population [US Census, 1990;
American Community Survey, 2006].
5. Temporal Analyses and Results
[22] Recognizing that various statistical procedures used in our study assume
normally distributed variables over the 108 months from 1995 to 2003, we evaluated the
climate predictors using the standardized coefficients of skewness, z1, and kurtosis, z2,
calculated as:
15
z1 =
(6/N)1/2
∑i=1
N
(xi - X)3 / N ∑
i=1
N
(xi - X)2 / N
-3/2
and
z2 =
(24 / N)1/2
∑i=1
N
(xi - X)
4 / N ∑
i=1
N
(xi - X)
2 / N
-2
- 3
where the resulting z values are compared against a t-value deemed appropriate for a
selected level of confidence (e.g., for N=108, t=1.98 for p=0.05 and t=2.63 for p=0.01). If
the absolute value of z1 or z2 exceeds the selected value of t, a significant deviation from
the normal curve is confirmed. Otherwise, no statistically significant deviation from a
normal distribution is determined (the null hypothesis that the samples came from a
normal distribution cannot be rejected). In addition, we tested the predictors for
deviations from normality using the Kolmogorov-Smirnov one-sample test in which the
predictor is tested against another variable defined as having a normal distribution.
[23] The results indicated that the precipitation anomaly time series contained a
significant deviation from normality with positive skewness and positive kurtosis. A
modified square root transformation in which the sign of the original value is retained
(e.g., a value of -4 transforms to -2) produced a times series without the substantial
deviation from a normal distribution.
[24] For each census tract, the monthly water-consumption values (% of normal)
were compared to the temperature anomalies, transformed precipitation anomalies, and
16
PHDI values. The three climate variables shared little variance (<10%) over the 108
months and can be treated as nearly perfectly independent variables. The Pearson
product-moment correlation coefficients between the temperature anomalies and the
water-consumption percentages through time average 0.32 over the 230 tracts and range
from near zero to 0.51. Therefore, over the 230 census tracts, temperature anomalies
(with the annual cycle removed) generally explain approximately 10% of the variance in
the monthly percentage of normal water use; some tracts have no variance explained by
temperature while others have over 25% of the variance accounted for by the temperature
anomalies. Similarly, the correlation coefficients for the transformed precipitation
anomalies and water-consumption percentages average -0.20 and range from near zero to
-0.52. As we would expect, lower-than-normal precipitation results in higher water use,
although the effects vary substantially across the city. The coefficients between the
consumption values and PHDI through time over the 230 tracts average -0.26 and range
from near zero to -0.51. Not surprisingly, more severe drought conditions stimulate
higher water use. The Pearson product-moment correlation coefficients show the
expected pattern detailed in Balling and Gober [2007] that water consumption increases
when temperatures are relatively high, precipitation levels are low, and drought
conditions exist.
[25] Multiple regression analysis was used to determine the portion of variance
through time in water consumption in each census tract that can be explained by the three
climate-related variables. The coefficient of determination, R2, reveals the portion of
variance in monthly water-consumption percentages of normal explained over the 108
months by the three climate-related variables. The R2 values average 0.25 but range from
17
0.01 to 0.72, indicating that some tracts have effectively no climate sensitivity for their
water consumption while others have a substantial sensitivity to variations in climate. If
we use the 0.05 level of confidence, 80 of the census tracts showed no statistically
significant association between residential water consumption and the climate variables.
A map of explained variance levels (R2 values) shows considerable spatial variability
throughout the metropolitan area (Figure 7), with the highest levels of climate sensitivity
occurring in affluent neighborhoods in the City’s northeastern sector and in new
neighborhoods on the urban fringe. Moran’s I Index, which is a measure of whether a
pattern is clustered, dispersed, or random, equals 0.07 and indicates that spatial
autocorrelation is significant at p<0.01 level of confidence, but indicating only a slight
tendency for spatial clustering of similar values. The distance weights used in computing
the index were based on distances between polygon centroids.
6. Spatial Analyses
[26] We treated the multiple R2 values over the 230 tracts as the primary
dependent variable (a climate sensitivity variable) and the socio-demographic and land-
use indicators as independent variables. However, using the standardized coefficients of
skewness and kurtosis, along with the Kolmogorov-Smirnov one-sample test, we found
that the socio-economic and land use variables typically contained significant deviations
from a normal distribution. We therefore used Spearman rank-order correlation
coefficients to reveal the interrelationships among the independent variables (Table 2).
Particularly strong relationships exist between income and the presence of pools,
household size and percent Hispanic, and percent Hispanic and income. As a result of the
18
multicollinearity, we subjected the independent variables to a normalization procedure
and principal components analysis and derived two factors with eigenvalues greater than
1.0 (our results did not change appreciably with or without the normalization). Together
they explained 78 percent of the variation in socio-demographic and land use
characteristics (Table 3). Component 1 captures nearly 60% of the variance in the data
and shows that high-income, small, non-Hispanic households tend to live in areas with
large lots, swimming pools, and irrigated vegetation. This is the dominant variance
pattern in the spatial data and the one we would expect to be related to the climate
sensitivity variable. Component 2 is difficult to interpret, given that the highest loading
was 0.58 showing that no one variable shared any more than 33% of its variance with this
component that explained less than 20% of the variance in the data. We could see no
reason why this component would be related to the climate sensitivity variable.
[27] The spatial variance in the climate sensitivity variable (the multiple R2 values
for each of the 230 tracts) should be significantly related to spatial variance in the land-
use and socio-economic variables. As seen in Table 4, all of the independent variables,
with the exception of the number of residents per household, are significantly related to
spatial variation in the climate sensitivity of water consumption. Climate sensitivity
increases with high income, large lots, irrigated landscaping, and the presence of
swimming pools; it decreases with the percent Hispanic. The relatively strong
relationship between climate sensitivity and Component #1 reinforces these patterns in
the data (Figure 8). The failure of household size to be significant is consistent with the
idea that the number of residents in the household influences indoor rather than outdoor
water use, and it is outdoor use that drives
19
climate sensitivity.
[28] Each multiple regression equation (one for each of the 230 tracts) linking
water consumption to climate variations over the 108 months has standardized partial
regression coefficients for each of the three independent variables (temperature
anomalies, modified square root of precipitation anomalies, PHDI). The standardized
partial regression coefficients range from -1.0 to 1.0 and are similar to correlation
coefficients between water consumption and the given climate predictor variable. The
standardized partial regression coefficients for the temperature anomaly variable are
positively and significantly related to the presence of swimming pools, suggesting that
the presence of pools increases the sensitivity to temperature. This is not surprising given
that automatic refill devices keep pool levels constant under a variety of temperature
conditions. The sensitivity of water consumption to variations in temperature is also
positively related to Component #1, suggesting that overall affluence of a census tract
(which is highly related to swimming pool variable) increases sensitivity to temperature.
[29] The general lack of significant spatial relationships between the standardized
regression coefficients for precipitation anomalies (showing the strength of the
relationship between water consumption and precipitation through time) and the socio-
economic variables may in part be due to the sporadic nature of precipitation, particularly
for the desert environment. A rain event could occur late in the month and produce a
positive monthly anomaly, when in reality, the bulk of the month was actually dry.
[30] The PHDI variable integrates both temperature and precipitation through
longer periods of time and should provide the clearest climate sensitivity signal. For
example, the first-order (one month) autocorrelation coefficients for the temperature and
20
precipitation anomalies are 0.20 and -0.10, respectively, while the first-order
autocorrelation through time for PHDI is 0.91 (Figure 6). As seen in Table 4, the socio-
economic variables along with Component #1 are strongly related to spatial variations in
the standardized partial regression coefficients for the PHDI variables. Recall that a
negative partial regression coefficient between PHDI and water consumption indicates
that water consumption increases in dry periods when PHDI values are negative. In an
absolute sense, the strongest control on sensitivity of water consumption to drought
conditions are Component #1 and more specifically, the swimming pool variable (Figure
9). The plot clearly shows that as the presence of pools increases, the absolute values of
the negative standardized regression coefficients become larger. Similarly, higher
income, lot size, and percent mesic (all well captured by Component #1) significantly
increase the sensitivity of water consumption to drought.
7. Conclusions
[31] In this investigation, we explored the sensitivity of single-family residential
water consumption to variations in climate throughout several hundred diverse census
tracts in Phoenix, Arizona. We use a series of statistical procedures to reveal the
following:
(1) Consistent with previous studies, we found that residential water consumption
in Phoenix is significantly (p<0.05) related to variations in climate. As expected, water
consumption generally increases when (a) temperatures are above normal, (b)
precipitation is below normal, and (c) the area is in a period of drought. Water
consumption is most strongly related to the drought variable, which is expected because:
21
(a) the drought variable integrates climate effects of variations in both temperature and
precipitation, and (b) the drought variable provides the best overall picture of the
vegetation’s demand for irrigation water and the rate of pool evaporation.
(2) We found that the sensitivity of water consumption to variations in climate
varies substantially from one census tract to another. One third of the tracts have
effectively no sensitivity to climate variations, while in one tract, 72% of the variation in
water consumption is explained by variations in climate.
(3) Greater sensitivity to atmospheric conditions occurred in census tracts with
many pools, a high proportion of high-income residents, larger lot sizes, and a high
proportion of irrigated mesic landscaping. Low climatic sensitivity occurred in
neighborhoods with a relatively high proportion of Hispanic residents.
(4) Generally, the socio-economic and land use variables were weakly related to
how sensitive water consumption is to variations in either temperature or precipitation
alone but significantly related to the sensitivity to drought. The overall affluence of a
census tract significantly increases the sensitivity to drought, with the most important
variable being the percentage of lots in a census tract with swimming pools.
[32] In the coming few decades, Phoenix is expected to double its human
population, the urban heat island effect will increase local temperatures, and the ongoing
buildup of greenhouse gases should further increase temperatures and decrease
precipitation in the region. The net result is that ever more residents may be faced with
ever more droughts in the future. Should these climate changes occur (and they seem to
be occurring at present), water demand in Phoenix will certainly increase, and, as shown
in this study, the increase will be greatest for the more affluent census tracts in the city.
22
[33] A great challenge for the city is to grow in a way that puts less stress on its
water resources and reflects its status as a desert city. One obvious strategy is to
emphasize development that is less climate-sensitive, i.e. development with smaller lots,
fewer swimming pools, and less irrigated vegetation. Public discourse about water
conservation tends to emphasize the wiser use of indoor water, and has tended to ignore
the very important fact that outdoor use represents a much larger share of households’
water budget. Our study suggests that manipulation of the urban form by encouraging
higher densities, restricting the use of irrigated landscaping, and limiting the number of
pools would lower climate sensitivity and make the city’s water use more resilient in the
face of climate change and urban heat island effects.
Acknowledgments
This material is based upon work supported by the National Science Foundation under
Grant No. SES-0345945 Decision Center for a Desert City (DCDC). Any opinions,
findings and conclusions or recommendations expressed in this material are those of the
authors and do not necessarily reflect the views of the National Science Foundation.
23
References
Agthe, D. E. and R. B. Billings (1997), Equity and conservation pricing policy for a
government-run water utility, Aqua- J. Water Supply Res. Tech., 46(5), 252-260.
Alley, W. M. (1984), The Palmer drought severity index-limitations and assumptions. J.
Clim. Appl. Meteorol., 23, 1100–1109.
American Community Survey (2006), Detailed Tables,
<http://factfinder.census.gov/servlet/DatasetMainPageServlet?_program=ACS&_
submenuId=&_lang=en&_ts=> (19 October 2007).
Arizona Department of Economic Security (2007), Arizona Population Projections: 2006
to 2055, Arizona Workforce Informer,
<http://www.workforce.az.gov/?PAGEID=67&SUBID=138> (19 October 2007).
Balling, R. C. Jr., and P. Gober (2007), Climate variability and residential water use in
Phoenix, Arizona. J. Appl. Meteorol. Climatol., 46(7), 1130-1137.
Berry, D. W., and G. W. Bonem, (1974), Projecting the municipal demand for water.
Water Resour. Res., 10, 1239–1241.
Billings, R. B., and D. E. Agthe (1980), Price elasticities for water: A case of increasing
block rates, Land Econ., 56(1), 73-84.
Billings, R. B., and D. E. Agthe (1998), State-space versus multiple regression for
forecasting urban water demand. J. Water Resour. Plan. Manage., 124, 113–117.
Billings, R. B., and W. M. Day (1989), Demand management factors in residential water
use: The southern Arizona experience, J. Amer. Water Works Assoc., 81(3), 58-
64.
24
Campbell, H. E., R. M. Johnson, and E. H. Hunt (2004), Prices, devices, people, or rules:
The relative effectiveness of policy instruments in water conservation, Rev. Policy
Res., 21(5), 637-662.
Cochran, R., and A. W. Cotton (1985), Municipal water demand study, Oklahoma City
and Tulsa, Oklahoma, Water Resour. Res., 21(7), 941-943.
Dalhuisen, J. M., R. Florax, H. L. F. de Groot, and P. Nijkamp (2003), Price and income
elasticities of residential water demand: A meta-analysis, Land Econ., 79(2), 292-
308.
Domene, E., and D. Saurí (2006), Urbanisation and water consumption: Influencing
factors in the metropolitan region of Barcelona, Urban Studies, 43(9), 1605-1623.
Fotheringham, A.S., C. Brunsdon, and M. Charlton (2002), Geographically Weighted
Regression: The Analysis of Spatially Varying Relationships, Wiley, New York.
Gammage, G. (1999), Phoenix in perspective: Reflections on developing the desert, 180
pp., Herberger Center for Design Excellence, College of Architecture and Urban
Design, Arizona State University, Tempe, AZ
Gegax, D., T. McGuckin, and A. Michelsen (1998), Effectiveness of conservation
policies on New Mexico residential water demand, New Mexico J. Science, 38,
104-126.
Gober, P. (2006), Metropolitan Phoenix: Place making and community building in the
desert, 233 pp., University of Pennsylvania Press, Philadelphia, PA.
Guhathakurta, S., and P. Gober (2007), The impact of the Phoenix urban heat island on
residential water use, J. Am. Plan. Assoc., 73(3), 317-329.
25
Guttman, N. B. (1991), A sensitivity analysis of the Palmer hydrologic drought index.
Water Resour. Bull., 27(5), 797–807.
Gutzler D. S., and J. S. Nims (2005), Interannual variability of water demand and
summer climate in Albuquerque, New Mexico, J. Appl. Meteorol., 44(12), 1777–
1787.
Kallis, G. (1999), Geography of Metropolitan Areas and the Use of Water, Report
prepared for the METRON project.
Karl, T. R., and R. W. Knight (1985), Atlas of monthly Palmer hydrological drought
indices (1931–1983) for the contiguous United States. Historical Climatology
Series 3–7, 319 pp., National Climatic Data Center, Asheville, North Carolina
Karl, T. R., and C. N. Williams, Jr. (1987), An approach to adjusting climatological time
series for discontinuous inhomogeneiti, J. Appl. Meteorol., 26(12), 1744–1763.
Karl, T. R., C. N. Williams Jr., P. J. Young, and W. M. Wendland (1986), A model to
estimate the time of observation bias associated with monthly mean maximum,
minimum and mean temperatures for the United States, J. Appl. Meteorol., 25(2),
145-160.
Karl, T. R., C. N. Williams Jr., F. T. Quinlan, and T. A. Boden, (1990), United States
historical climatology network (HCN) serial temperature and precipitation. Dept.
of Energy, Oak Ridge National Lab. ORNL/CDIAC-30, NDP-019/R1, 83 pp. plus
appendixes.
Loh, M., and P. Coghlan (2003), Domestic Water Use Study in Perth, Western Australia
1998-2001, 36pp., Water Corporation, Perth, WA.
Maidment, D. R., and E. Parzen (1984), Time patterns of water use in six Texas cities. J.
Water Resour. Plan. Manage., 110, 90–106.
26
Martin, W. E., and S. Kulakowski (1991), Water price as a policy variable in managing
urban water use: Tucson, Arizona, Water Resour. Res., 27(2), 157-166.
Martínez-Espiñeira, R., and C. Nauges (2004), Is all domestic water consumption
sensitive to price control? Applied Econ., 36, 1697-1704.
Mayer, P. W. and W. B. DeOreo (eds.), (1999), Residential End Uses of Water, American
Water Works Association Research Foundation, Denver, CO.
Michelsen, A. M, J. T. McGuckin, and D. Stumpf (1999), Nonprice water conservation
programs as a demand management tool, J. Am. Water Res. Assoc., 35(3), 593-
602.
Muller, P. O. (1995), Transportation and Urban Form: Stages in the Spatial Evolution of
the American Metropolis in S. Hanson, ed., The Geography of Urban
Transportation, 424 pp., Guilford Press, New York, NY.
Newman, J.E., and J.E. Oliver (2005), Palmer Index / Palmer Drought Severity Index, in
J.E. Oliver, ed., Encyclopedia of World Climatology, 571-573, Springer,
Dordrecht, Netherlands.
Palmer, W. C. (1965), Meteorological drought. US Weather Bureau Res. Paper 45,
58pp., Washington, District of Columbia.
Phoenix Water Resources Plan (2005), Water Services Department, Water Resources and
Development Planning Section.
Quayle, R. G., D. R. Easterling, T. R. Karl, and P. Y. Hughes (1991), Effects of recent
thermometer changes in the Cooperative Station Network, Bull. Am. Meteorol.
Soc., 72(11), 1718–1723.
27
Rhoades, S. D., and T. M. Walski (1991), Using regression analysis to project pumpage.
J. Am. Water Works Assoc., 83, 45–50.
Stefanov, W. L., M. S. Ramsey, and P. R. Christensen (2001), Monitoring urban land
cover change: an expert system approach to land cover classification of semiarid
to arid urban centers. Remote Sens. Environ., 77(2), 173–185.
Seager, R., M. Ting, I. Held, Y. Kushnir, J. Lu, G. Vecchi, H.-P. Huang, N. Harnik, A.
Leetmaa, N.-C. Lau, C. Li, J. Velez, and N. Naik (2007), Model projections of an
imminent transition to a more arid climate in Southwestern North America,
Science, 316(5828), 1181-1184.
Wentz, E., and P. Gober (2007), Determinants of small-area water consumption for the
City of Phoenix, Arizona, Water Resour. Manage., 21, 1849-1863.
Wilson, L. (1989), Addition of a climate variable to the Howe and Linaweaver western
sprinkling equation. Water Resour. Res., 25(6), 1067–1069.
Woodard, G. C., and C. Horn (1988), Effects of Weather and Climate on Municipal
Water Demand in Arizona, Report prepared for the Arizona Department of Water
Resources and Tucson Water, Division of Economic and Business Research,
College of Business and Public Administration, University of Arizona, Tucson,
Arizona, August, 1988.
US Bureau of the Census (1990), Decennial Census, Summary File 1, Detailed Tables,
<http://factfinder.census.gov/servlet/DatasetMainPageServlet?_program=DEC&_
tabId=DEC2&_submenuId=datasets_1&_lang=en&_ts=210863607521> (19
October 2007).
28
Table 1. Median Single-Family Residential Lot Size (m2) in Phoenix, 1900-2000
Year 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 Lot Size
604 639 639 654 692 699 701 763 725 625 608
Source: DCDC using records from the Maricopa County Assessor’s Office
29
Table 2. Spearman rank-order correlation coefficients among the land-use and socio-economic variables across the 230 census tracts. Absolute values >0.14 and >0.18 are significant at the 0.05 and 0.01 levels of confidence.
Mesic
Household Size
Lot Size
Pools
Percent Hispanic
Household Income
.30 -.58 0.50 0.85 -0.61
Mesic -0.43 0.52 0.29 -0.30 Household
Size -0.32 -0.45 0.78
Lot Size 0.53 -0.28 Pools -0.71
30
Table 3. Unrotated Principal Component Loadings and Communalities Variable Component 1 Component 2 Communality Household Income .890 .199 .832 Mesic .633 .331 .510 Household Size -.696 .571 .811 Lot Size .714 .580 .846 Pools .872 .023 .760 Hispanic -.806 .525 .925 Eigenvalue 3.595 1.089 4.684 % Variance .599 .181 .781
31
Table 4. Non-parametric Spearman rank-order correlation coefficients between socio-economic predictors for 230 census tracts and (a) the total temporal variance in water consumption explained by all three climate variables, and the sensitivity of water consumption to changes in (b) temperature, (c) precipitation, and (d) PHDI (values significant at the 0.05 level are in plain text and those significant at the 0.01 level are in bold text)
Predictor
Total Variance
Temp
Prec
PHDI
Income / Household 0.23 -0.53 Percent Mesic 0.18 -0.26
Number in Household 0.22 Lot Size 0.20 -0.15 -0.29
Pool Percent 0.32 0.16 -0.62 Percent Hispanic -0.27 0.49 Component #1 0.30 0.17 -0.63
32
Figure 1. Map of the ratio of summer to winter mean water consumption
33
0
2
4
6
8
10
12
J F M A M J J A S O N D
Month
Eva
po
rati
on
(m
m/d
ay)
Figure 2. 2003 monthly pan evaporation rates (mm) in the Phoenix area (specifically,
the Arizona State University campus in Tempe).
34
Figure 3. Map showing the “Phoenix” study area within the much larger metropolitan
area.
35
Figure 4. Map of single-family residential water consumption (liters per year) by census
tracts.
36
0
10,000
20,000
30,000
40,000
50,000
60,000
70,000
80,000
90,000
J F M A M J J A S O N D
Month
Co
nsu
mp
tio
n (
lite
rs)
Figure 5. Mean monthly single-family residential water consumption (liters) in the
Phoenix metropolitan area.
37
-6
-4
-2
0
2
4
6
8
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
Time
PH
DI
Figure 6. Monthly PHDI values from January 1995 to December 2003.
38
Figure 7. Map of R2 values relating single-family residential water consumption to
variations in climate.
39
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
-2 -1 0 1 2 3 4
Component #1
Exp
lain
ed V
aria
nce
Figure 8. Plot of explained variance (water consumption explained by climate) versus
component #1 (highly related to income and the presence of swimming pools).
40
Figure 9. Water consumption sensitivity drought level (PHDI) versus percent swimming
pool coverage by census tract. The larger negative values for PHDI Sensitivity indicate a
stronger relationship between water consumption and PHDI.
-0.6
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0 20 40 60 80 100
Pool Percentage
PH
DI
Sen
siti
vity
Recommended