Estimating domestic self-supply groundwater use in urban continental Africa

Urban population of continental Africa was taken from three sources: (i) the 2015 Gridded Population of the World version 4 with a 1 km spatial resolution (CIESIN (Center for International Earth Science Information Network) 2017),(ii) the 2015 population from the Global Human Settlement Population layer (GHS-POP) with a resolution of 250 m, and (iii) the 2015 gridded population data from the High Resolution Settlement Layer (HRSL) with a resolution of 30 m (Facebook Connectivity Lab and Center for International Earth Science Information Network—CIESIN—Columbia University 2016). More details are in (table S1 available online at stacks.iop.org/ERL/15/1040b2/mmedia of the supplementary information (SI). Definitions on urban population densities differ—we considered a gridded area urban when its population density was > 300 inhabitants/km². The total urban population for continental Africa obtained in this way was 470 million, which closely corresponded to the 2015 estimates from UN-DESA and UN-HABITAT (468 and 471 million, respectively).

Groundwater potential was taken from GIS layers of the Groundwater Maps of Africa (Macdonald et al 2012). We used layers groundwater storage, depth to groundwater and aquifer productivity. The data covers continental Africa (without islands) with a spatial resolution of 5 km.

Surface waters were obtained from OpenStreetMap (OSM; downloaded from http://download.geofabrik.de/in June 2019 and January 2020), using all features (values) available in the categories (keys) waterways and water bodies.

Socioeconomic status as a spatially available parameter does not exist, so we developed a proxy, whereby we assumed that low socioeconomic areas depended on springs and hand-dug wells, which are controlled by depth to groundwater. In addition, we assumed that middle and high socioeconomic status areas can also access private boreholes and that their access is controlled by aquifer productivity (> 1 l s⁻¹; see figure 1). To develop the proxy, spatial distributions and concentrations of luxury amenities and green areas, available in OSM, were determined in QGIS, using Kernel density functions (QGIS Development Team (2020) QGIS User Guide Release 3.4 2020). Luxury amenities include hotels, banks, ATMs, swimming pools, cinemas, jewelleries, shopping malls and museums. Green areas such as parks and golf courses were also included, as the distribution of green spaces in cities is influenced by the socioeconomic status of neighbourhoods (Rigolon et al 2018). The resulting density map was then classified according to income distribution, creating a distribution of high, middle and low socioeconomic status, guided by the World Bank generated indicator 'Income share held by subgroups of population indicated by deciles or quintiles' (World Bank 2018). An example of spatial distribution of luxury amenities and green areas as well as a map of the spatial socioeconomic status of Kampala (figure S1) and guiding text (chapter S2.1) is provided in the SI.

Zoom In Zoom Out Reset image size

Spatial information on access to a public water supply network does not exist, so we developed another proxy, whereby we assumed, that new and/or more distant (from the city centre) urban developments will be less likely to have public water infrastructure, and therefore more likely depend on self-supplied groundwater. The steps to arrive at this proxy, including a detailed example (Kampala) are described in chapter S2.2 (SI). Briefly, based on data for network supply coverage and 2015 urban population data, we estimated the number of inhabitants served by a networked water supply. Secondly, we determined the areal extent of the city as a function of time from 1995–2015 with a 5 to 10-year frequency. Thirdly, the time-dependent urban boundaries of each city were overlain on the 2015 Gridded Population map (see above).

We then determined the 2015 urban population within each of those boundaries and which best compared to the 2015 inhabitants covered by the public water supply. The area inside the best fitting urban extent boundary was assumed to have access to a piped network and the area outside to have a poor public network coverage and/or reliability. The latter was named lag-area of the public water supply. Our method assumes that development of the public supply network follows urban sprawl and outward growth and only depends on time as the factor—though this is certainly not the whole picture. Network development (or lack of thereof) also depends on socioeconomic status (e.g. new affluent areas may be connected sooner than new low-income areas and slums) or on hydraulic characteristics, like elevation, or presence of natural hydraulic barriers (mountains or ridges). In addition, water demand management and water resources constraints are increasingly setting limits for further expansion of piped distribution. However, for simplicity, we did not include these parameters in the determination of the lag-area.

${G_M}$ is the maximum urban population to use groundwater obtained via self-supply in an area across the entire African continent, or, the urban population potentially using groundwater via self-supply. ${G_M}$ [inhabitants/km²] was determined by: for each km².

$$\begin{equation}{G_M} = {F_M}U.\end{equation} \tag{ 1 }$$

${F_M}$ indicates the maximum proportion of the population that could access groundwater via self-supply. It has a value between 0 and 0.95, based on a conditional algorithm depicted in figure 1 (fourth column from the left). The algorithm considers the parameters [groundwater storage], [depth to groundwater], and [aquifer productivity]. ${F_M}$ was given a value dependent on the combination of these parameters.

In order to reduce computation time, we reduced the range of ${F_M}$ values into four discrete values: 0, 33%, 67%, and 100% of the maximum range value. Together, these values define the maximum chance of using groundwater obtained via self-supply (Tables S5 and S6 (SI)). $U$ is the urban population per km² (Gridded Population of the World version 4). There are different ways to determine the extent of a city or urban area (e.g. areal, administrative boundaries). We chose to define the urban extent as a cluster of contiguous 1 km² 'pixels' in the Gridded Population of the World version 4 closest to the location of the city marker available in OSM with a population > 300 inhabitants/km². In case of a particular city or one urban area, ${G_M}$ was expressed as a percentage of the total population of that city by dividing by its total population, which we named ${G_{M\% }}$. We determined ${G_M}$ for the entire African continent and ${G_{M\% }}$ for a selected number (10; see below) of cities.

${G_L}$ [inhabitants/km²] was defined as the likely UUGS in an area across the entire African continent. ${G_L}$ was determined by:

$$\begin{equation}{G_L} = {F_L}U\end{equation} \tag{ 2 }$$

where ${F_L}$ indicates the ratio or proportion of the population that are likely to use groundwater via self-supply. In this case ${F_L}$ was determined by the parameters [groundwater storage], [depth to groundwater], [aquifer productivity factor], [proximity to surface water], [socio-economic status], and [lag-area of the public water supply] (figure 1). In addition, in the algorithm, we included the presence of imposed groundwater use restriction measures. This is to reflect the fact that in some cities in Africa, households are neither allowed to drill nor to abstract groundwater (Kampala), or there are prohibitively high customs duty on imports of drilling equipment (Addis Ababa) rendering drilling very expensive.

We used three sets of ranges of ${F_L}$: from 0–0.35 indicating a low likelihood, from 0–0.65 indicating an average likelihood, and from 0–0.95 (identical to ${F_M}$) indicating a high likelihood of using groundwater through self-supply. Each range was made discrete by division into four values at 0, 33%, 67% and 100% of the maximum range value (see Tables S7-S9 (SI)).

When multiplied with $U$, these three sets of ${F_L}$ values yielded ${G_L}$ as arrays of population numbers per km². In case of Addis Ababa and Kinshasa, we used GHS-POP, while HRSL was used for the other cities. Also in this case, ${G_L}$ can be expressed as a percentage of the total population of a city, which we then named ${G_{L\% }}$. Because not all data is available for each city in Africa, we determined ${G_{L\% }}$ for cities with comprehensive data on groundwater use (see below). This enabled an assessment of the methodology to calculate ${G_{L\% }}$ (see below).

The difference between ${G_M}$ and ${G_L}$ is the number of input parameters used and the number of sets of values of the parameter $F$ (either ${F_M}$ or ${F_L}$). In the case of ${G_M}$, only hydro(geo)logical parameters were used, and therefore we consider ${G_M}$ the maximum or potential UUGS. In addition, one set of ${F_M}$ values ranging from 0–0.95 was used. For ${G_L}$, in addition to hydro(geo)logical parameters, the lag-area presence of a public supply network, socio-economic status, and—if applicable—imposed groundwater abstraction restrictions were taken into account. Furthermore, three sets of ${F_L}$ values were used. Therefore, we consider ${G_L}$ to more accurately indicate the urban population that actually depends on groundwater for one or more uses than ${G_M}$. We furthermore applied the following set of rules: 1) Only in case of groundwater depths > 25 m, the parameter aquifer productivity (< or > 1 l s⁻¹) was applied. We assumed low productive aquifers would still be tapped at shallow depth, 2) Proximity to surface water only affected the low economic status. We assumed middle and high economic status did not use surface water, 3) Restriction measures applied to middle and high economic status, but not to low economic status, and finally, 4) If groundwater was available it would be used, dependent on low, average or high likelihood conditions, except in case of restriction measures. Because the F parameter ranged between 0 and at most 0.95 (high likelihood conditions) instead of 1.0, this final assumption did not imply that in case of groundwater availability, the population was limited to using only groundwater. Still, a fraction $1 - F$ of urban inhabitants could use other sources, although we did not specifically mention them (e.g. rainwater harvesting). Fig. S4 (SI) gives a worked example of the algorithm.

Comparison of the algorithm used (figure 1) with survey data was carried out only for ${G_L}$, since it was considered to be accurately calculating the UUGS. In order to validate our results, we compared ${G_L}$ with the survey data from the 'Performance Monitoring and Accountability 2020 project' (PMA2020; Boyle et al 2019) for seven cities in SSA. PMA2020 uses innovative mobile technology to support low-cost, rapid-turnaround surveys monitoring key health and development indicators. Data are collected at household and health facility levels via mobile phones through a network of female data collectors. It uses a two-stage cluster design with residential area (urban vs. rural) and sub-regions as strata, and standardized questionnaires to gather data about households that are comparable across program countries and consistent with existing national or regional surveys.

To our knowledge, the PMA2020 is one of the few survey instruments, which consistently collects information on the use and type of secondary water source for specific cities and makes the data readily available. We decided not to compare our results with the data from regional surveys, the basis for the earlier mentioned JMP. The JMP team has recommended that the 2018 version of the Multiple Indicator Cluster Surveys (https://washdata.org/unicef-multiple-indicator-cluster-surveys-mics) ask about the main source of drinking water used and the source of water used for other purposes, so we expect that more data sets on secondary source water will become available in the future. Furthermore, for Arusha we made use of fieldwork surveys carried out within our group (Komakech and de Bont 2018).

For the two proxies we developed we carried out a (modest) sensitivity analysis for the case of Kampala. For the proxy socioeconomic status, we changed the deciles that define the boundaries of high, middle and low socioeconomic status. For the proxy lag area of the public supply, we changed the (time dependent) urban boundary, thereby changing the areas served by the public supply and not served.

Given the distribution of hydrogeological conditions and our definition of urban (>300 inhabitants/km²), we estimated that 369 million people living in urban areas in mainland Africa could potentially use groundwater to meet their domestic needs (figure 2). This represents 79% of the total urban population of mainland Africa (470 million). The model shows that, because the resource is present, there is considerable potential to use groundwater in urban parts of Nigeria; around the lakes in Eastern Africa (e.g. Kenya, Uganda and Rwanda);in the Nile delta and around the Nile valley in Egypt; in Ethiopia, more scattered, in northern-eastern South Africa and Zimbabwe; and along the northern African coast, mainly in Algeria and Morocco.

Zoom In Zoom Out Reset image size

Based on availability of survey data, we selected seven cities for which we determined ${G_{M\% }}$- and ${G_{L\% }}$ values (top 7 rows of table 1). The maps (figures S5-S25 in the SI) give a spatially distributed overview of input data per city. ${G_{M\% }}$-values ranged between 52 and 90%, and ${G_{L\% }}$ values (${F_L}$ set) ranged between 5 and 76%. The average of the ${G_{M\% }}$-values in table 1 was 79% and the average of the ${G_{L\% }}$-values (${F_L}$ set) 31%, which indicated a difference of some 48 percentage points between the likely and the maximum population using groundwater obtained via self-supply. The differences between ${G_{M\% }}$ and ${G_{L\% }}$ were primarily due to the presence of a public supply network and a population with an average to high socio-economic status, who were—for the better part—assumed to be able to pay for a public supply connection. In the case of Kampala and Addis Ababa, the difference between ${G_{M\% }}$-and ${G_{L\% }}$ was high (50% or more) due to a combination of favourable hydrogeological conditions for groundwater use (${G_{M\% }}$ is high: 75%–90%), while a piped network is covering a large part of the cities with household and yard connections, public standpipes and kiosks. Thus, there is less need to practice self-supply. As mentioned, in Kampala there are also restrictions to drill and in Addis Ababa, drilling equipment import restrictions make drilling expensive (pers. comm. of one of the authors with the Ministry of Water, Irrigation, and Energy, and with the Addis Ababa Water and Sewerage Authority).

Based on our time-dependent analysis of the urban population expansion, we observed that all cities in table 1 were rapidly increasing in size. When networked supply development lags behind, they develop important self-supply practices involving groundwater, if available. Per city in table 1, the difference between minimum and maximum ${G_{L\% }}$ ranged from 17 percentage points in the case of Kampala (25%–8% = 17%) to 51 percentage points in the case of Lagos. These higher differences of 51 percentage points, 49 percentage points (Kano), 47 percentage points (Arusha) we primarily attributed to favourable hydrogeological conditions (groundwater storage, productivity, and depth to groundwater) on the one hand and low network supply coverages on the other. This enables self-supply. n contrast, low differences of 17 percentage points (Kampala) and 18 percentage points (Addis Ababa) were mainly caused by the above described high public coverages on the one hand and restrictive measures on the other hand.

Survey data indicate that 3%–83% of the urban population use groundwater obtained via self-supply in the seven selected cities (table 1). When comparing ${G_{L\% }}$-values with the PMA survey data, we observed that for Arusha, Kampala, Kano, and Lagos, ${G_{L\% }}$-values were closest to the PMA survey data in case ${F_{L\_max{\text{ }}}}$ values were used. For Addis Ababa and Nairobi ${G_{L\% }}$-values were closest to the PMA survey data if ${F_{L\_min}}$ values were used and for Kinshasa in case ${F_L}$ values were used. From this we concluded that with the three sets of ${F_L}$ values we used the bandwidth of likely UUGS in urban areas in continental Africa could be estimated. However, it was not possible to predict a more accurate 'likely UUGS' than a range of values, as presented in table 1. As an example, we calculated ${G_{L\% }}$ for 3 cities without PMA survey data: Maputo, Dodoma, and Lusaka (lower 3 rows of table 1). So, in Maputo, the bandwidth of likely UUGS was 8%–32%, in Dodoma 21%–65%, and in Lusaka 16%–53%.

The difference in ${G_{L\% }}$ when changing socioeconomic status, which was changed by adjusting decile boundaries defining wealth, ranged from 1%–9% compared to the modelled values (table S10 (SI)). When changing the lag area, the difference in ${G_{L\% }}$ ranged from 1%–3% compared to the modelled values (table S10). For the cases calculated, our geospatial model was more sensitive to changes in the socioeconomic distribution proxy than in the lag area. We attributed this to the greater dependence of low income areas on groundwater on the one hand and because low socioeconomic status pixels were more abundant in the more recent parts of the city, not affected by the lag area changes. From this modest sensitivity test, we concluded that changes in the two proxies we developed had effect on ${G_{L\% }}$, but this effect was limited, and our geospatial model appeared to be fairly robust.