Introduction

Cities around the world are facing challenges to managing water resources since freshwater sources are diminishing, droughts as well as other hydrometeorological extreme events are increasing their frequency, causing the development of new supplies to become environmentally and politically unfeasible (^{Baumann & Boland, 1998}). These challenges are related to a constant escalation of urban water demand (^{Saurí, 2013}), linked to the growing population residing in cities that represents half of the world’s population (^{UN, 2018}). Integrated Urban Water Management (IUWM) emerged as an alternative paradigm to tackle these issues by integrating three components of urban water services: supply, drainage, and sanitation. IUWM seeks to integrate land-use planning and economic development to overcome fragmentation in public policy formulation to accomplish economic, social, and environmental goals (^{Bahri, 2011}). This paradigm carries out sustainable development within the urban water sector (^{Tiburcio & Perevochtchikova, 2012}).

IUWM is a demand-side paradigm that conceives water needs to be multifaceted and not merely a question of quantity, in contrast to the traditional water management paradigm (^{Mitchell, 2006}). Therefore, all water uses are considered, as well as quality and quantity for final uses (^{Pinkham, 1999}). These conceptions target the supply component of IUWM, highlighting the need to balance water supply and demand by focusing on reducing water uses and losses, including unaccounted water in a system (^{Baumann & Boland, 1998}). The goals of this balance are meeting human and ecosystem needs, in addition to selecting the most efficient options to manage water (^{Saurí, 2013}; ^{Chang, 2017}). Urban water demand takes place in natural and social systems at different temporal and spatial scales with different dynamics (^{House-Peters & Chang, 2011}). Both systems must be considered to attain the integration of water management components (^{Tiburcio & Perevochtchikova, 2012}).

Different micro and macro-scale urban water demand determinants have been identified in case studies from Australia, Europe, and the United States (^{Worthington & Hoffman, 2008}; ^{March & Saurí, 2009}). In contrast, case studies from urban areas of developing countries are still lacking (^{Saurí, 2013}). In Latin America just two studies were found, one in the City of Fortaleza, Brazil (^{De-Maria-André & Carvalho, 2014}), and the other for the City of Hermosillo in Mexico (^{Ojeda, Álvarez, Ramos, & Soto, 2017}).

One macro-scale determinant of residential water use is urban density (^{Chang, 2017}), which has been found to reduce water use (^{Saurí, 2013}). Case studies carried out in the cities of Austin, Phoenix, Portland, and Salt Lake City are evidence of this statement (^{Chang, Hossein, & Shandas, 2010}; ^{Shandas & Parandvash, 2010}; ^{Stoker et al., 2019}). Similar evidence was found in the city of Auckland, New Zealand (^{Ghavidelfar & Shamseldin, 2017}). In these cases, urban density is negatively associated with water use, while the presence of gardens and pools tend to increase it, as both are typical characteristics of low urban density dwellings (^{Domene & Saurí, 2006}; ^{Wentz & Gober, 2007}; ^{De-Maria-André & Carvalho, 2014}; ^{Kontokosta & Jain, 2015}).

Income and household size are the most mentioned socioeconomic determinants of water use, which typically are positively related (^{Arbués, García-Valiñas, & Martínez-Espiñeira, 2003}; ^{Arbués & Villanúa, 2006}; ^{Domene & Saurí, 2006}; ^{Wentz & Gober, 2007}; ^{Worthington & Hoffman, 2008}; ^{De-Maria-André & Carvalho, 2014}; ^{Kontokosta & Jain, 2015}). Some other relevant determinants have been studied, such as inhabitants with college education and the number of internal water connections (^{Babel, Das Gupta, & Pradhan, 2007}); race and ethnicity, and land use regulation (^{Turner & Ibes, 2011}); gender and the number of bathrooms (^{De-Maria-André & Carvalho, 2014}; ^{Ojeda et al., 2017}); and energy consumption (^{Kontokosta & Jain, 2015}). It is relevant to consider the role of water measurement, as it reduces residential water demand (^{Worthington & Hoffman, 2008}). Evidence on this point has been found in three Californian cities (^{Tanverakul & Lee, 2015}), as well as in the city of Hermosillo, Mexico (^{Ojeda et al., 2017}).

Different temporal and spatial scales have been used in all these studies, ranging from parcels to city level, and also from daily to annual time series. Depending on the target of required actions, spatial and temporal scales are chosen, as determinants of water demand vary over space and time (^{Billings & Jones, 2008}; ^{Chang, 2017}). Spatial research on urban water demand has increased since 2001; it has allowed to understand the roles of location and adjacency to explain water use (^{House-Peters & Chang, 2011}).

Some studies have used spatial lag and spatial error regression analyses with units such census tracts (^{Chang et al., 2010}; ^{House-Peters, Pratt, & Chang, 2010}; ^{Chang, Bonnette, Stoker, Crow-Miller, & Wentz, 2017}; ^{Stoker et al., 2019}). These two analyses are spatially stationary and global. In turn, Geographically Weighted Regression (GWR) is a spatially explicit non-stationary local analysis that approaches varying spatial relationships among variables, giving local statistics for each spatial unit under study (^{Fotheringham, Brunsdon, & Charlton, 2002}). GWR has been employed to estimate local coefficients associating water consumption to different single family residential variables, across census tracts in Phoenix (^{Wentz & Gober, 2007}). This analysis has also been used with multi-family residential water use in New York City (^{Kontokosta & Jain, 2015}).

As previously mentioned, urban water demand determinants have been lightly studied in Latin America; this is the case of Mexico City, one of the megacities of this region and the capital of Mexico (^{Perevochtchikova, 2015}). The relevance of studying these determinants, emerge as the city is lacking a precise diagnosis on this topic (^{GDF, 2007}; ^{GDF, 2012}), while demand has been found to be a pressure over groundwater resources which threatens the long term sustainability of the city (^{Martínez, Escolero, & Perevochtchikova, 2015}). Summarizing, the city has been facing multiple challenges such as vulnerable water supply infrastructure (^{Escolero, Kraslisch, Martínez, & Perevochtchikova, 2016}), social and political conflicts (^{Perló & González, 2009}), water shortages and intermittent water supply (^{Morales & Rodríguez, 2009}), as well as varying access to water services (^{Jiménez, Gutiérrez, Marañón, & González, 2011}). Comprehending the demand-side of water management is the first step towards the development of water conservation strategies applying IUWM principles.

In this sense, the present study proposes the analysis and estimation of residential water demand for the year 2010 in Mexico City based on a local scale, considering socioeconomic, urban density, and access to water services determinants applying two models: one global and stationary and the other local, spatially explicit and non-stationary. The contribution of this work consists in proposing a useful tool to better comprehend the complex water crisis of Mexico City from a water demand-side view, as well as to advice the decision-making process from the IUWM principals.

Water Management in Mexico City

Mexico City has 8.85 million inhabitants, it spreads over an area of 7 854 km² housing 22 815 504 people (^{INEGI, 2012}). This urban complex is located inside the Hydrological Administrative Region (HAR) XIII “Waters of the Valley of Mexico” in terms of national water management (Figure 1). The HAR XIII presents the highest degree of pressure over water resources (141 %), and has the lowest annual water availability of the country with 144 m³ per person. Mexico City generates 16.97 % of the national Gross Domestic Product (GDP), while its per capita water availability is only 55 m³ (^{Conagua, 2018}).

Figure 1 Mexico City location (Source: self-elaborated with data from ^{INEGI (2012)}; ^{Conagua (2018)} and information form Conagua website https://app.conagua.gob.mx/consultarepda.aspx). 

Water management in Mexico City is complex, as it involves actions from different institutions at national and local levels (^{Perevochtchikova, 2015}). The Mexican National Water Commission (Conagua, abbreviation in Spanish) is a federal institution in charge of allocating permits on water volumes to the national water users (^{Conagua, 2009}); while the Waters System of Mexico City (Sistema de Aguas de la Ciudad de México, abbreviation in Spanish: Sacmex) is the local water utility in charge of providing water services in Mexico City (^{ALDF, 2003}). Sacmex is a national water user under Conagua regulation. Both insitutions are involved in Mexico City water supply, as infrastructure (Figure 1) links artificially three different basins (Cutzamala, Lerma and the Basin of Mexico). Average water supply for the period 2008-2013 was 31.33 m³/s (^{Sacmex, 2014b}), it was withdrawn from different surface and groundwater sources that are managed by either Conagua or Sacmex as Table 1 summarizes. Groundwater from inside Mexico City’s territory is the most important source, the remaining water quantity is obtained from distant sources.

The Lerma and Cutzamala Systems were built during the 20^th century to deal with the increasing population water needs of Mexico City, as well as to diminish land subsidence inside the Basin of Mexico due to groundwater extraction (^{Perló & González, 2009}). This situation pictures the traditional water management paradigm, as decisions were taken by only conceiving water demand in terms of population growth (^{Tiburcio & Perevochtchikova, 2012}), which increased from 0.37 million people in 1900 to 8.85 million in 2010 (^{GDF, 2012}; ^{INEGI, 2012}). Despite these efforts, all water supply systems are currently working under high restrictions, being the Cutzamala System the most vulnerable water source (^{Escolero et al., 2016}), while local groundwater extraction is causing severe drawdowns of potentiometric levels in the Basin of Mexico (^{Martínez et al., 2015}).

Besides this issue ^{Morales and Rodríguez (2009)}, based on Conagua Water Statistics in Mexico bulletin of 2004, mention the existence of a 10.24 m³/s water deficit between natural water availability and allocated quantity that causes water restrictions and intermittent water supply (tandeo in Spanish) across the city. In parallel, according to ^{Jiménez et al. (2011)} access to water services does not spread evenly in Mexico City. These authors define the most proficient level of access to water services as a 24 hour serving internal water connection, with highly drinkable water quality. Water losses are another relevant issue since around 40 % of Mexico City water supply is estimated to be lost due to leakage (^{Perló & González, 2009}; ^{GDF, 2012}; ^{Sacmex, 2019}).

In addition, since 1994 the local government has focused on increasing water demand measurement (^{Martínez, 2004}), but this target has not been fully accomplished. Average water measurement coverage for the period 2008-2013 ranged for 63 % of residential water users, 70 % for non-residential users, and 82 % for the mixed water use (^{Sacmex, 2014a}). These users refer to a person hiring water services from Sacmex according to the local legislation (^{ALDF, 2003}). Based on the 2010 national census data from the National Institute of Statistics and Geography (INEGI), 2.4 million dwellings were reported to have an internal water connection, while Sacmex counted 2 million residential water users, out of which 1.29 million had a metering device (^{GDF, 2012}: 35). For the year 2010, 64.25 % of residential water users had a metering device generating 9.95 m³/s of residential water demand (^{Sacmex, 2014a}). Figure 2 shows a comparison of water supply and metered water uses for the period 2008-2013.

Figure 2 Water uses and water supply in Mexico City, 2008-2013. Source: self-elaborated with data obtained from the transparency portal of ^{Sacmex (2014a)} and ^{Sacmex (2014b)}. 

Although the IUWM principles have been introduced as a guideline for Sacmex, the local governmental programs mainly focus on infrastructure development strategies (^{ALDF, 2003}; ^{GDF, 2012}). Particularly, water demand is mainly considered an issue of population growth and no other determinant or diagnosis is given (^{GDF, 2012}). In contrast, population growth in the city has been showing a stabilizing trend that is projected to decrease until 2030 (^{Downs, Mazari-Hiriart, Domínguez-Mora, & Suffet, 2000}; ^{Perló & González, 2009}; ^{Conagua, 2018}). This highlights the relevance of approaching water demand determinants other than population growth to understand residential water demand comprehensively.

Water demand in Mexico City has been lightly studied in comparison to other topics around water management. ^{Downs et al. (2000)} forecasted water demand for the whole Metropolitan Area of Mexico City until 2015, based on a probabilistic model of population growth extrapolating some socioeconomic characteristics of one district (alcaldía in Spanish) within the city. Their results accounted for 26 m³/s in Mexico City, which is similar to an estimation of 22.3 m³/s reported by Sacmex in GDF (2007:13). In addition, per capita water use correlates positively with socioeconomic variables and negatively to household size (^{Montesillo, 2004}; ^{Dávila & Constantino, 2007}). These studies were performed at a municipality and district level comprising huge spatial heterogeneity; on the other hand, the use of measured data is not mentioned either. In contrast, ^{Ramos-Bueno (2016)} analyzed measured data of residential water demand on a neighborhood scale. However, this study is still limited as intermittent water supply was not included and also global spatially stationary techniques were used.

Data and methods

Water demand is defined as “the total volume of water necessary or needed to supply customers within a certain period of time” (^{Billings & Jones, 2008}: 4). In this context, empirical studies referring to either residential water demand (^{Downs et al., 2000}; ^{Arbués & Villanúa, 2006}; ^{Babel et al., 2007}; ^{Wentz & Gober, 2007}; ^{Shandas & Parandvash, 2010}; ^{Turner & Ibes, 2011}; ^{De-Maria-André & Carvalho, 2014}; ^{Ashoori, Dzombak, & Small, 2016}; ^{Ghavidelfar & Shamseldin, 2017}); water consumption (^{Domene & Saurí, 2006}; ^{Chang et al., 2010}; ^{House-Peters et al., 2010}; ^{Ojeda et al., 2017}); or water use (^{Kontokosta & Jain, 2015}; ^{Chang et al., 2017}), are used interchangeably. The present study uses the term residential water demand which corresponds to measured residential water use during 2010.

Results

Descriptive statistics of input variables appear in Table 3; they show data transformed into logarithmic form.

Correlation matrix and variables spatial distribution

Figure 3 shows the correlation matrix of all variables, which are significantly correlated to the dependent variable. Dwellings with internal water connection (VA) and residential water demand (AN) are strongly and positively correlated. Household size (TH) and intermittent water supply (TA) exhibit negative signs, showing an inverse correlation with residential water demand. The highest sign of the whole matrix is between the Social Development Index (ID) and household size. To avoid multicollinearity issues household size was excluded from the OLS and GWR analyses, remaining just five independent variables for these procedures. It is relevant to notice the negative sign of this correlation.

Figure 3 Correlation matrix between all variables under analysis: Annual residential water demand (AN), residential users with metering device (PC), total dwellings with internal water connection (VA), average Social Development Index (ID), intermittent water supply (TA), household size (TH) and dwelling density (DV). 

The LISA analyses for the dependent variable and the five remaining input variables appear in Figure 4. Results from the Moran’s I index were positive values ranging from 0.34 to 0.65; this indicates that the variables under study present significant spatial patterns across the city neighborhoods. Maps locating high and low values of each variable appear in dark red and light blue, respectively.

Figure 4 Local Indicators of Spatial Autocorrelation (LISA) for the dependent and independent variables for the analysis of annual residential water demand in Mexico City neighborhoods (clusters of high values are in dark red and clusters of low values are in light blue). 

The Average Social Development Index and intermittent water supply exhibited the Moran’s I highest values of 0.65 and 0.57, respectively (Figures 4D and 4E). They were followed by the percentage of residential water users with a metering device (Figure 4B), which had an autocorrelation of 0.49. The remaining three variables had Moran’s I values around 0.34 and 0.37, which are moderate but still statistically significant. Figures 4A and 4C show similar spatially clustered patterns, but they both differ from Figure 4B, showing different spatial patterns among the location of dwellings, metering device coverage, and high levels of residential water demand.

Cluster of high values for the ID in Figure 4D are mainly located in the western side of the city, while low values appear in the southern and eastern sides. Some high residential water demand values from Figure 4A overlap with some others of the Social Development Index in Figure 4D. The case of intermittent water supply, shown in Figure 4E, locates a spatial pattern of incidence mainly in the southwestern edge of the city, which corresponds to the mountain limits of the Basin of Mexico (Figure 1). Finally, dwelling density in Figure 4F shows a spatial cluster of low density in the western area and some neighborhoods in the southern part of the city.

Geographically Weighted Regression results

Results from GWR analysis stemming from Equation (2) of section “Independent variables selection” appear in Table 5. They show a little improvement over the R² compared to the OLS, as it is 2 % higher. In contrast, the AIC of the GWR analysis is smaller than the OLS, suggesting better adjustment. Signs and magnitudes form the coefficients vary according to their location, contrasting with results in Table 4.

Table 5 Geographically Weighted Regression (GWR) coefficients of a log-log form model to estimate residential water demand in Mexico City neighborhoods (obtained from Equation (2)). 

Variable	Minimum	Median	Maximum	Range
Intercept	0.25	2.42	5.11	4.86
Residential water users with metering device (%)	0.05	0.85	1.18	1.13
Dwellings with internal water connection (total count)	0.91	0.98	1.07	0.15
Average Social Development Index (values from 0 to 1)	0.78	1.47	2.24	1.45
Dwelling density (ha)	-0.35	-0.06	0.21	0.57
Intermittent water supply (dummy)	-5.49	-0.32	0.22	5.71
R2	0.89
AIC	2 091.96
n	1 346

The extent of the control variable coefficients grows from 0.05 to 1.18, showing bigger variation than dwellings with internal water connection. Dwelling density and intermittent water supply exhibit most of their coefficients with negative signs. In contrast, some spatial units have positive signs, meaning that water demand would increase in terms of these two variables. Elasticities of the Social Development Index vary from 0.78 to 2.24; although this determinant has one of the highest values it is not the case for all spatial units. These are some relevant differences that contrast with the global stationary OLS analysis. Local coefficients of the study area are displayed in Figure 5.

Figure 5 Quantile spatial distribution of local coefficients from the Geographically Weighted Regression (GWR) model to estimate residential water demand in Mexico City neighborhoods. 

Figure 5A displays local coefficients of the control variable, where it can be observed low values in the central area while high values are located on the southwestern side; this means that water demand is less elastic to metering devices in central neighborhoods in contrast with the peripheral areas. The case of dwellings with internal water connection in Figure 5B shows a spatial pattern of high elasticity values in the north that diminishes southwards, although the lowest values are clustered in the western side; this determinant exhibits the lowest range of spatial variation (Table 5). The Social Development Index in Figure 5C shows high levels of elasticity with residential water demand in the central-western side of the city, while the lowest values are located in three separate clusters, one in the central area, and the other two in the eastern and western peripheries.

Dwelling density in Figure 5D shows some positive signs in the southeastern peripheral neighborhoods, meaning these determinants would increase residential water demand in these areas, although these coefficients are not high, a 10 % increase would be translated into 2 % of extra demand. The case of intermittent water supply is relevant to acknowledge, as it presents the highest spatial variation. Since this determinant represents a restriction to access to water services, the central and northern areas would be more affected in case those neighborhoods had this type of water supply, which would create a substantial reduction of residential water demand in the neighborhoods having the highest values of the dependent variable.

The characteristics of the spatial clusters from Figure 5 appear in Table 6. Local values for residential water demand, average percent of residential water users with metering device, total number of dwellings with internal water connection, average household size, average Social Development Index, average dwelling density and the total number of neighborhoods with intermittent water supply were assembled in terms of the elasticities calculated by the GWR analysis.

Table 6 Characteristics of spatial clusters in terms of quantile elasticities distribution. Spatial clusters: Residential users with metering device (PC), total dwellings with internal water connection (VA), average Social Development Index (ID), dwelling density (DV) and intermittent water supply (TA). 

Spatial Cluster	Elasticities	Annual residential water demand (million cubic meters)	Residential water users with metering device (%)	Dwellings with internal water connection (per thousand)	Average household size (person per dwelling)	Average Social Development Index	Dwelling density (per hectare)	Total neighborhoods with intermittent water supply
PC	0.051 - 0.381	87.37	64.35	565.40	3.34	0.86	49.14	0
	0.381 - 0.696	67.90	76.25	405.49	3.55	0.85	47.28	2
	0.696 - 0.936	72.12	79.27	509.01	3.71	0.82	48.50	20
	0.936 - 1.001	52.83	71.83	415.97	3.82	0.80	37.83	19
	1.001 - 1.178	33.12	58.61	275.08	3.86	0.76	29.22	87
VA	0.912 - 0.937	34.46	73.21	233.43	3.92	0.76	43.27	20
	0.937 - 0.967	46.93	68.38	298.44	3.61	0.83	31.80	44
	0.967 - 1.003	71.95	68.74	537.49	3.69	0.81	42.18	40
	1.003 - 1.027	85.95	70.88	579.09	3.53	0.85	48.92	6
	1.027 - 1.067	74.03	69.06	522.50	3.52	0.84	45.81	18
ID	0.784 - 1.201	60.16	58.87	517.49	3.87	0.75	37.58	55
	1.201 - 1.416	59.77	65.62	415.20	3.60	0.82	42.81	39
	1.416 - 1.541	56.12	74.97	367.50	3.62	0.84	38.33	7
	1.541 - 1.730	60.34	75.45	422.16	3.68	0.83	42.94	26
	1.730 - 2.237	76.94	75.42	448.60	3.49	0.85	50.34	1
DV	-0.354 - -0.249	43.77	71.26	281.34	3.77	0.80	43.54	60
	-0.249 - -0.087	75.01	71.21	501.41	3.49	0.85	43.11	31
	-0.087 - -0.038	70.71	74.56	461.16	3.51	0.86	44.32	5
	-0.038 - 0	106.51	69.56	765.01	3.58	0.83	46.77	24
	0 - 0.214	17.32	60.64	162.03	4.08	0.71	27.72	8
TA	-5.487 - -0.783	79.07	71.11	519.95	3.38	0.87	45.14	0
	-0.783 - -0.512	73.52	72.27	449.33	3.52	0.83	53.41	22
	-0.512 - -0.246	48.34	66.49	355.85	3.77	0.80	41.50	49
	-0.246 - 0	91.03	76.16	635.90	3.66	0.84	39.59	49
	0 -0.221	21.36	59.38	209.91	4.04	0.72	29.47	8
Mexico City		313.33	70.06	2170.95	3.65	0.82	42.40	128

High residential water demand clusters are characterized by the highest number of dwellings with internal water connection, small average household size, high average Social Development Index, and high dwelling density. These areas are also characterized by a low number of neighborhoods with intermittent water supply. According to Table 6, high coverage of metering devices does not necessarily show the highest levels of residential water demand.

Residential water demand prediction and estimation

Figure 6 shows the predicted values of residential water demand stemming from both models. These results are compared to observed data from Sacmex; in both cases an underestimation is present. Two estimations of residential water demand in Mexico City are presented based on coefficients from OLS and GWR analyses. Estimations yielded a result of 13.70 m³/s with OLS coefficients and 15.33 m³/s with GWR coefficients.

Figure 6 Observed, predicted, and estimated annual residential water demand with OLS and GWR models in Mexico City, 2010 (Estimated residential water demand columns already include unaccounted residential water quantity). 

Residential water demand estimation from the OLS model represents 43.43 % of total supply to Mexico City (31.54 m³/s) while the GWR estimation accounts for 48.59 %. These results highlight the need to address unaccounted water quantity as well as water losses in comparison to water supply, to create a demand-side view of water management encompassing IUWM principals to impact on the water supply component.

Discussion

Mexico City’s central area is the top priority for residential water demand reduction. It is characterized by a cluster of neighborhoods with high values of residential water demand with low levels of metering coverage. These values are also related to high numbers of residential dwellings with internal water connection, irrespective of their metering coverage. This point is similar to the findings of ^{Tanverakul and Lee (2015)} and ^{Ojeda et al. (2017)}, where places without measurement tended to use more water in comparison to areas with good metering coverage. Peripheral areas, although showing high elasticities for residential water demand in terms of metering coverage, also show high levels of intermittent water supply which represent a restriction of access to water services. This point highlights differences in water demand in the center and the edges of the city, although both account for a similar level of metering water coverage. Management actions must take this point into consideration.

Residential water demand in Mexico City is linked to public services provision. High levels of residential water demand are associated with high levels of the Social Development Index, small household size, and low incidence of intermittent water supply. Similar results were reported by ^{Montesillo (2004)}, and ^{Dávila and Constantino (2007)}, where similar socioeconomic variables were correlated to water use, while ^{Jiménez et al. (2011)} mentioned poverty levels were related to defficient access to water services. Following the spatial clusters of the index, a central-western pattern of neighborhoods with high values is found which might be related to the urban development pattern of the city, as eastern neighborhoods are the result of consolidated slums that emerged without urban planning. As small household size appears in neighborhoods with high residential water demand, a study of water use intensity is warranted.

Dwelling density effects on residential water demand must be taken in a broader context. Although the global OLS analysis shows dwelling density with a negative sign, similar to ^{Chang et al. (2010)}, ^{Shandas and Parandvash (2010)}, ^{Ghavidelfar and Shamseldin (2017)}, and ^{Stoker et al. (2019)}, some negihborhoods exhibit positive coefficients as a result of the GWR analysis (^{Fotheringham et al., 2002}). These coefficients are located in the southeastern side, the rural area of the city, where densification would create extra water demand. High density areas are also those having high levels of residential water demand and Social Development Index. Although dwelling density might have the potential to reduce water demand other variables have stronger effects.

Intermittent water supply plays an important role in explaining residential water demand. Since deficient access to water services is a relevant concern in Mexico City, calculated local coefficients show that central and northern neighborhoods would be more affected when facing this type of supply. Neighborhoods in the southeastern side present positive coefficients to this variable, which coincide with the location of the Southern system of water supply. In turn, the Northern and Central systems provide around 10 % of total water supply, where the highest coefficients of intermittent water supply are located. This result suggests a high dependence of the central and northern areas on exterior sources, but it also suggests their vulnerability to intermittent water supply in case of happening, despite these areas are also characterized by high levels of infrastructure. Therefore, the water deficit mentioned by ^{Morales and Rodríguez (2009)} and the low water availability statement from the local water management programs (^{GDF, 2007}; ^{GDF, 2012}), are not a mere question of quantity. It is still needed to understand the cause of intermittent water supply reported in the local government official newspaper, as well as to explore the effects of hourly supply frequency and water quality issues on residential water demand.

As IUWM states the need to address water demand in a multifaceted way (^{Mitchell, 2006}) and also to look for the best options to balance water supply and demand (^{Baumann & Boland, 1998}), the case of Mexico City should focus on the spatial distribution of socioeconomic characteristics, dwelling density and access to water services instead of just population growth. This is one turning point to conceive water management from a demand-side view.

Conclusions

In this study a diagnosis and two estimations of residential water demand in Mexico City were exposed, based on a OLS multiple regresion and a GWR models. Socioeconomic and access to water services variables resulted the most important determinants of residential water demand rather than urban density, highlighting the IUWM principal of water demand being more than a question of quantity; these type of analyses are useful to impact in urban water supply management, as well as on the evaluation of different sources. Central neighborhoods are the priority area to implement water demand reduction. The exploratory estimations of residential water demand for the whole city offer an insight on the need to address water losses. Resultant coefficients are helpful to advice public policymakers on decisions related to water management, to face the current water challenges of Mexico City.

More research is still needed on access to water services and its relation with residential water demand since variables such as hourly water frequency provision and drinking water quality would give more insights. Balance of water demand and supply must be approached from a spatial view, as provision areas are not the sites where water is consumed. Residential water use intensity must be studied as neighborhoods with small household sizes are also those consuming high volumes of water. Another relevant issue to examine is the role of water losses in explaining water demand. These points constitute a complement to the results obtained from this study to encompass a water demand-side view from IUWM for Mexico City.