Kuwait household water demand in 2050: Spatial microsimulation and impact appraisal

Household water demand has increased dramatically in Kuwait over the last few decades, due to rapid population growth and changing lifestyles. Avoiding a water deficit through a supply‐side approach has been the default strategy in Kuwait, yet this approach is unsustainable, associated with declining groundwater levels, and reliance on desalination that results in major carbon emission and environmental impact and that takes a large and growing share of oil revenues. In this study, we forecast household water demand in Kuwait to 2050 under a Business‐As‐Usual (BAU) scenario and evaluate the economic and environmental impacts. A spatial microsimulation, constrained by the national population projection of the Kuwait Institute of Scientific Research (KISR), was developed to overcome data limitations in forecasting household demand. Results show a 45% increase in water demand by 2050, to 664.1 million cubic metres (MCM), relative to the 2019 base year. Annual production costs increase from 1.39 billion USD in 2019 to 1.99 billion USD by 2050, whilst carbon emissions increase from 10.85 to 15.54 million tonnes/year. These results should alert policymakers to the potential impacts of the growing water demand and provide further support for water conservation action to reduce demand.


Highlights
• Kuwait's household water demand forecast under the current situation is an indicator to inform policymakers of the urgent application of water conservation measures.
• A behavioural spatial microsimulation model has been applied to forecast household water demand at a disaggregate household-level.
• Forecast output has shown that water demand will increase by 45% by 2050, to 664.1 million cubic metres (MCM), relative to the 2019 base year.

| INTRODUCTION
Kuwait is located in the northeast of the Gulf Cooperation Council (GCC) region, one of the driest regions in the world, characterized by an extremely poor endowment of freshwater resources, low precipitation and high temperatures and evaporation (Al-Zubari et al., 2017).
There is therefore a very high soil moisture deficit, so only a small percentage of rainfall infiltrates into aquifers.There is no surface source of usable water, such as rivers or lakes (Mukhopadhyay & Akaber, 2018).Kuwait is amongst the most highly water-stressed countries in the world, with the lowest availability of renewable freshwater per capita (World Bank, 2005).The renewable resource is below 2 m 3 per capita per year, which judged by the Falkenmark water stress index places Kuwait in 'absolute water scarcity', making it one of the least water secure countries (Rijsberman, 2006).
The severity of freshwater shortage began at the turn of the 1970s when Kuwait oil wealth led to exceptional economic and social transformation.Population grew rapidly, at 4.2% a year (Al-Zubari, 2002;Dawoud, 2005), accompanied by acceleration in agricultural development, industrialization and urbanization and changing consumer lifestyles (Abderrahman, 2000;Kotilaine, 2010).This transition led to a substantial increase in water demand, met initially by significant exploitation of groundwater aquifers, the only natural source of freshwater in the region (Dawoud, 2017).Population growth and a declining groundwater resulted in a dramatic drop in renewable freshwater per capita (Figure 1), which led Kuwait to construct desalination plants and draw down fossil groundwater (El Sayed & Ayoub, 2014;Saif, Mezher, & Arafat, 2014).Unconstrained by supply, water demand increased sharply.Whilst some conservation measures were introduced to protect the strategic fossil aquifers, demand and water stress continued to increase, in an unsustainable manner.This growing water stress led the state to invest a share of its oil wealth in desalination (Al-Hashemi et al., 2014;Shomar & Hawari, 2017), which overcame the immediate water scarcity (Dawoud, 2012;Mohamed, 2009).Today, desalination meets more than 90% of household and industrial needs (Aliewi and Alayyadhi, 2018;UN, 2019), such that Kuwait ranks sixth in the world for desalination capacity, amounting to 4% of global daily desalination (ADNEC, 2018;ESCWA, 2006).Growth in Kuwait's desalination capacity has followed growth in water demand and population.The state increased desalination capacity from 30 million cubic metres (MCM) in 1970 to 716 MCM in 2016, an increase of 2287%; over the same period population increased 498% (Al-Humoud & Al-Ghusain, 2003;MEW, 2017).This supply-side policy led to a rapid increase in per capita consumption (PCC).PCC in the household sector is the highest in the world and far above that of most other countries (Qureshi, 2020).Average household PCC is about 500 L per day (L/d) (Al-Ansari, 2013;Al-Zubari et al., 2017).In comparison, average household PCC in Germany and France is about 120 L/d, 156 L/d in England, 128 L/d in the EU and 310 L/d in the United States (Abu-Bakar, Williams, & Hallett, 2023;DiCarlo & Berglund, 2022;McCarton, O'Hogain, & Nasr, 2022;Parmigiani, 2015).
Urban land use in Kuwait accounts for 2.9% of the country by area, of which 2.5% is residential and commercial (Figure 2).The residential/ household sector has the highest water demand share (>60% of all demand) compared with other sectors and has the highest demand increase, of about 4.1% per annum (MEW, 2018).The dramatic increase in the household sector is attributed to (i) population growth and urbanization; (ii) tariff structures that do not cover water production costs and encourage wasteful consumption; and (iii) lifestyle changes, such as a greater focus on personal hygiene (e.g., more frequent showering).
Under the Kuwait government classification, the commercial sector also includes some residential units (largely high-rise buildings), so commercial areas also includes some household water demand that has been added in forecasting household demand.Household demand is expected to grow further as the government has a National Masterplan 'New Kuwait' that includes 13 new residential areas.These developments will comprise 1086 km 2 , which constitutes 6.1% of the total area of the country, more than twice the current residential area.These developments are due for completion by 2045 and will add further pressure on water resources.
Dependence on desalination to satisfy the increasing demand has adverse environmental and economic impacts.Routine discharge of hypersaline effluent harms the marine ecosystem (Jones et al., 2019;Von Medeazza, 2005) by raising sea water salinity 5-10 parts per thousand (Lattemann & Höpner, 2008), raising temperature 7-8 C and lowering dissolved oxygen (Mohamed, 2009), and by polluting with chlorine and un-ionized ammonia.Kuwait's Desalination has to date been fuelled by hydrocarbons and so contributes greenhouse gases (Al-Hashemi et al., 2014).Kuwait is ranked amongst the world's 14 worst countries in terms of carbon footprint with per capita emissions much higher than EU (Reiche, 2010) and OECD countries (Doukas et al., 2006).CO 2 emission are 25.2 t per capita/year, compared with 9.5 t per capita/year for the OECD and 8.9 t per capita/year for the EU (OECD, 2021).Generous subsidies have given rise to a huge gap between water system revenues and operating expenses in Kuwait (El Sayed & Ayoub, 2014).The net revenue covers only 6% of total production costs, far below full cost recovery.If this situation continues under current demand trajectories, subsidies will place an even heavier burden on the fiscal budget (Darwish, Al-Najem, & Lior, 2009).Kuwait already uses around 12% of its oil production to fuel desalination plants, a share predicted to rise to a staggering 50% by 2050 (Al-Rashed & Akber, 2015), which, in turn, will have a high opportunity cost.This strategy of turning nonrenewable fuel into water is unsustainable.Despite these problems, Kuwait is expected to invest further in desalination plant construction and expansion to meet growing water demand.
Currently, Kuwait has the lowest per capita freshwater availability in the GCC (and indeed in the world) yet its PCC is amongst the highest in the GCC (and world).Within the GCC, it has the highest share

| METHODS
In this research, several methods (Table 1) have been used to develop the Business-As-Usual (BAU) forecast for household water demand to  and 2.3).
To develop the disaggregated PHC demands, a process was applied to the PHC demand matrix (see Table 1) to impute PHC missing values (for dwelling type/household size classes), using non-linear power functions (see Section 3.1).

| Household baseline demand estimate
The imputed PHC demand matrix must be scaled to the national population using the household population matrix to give total national household water demand.This process is hindered by a lack of dwelling type data in the Kuwait national population census, preventing simple estimation of household size by dwelling type.This problem was addressed using a synthetic population microsimulation that constructs an artificial population with a distribution of characteristics that matches that in the observed population and that enables estimation of variable combinations that do not exist in the census data (Hermes & Poulsen, 2012).In effect, a good synthetic population simulation can reproduce the characteristics of a population allowing extension from a sample to an entire population, thus overcoming data limitations (Whitworth et al., 2017).It is common, for confidentiality reasons, for a census to have fewer details (variables) for individuals and/or households compared with sample surveys, but microsimulation allows these missing data to be imputed at the population level from the more detailed sample (Smith, Clarke, & Harland, 2009;Whitworth et al., 2017).
A Small Area Estimation (SAE) approach using a static spatial microsimulation method was used, considered amongst the most reliable of SAE methods (Ballas, Clarke, & Turton, 2003;Hynes et al., 2009;Tanton, 2014).The static spatial microsimulation linked the household population matrix (census 'macro data') with the PHC demand matrix (sample population 'microdata') through shared benchmarks to produce a synthetic population using Flexible Modelling Framework (FMF) software (Harland, 2013).The PHC demand matrix was then applied to the resulting synthesized population to estimate total national household water demand for the years 2013-2018 2 for which observed aggregate demand is known, allowing the BAU model to be validated.

| Forecasting BAU demand
The BAU forecast demand is driven by demography; hence, a population projection to 2050 for Kuwait is required.We used the

A. Observed Kuwaiti population 2018; B. TED/KISR projection of Kuwaiti population 2018
Developed a constant coefficient to be applied to the period of projection.
Procedure II: define the domestic worker proportions in Kuwaiti population To maintain consistency of Kuwaiti population projection compared with observed population.The domestic worker proportions will be added to the Kuwaiti projection.
Three constant coefficients were developed from the historical trends: (i) average of 2007-2018; (ii) average of 2014-2018, as a relative demographic change occurred; (iii) proportion of 2018 as it is the recent observed year.The developed coefficients were applied to the non-Kuwaiti population, then the obtained number will be subtracted from this population and added to the Kuwaiti population.

Procedure III: define non-Kuwaiti household population
To find the non-Kuwaiti population in the household sector, as the research targeted those inhabitants in the relevant sector.

Observed non-Kuwaiti population (2007-2018).
Three constant coefficients were developed from the historical trend: (i) average of 2007-2018; (ii) average of 2014-2018, as a relative demographic change occurred; (iii) proportion of 2018 as it is the recent observed year; The developed coefficients were applied to the total non-Kuwaiti population, then the obtained number will be representative to non-Kuwaiti in the household sector.
Procedure IV: determine population distribution by household size and the country's governorates (i) To set PHC demand for each household.
(ii) To allow spatial analysis amongst the country's governorates.
Percentile distribution of observed household size distribution was projected onto the forecast population after satisfying previous procedures; this method is derived from the scaling-up method.

Procedure V: define households by dwelling type
To specify a household's dwelling type, as dwelling type is a determinant of PHC demand.
A    3) with the influential outliers affecting the trendline and biassing the missing value assumption.Uncertainty exists as to whether these outliers occur because of measurement error in the water demand survey or if they represent actual variability in observed demands.However, due to the high influence of the outlier points in the slope and assumption process, these points were omitted and replaced by values derived by applying a regression fitting function (also used to derive any missing values-e.g., where there is no PHC for a given dwelling type/household size).

| Baseline demand estimate
To establish a base period that reflects the 'current' situation of water demand in Kuwait, water use coefficients from the PHC demand matrix are used to represent the primary driver of water demand and linked to associated population data6 ; this is represented by where BLDE refers to the baseline demand estimate; cgj is the Kuwaiti  The difference between Low-BB and the base scenario is 14.3 MCM in 2020, and between High-AA and the base scenarios is 18.5 MCM.
The household size distribution (scale parameter) has a major influence on forecast total demand but is a minor influence in terms of the demand proportion distribution for Kuwaiti and non-Kuwaiti.When applied to high, medium and low scales, considerable variation in total demand (Kuwaiti and non-Kuwaiti) is evident, but little variation arises due to the demand proportions between Kuwaiti and non-Kuwaiti.
Furthermore, the domestic workers' parameter has more influence on demand than the non-Kuwaiti parameter.
These costs are in common use in Kuwait desalination industry, and no more recent data are currently available.The aggregate water cost can be calculated from where APC is the aggregate water cost at time t; p the price per unit 7 (m 3 ); and q the quantity of water billed and collected at time t.For water revenue (benefits), a change in q represents the cost costumers pay per unit 8 (0.58 US$).To calculate the PHC cost/revenue per unit, the following has been used: where PCRPHC is the cost/revenue per PHC at time t; h is the total household population (or Kuwaiti/non-Kuwaiti households); and p is the total cost/revenue at time t.The price per unit comprises the production and delivery costs. 8 The tariff structure in Kuwait is uniform volumetric.
F I G U R E 8 Difference between Kuwaiti and non-Kuwaiti demand.
thermal desalination 9 of 23.4 kg/m 3 of water produced has been drawn from literature (Darwish, Al-Najem, & Lior, 2009;Dawoud, 2012;Fath, Sadik, & Mezher, 2013;Raluy, Serra, & Uche, 2004).CO 2 emission from desalination was estimated from where CE t is the carbon dioxide emission at time t; d t refers to water produced (applied to the grand total and also total Kuwaiti and non-Kuwaiti demands); and e t is carbon dioxide emission per unit production at time t.The PHC's CO 2 footprint was calculated from where HCF t is the household carbon dioxide footprint at time t; p is the household population (grand aggregate and aggregate Kuwaiti and non-Kuwaiti); and e is the CO 2 emission to equivalent household population (e.g., Kuwaiti) at time t.

F
I G U R E 1 Population and renewable freshwater per capita in Kuwait since 1970.Source: World Bank (2005); Al-Zubari et al. (2017).

F
I G U R E 2 Urban land use in Kuwait.T A B L E 1 Methods for forecasting future household demand used in the study.
of demand from the household sector (>60%), yet despite strong annual growth in demand (>4.1%) water demand in Kuwait remains little studied.Against this background, this study assesses future household water demand in Kuwait under current demographic trajectories and then assesses the environmental and economic consequences of this changing demand.By understanding demand futures under current development trends, we hope to support officials and policymakers in Kuwait in taking appropriate conservation actions to avert a future water crisis.Our findings are of relevance to other countries in the GCC region, which share a similar range of scarcity and growth pressures.
2050.A model building process (Figure3) is used in which demographic forecasts are applied in conjunction with household water use coefficients, with both disaggregated to reflect household characteristics, and techniques used to overcome missing data.The main steps, detailed further below, are (i) demographic disaggregation of per household consumption (PHC) with trendline fitting imputation to complete the set of PHC coefficients by household type/size;(ii) generation of aggregate water demand for the base period(2013-  2018); and (iii) projection of the household BAU demand forecast to 2050.2.1 | Disaggregate PHC trendline imputationNo forecast of water demand in the household sector exists for Kuwait; there are only forecasts for the entire population for all country's sectors.We produced a forecast using spatial microsimulation(Birkin & Clarke, 2011), a method not previously used in water sector in the GCC, that allows observed aggregate PHC demand to be disaggregated by household characteristics (and which in later work enabled scenario modelling with representation of micro-components).This decomposed demand was constrained by an official population projection to generate the BAU household forecast.To develop the baseline demand estimate, two datasets were used: first, a survey of household water demand conducted in 2013 by the Central Statistical Bureau (CSB).The survey is based on probability sampling-a combination of cluster and stratified techniques, involving 2961 households comprising Kuwaiti and non-Kuwaiti households (the household status), household size and dwelling type for each household distributed over the country's governorates, with demand recorded as PHC (hereafter the 'PHC demand matrix').The dwelling type is recorded as a villa, floor or apartment in a villa, an apartment, a traditional house or an annex. 1 The second dataset is the household census (2007-2018), provided by the Public Authority and Civil Information (PACI), and F I G U R E 3 Overview of the modelling framework developed for the study.referred to hereafter as the 'household population matrix'.This matrix includes household status and household size distributed over the country's governorates but lacks the dwelling type variable of the water demand survey, which is an important influence on demand.Both matrices identify non-Kuwaiti inhabitants (domestic workers), which are a large part of the population and play a major role in Kuwaiti households' demand.These demographic variables are further discussed below (Sections 2.2 projection of the Techno-Economic Division (TED) of the Kuwait Institute for Scientific Research (KISR) in preference to that from the United Nations (UN), 3 as the latter underestimates population with respect to observed data.The TED/KISR projection employs a cohort component method that ages the population with annual representation of births, deaths and migration.This is considered preferable to other forecasts for Kuwait that use mathematical (arithmetic, geometric, logistic) trend projection as these are less successful at representing underlying demographic processes(Gawatre, Kandgule, & Kharat, 2016).Furthermore, the TED/KISR projection builds on the same PACI data structure as the 2013-2018 observed base period described above and differentiates national (Kuwaiti) from expatriate populations.This helps in forecasting as these groups differ very significantly in their household water demand (as evidenced by the CSB, 2013 4 water use survey).The TED/KISR projection was subject to further work (Table2) to closely fit the projection to the requirements of the demand forecasting.Because TED/KISR produced only a single-variant projection, a deterministic sensitivity test was applied using several population coefficients based on historical observations.Sensitivity of forecast demand to changes in input parameters, in the range of ±15% (a value suggested byBillings & Jones, 2011), was assessed to identify the relative influence of input variables on demand.In this way, the three most important independent variables for sensitivity testing were identified as domestic worker share, non-Kuwaiti population share and household size variability, which we define here as 'household size distribution'.These variables were manipulated and combined to produce 12 possible population variants, which fed into 12 BAU demand forecasts.For the domestic workers and non-Kuwaiti population share, parameter values were derived from historical observations (2007-2018) using a naïve 5 value (A) and a weighted moving average value (B).For domestic workers, parameters values are 60.08% (A) and 55.59% (B) reflecting the share of the Kuwaiti population that are domestic workers.Domestic workers are individuals employed by a householder and resident in a household and include maids, cooks and gardeners.We address them as a separate group in the model because the rate of growth of this population group, which now represents a large proportion of households, is considerably higher than for Kuwaitis.These domestic workers are assigned to the Kuwaiti household population in the study's demography matrices.The non-Kuwaiti household population share is similarly represented by the parameters (A) 48.64% and (B) 45.48%.In Kuwait, a significant share of the non-Kuwaiti population is not associated with2 The estimation starts with 2013 as the base year since the CSB survey is conducted in 2013.3TheUN projection also uses the cohort component method and is applied to provide multivariant population projections.Source: https://population.un.org/wpp/Graphs/.household water demand, as they work within the agricultural (or other commercial) sector, and their personal water use is thus recorded as an agricultural (or other commercial) demand.Thus, we need to understand the share of non-Kuwaiti population that are resident in households for which we are forecasting demand.The household size distribution parameter represents variability in the observed household size distribution, which has changed relatively quickly in Kuwait.The 'High' variant represents the observed household size distribution, averaged over an observed historic period(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018).In this period, the earlier years have a higher proportion of larger household sizes (>7 persons), and conversely, more recent years have a higher share of smaller household sizes (<7 persons).In averaging over the historical period, more weight is given to larger household sizes than is consistent with current, smaller household sizes, which will act to elevate forecast water demand (hence the 'high' variant label).A 'Medium' variant represents the percentile distribution of household sizes where a weighted moving average technique has been used.In this variant, the more recent years (smaller household size proportions) have more weight; that is, more recent years are considered more representative than the earlier ones.A 'Low' variant represents the household size distribution of 2018, the latest year for which observed household size distribution data are available.This variant gives the most weight to the smaller household size distribution, thus implies a lower household water demand.T A B L E 2 Enhancement of TED/KISR population forecast to support water demand forecasting.Objective/reason Dataset Method Procedure I: Kuwaiti population uplift (bias correction) To overcome underestimation of the TED/KISR projection compared with the observed population (2013-2018).

F
I G U R E 4 Function fitting to derive missing values for the Kuwaiti household dwelling categories.F I G U R E 5 Function fitting to derive missing values for the non-Kuwaiti household dwelling categories.

3
| RESULTS AND DISCUSSION 3.1 | Missing PHC values imputation Missing values in the PHC demand matrix were assumed by fitting power functions to the observed data (Figures 4 and 5).This process uses PCC values in preference to PHC values as they display a better fit; household size is then used to derive the missing PHC values.Assumed values in the Kuwaiti household dwelling categories fit well (r 2 > 0.90) in most categories, whereas non-Kuwaiti household categories have a fair fit in most categories.The imputation of missing values was satisfactory after application of the influential and leverage points detection tests in the PHC demand matrix.Tests to detect influential and leverage points are used in trendline curve fitting where different trendline functions are applied (over several iterations) with and without extreme observations (outliers) to see how these observations affect the trendline orientation.By applying this test, influential and leverage outlier points in the PHC matrix were detected, and the best fit trendline function (power) selected.Outlier points affect the orientation of the slope and drag the trendline towards its location in whatever trendline is being performed.Five outlier points were detected across the PHC matrix (Table

3. 4 |
Wider economic and environmental appraisal3.4.1 | Economic impact assessmentTo assess the economic implications of the forecast BAU demand, water cost is calculated as production cost per m 3 (2 USD), plus the cost of delivery to end-user (1 USD), giving 3 USD per m 3 of end-use consumption(Al-Damkhi, Abdul-Wahab, & Al-Nafisi, 2009;

F
I G U R E 1 0 CO 2 emissions under Business-As-Usual (BAU) scenario.F I G U R 9 The cost of production and revenue in Business-As-Usual (BAU) scenario.billion USD in 2050, which represents a very high opportunity cost if it uses around 50% of its daily oil production.This situation would lead to an economic dilemma as Kuwait's economy is a rentier economy (oil production represents over 90% of the government's income).This strategy of turning non-renewable fuel into water, with the added environmental impacts of carbon emission and marine discharge from desalination, is wholly unsustainable.Perversely, whilst the GCC countries have the lowest freshwater resource per capita in the world, desalination coupled with very low rates of cost recovery has led to the highest per capita water use in the world (typically 3-4 times those in much wetter Europe).Despite this immense challenge, there are clearly opportunities to transition to a more sustainable water future, through a range of intervention measures.First is the need to introduce full cost water pricing, with attention given to development of appropriate social tariffs(Barberán & Arbués, 2009).Second, whilst all households are currently fitted with 'dumb' (mechanical) water meters, introducing smart metering technology should increase consumer's awareness and in conjunction with pricing measures lead to behaviour change and water conservation(Koech, Cardell-Oliver, & Syme, 2021).Third is consumer education on water, with awareness raising campaigns to encourage consumers to save water through behaviour change and technology measures (e.g., upgrading water appliances to more efficient ones).Collectively, implementation of such conservation measures in less water stressed regions has shown demand reductions of 40-60% are possible(Abu-Bakar, Williams, & Hallett, 2021;Syme, Nancarrow, & Seligman, 2000).Such savings should then be readily achievable in Kuwait, but the government needs to grow the water conservation industry, so households can access appropriate guidance, buy water efficient appliances and fittings and have access to suitably qualified installers and advisers.Finally, whilst Kuwait's estimated network distribution leakage remains below that observed in many European countries, losses in the range of 7-15%(Akber & Mukhopadhyay, 2021) remain significant; hence, a leakage reduction programme is important.This can be further supported through phasing in of smart water metering.These conservation measures need to be introduced as part of a wider water conservation strategy, recognizing their mutually reinforcing nature.Historically, such measures have been ignored or at best marginalized in Kuwait's water management, where like the wider GCC region, the transition to a sustainable water future lies in embracing demand side management.
demand (PHCl/d) by spatial distribution g and dwelling type j for each household size h on a daily basis d; egj is the non-Kuwaiti demand (PHCl/d) by spatial distribution g and dwelling type j, divided by a billion (litres per MCM), then, multiplied by days per year t (e.g., 2015) to get the aggregate demand for year t.Equation 1 was applied for each year in the 2013-2018 baseline demand estimate period, where the results show an expected increase in demand with household population increase.Figure6shows the observed aggregate household demand (2014-2017) and estimated aggregate household demand against observed household population.This base period estimation was compared with observed demands using the mean absolute percentage error (MAPE) statistic to validate the estimation.The MAPE of 2.36% indicates a good a model estimate; hence, we take this baseline period model forward into the demand projection BAU to 2050.

6
Estimated household water demand.Kuwait water demand forecasts under different population projections (MCM/year).