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