Climatic trends
Precipitation (seven meteorological stations) and evaporation (five meteorological stations) data available for the past 44 years did not reveal any statistically significant trend (Fig. 4a,c). Linear regression analysis with SPSS v26 (Table 4) showed insignificant change in annual rainfall (R2 of 0.071) and evaporation (R2 of 0.224), yielding values of R2 of 0.071 and 0.224, respectively. This means that only 1 and 22% of the observed variance in precipitation and evaporation respectively, can be predicted from a linear trend. In contrast, temperature time series from six meteorological stations showed a more significant increasing trend of ~0.3 °C/decade over the 34 years of records available (Fig. 4b). The R2 of 0.558 (Table 3) indicated that 56% of the observed variance in temperature can be predicted from a linear trend. This analysis implies high uncertainty in the climate trends in the region, and as a result, existing data do not allow for confidently estimating a possible influence on the long-term evolution of groundwater recharge.
Table 3 SPSS linear regression parameters for the analysed averaged annual climatic data at 95% confidence interval (R2 is the coefficient of determination demonstrating whether observed values are regressed on predictions and Vis versa, while R is the correlation between the predicted values and the observed values). SD standard deviation Land-use change
The classification of land-use over the NAS focused on assessing changes in the spatial extent of built-up areas (urbanisation), forests and other vegetation (grasslands, agricultural land, and shrubs) over the 2000–2017 period (Fig. 5a,b). This is further summarised in Fig. 5c, revealing: (1) a 70% increase of built up areas from 14.5% coverage in 2000 to 24.2% in 2017, (2) a 23% decrease in forest cover from 2000 at 20.3–15.7% in 2017, and (3) a small change in the combined coverage of grassland, agriculture and shrubs. The observed reduction of agricultural surfaces at the expense of constructed areas around the main urban centres mostly relate to coffee and tea farms and are likely to modify any hydroclimatic influences on the spatio-temporal patterns of groundwater recharge. The conversion from farmlands to housing also brought about more drilling of boreholes thereby expecting to be modifying the magnitude and spatial distribution of groundwater abstraction points.
Groundwater abstraction and borehole permit application trends
Early 2010 corresponded with the start of ongoing WRA efforts to obtain systematic recording of abstraction rates through progressive installation of a metering system (master meter readings). Master meters are installed at the source point (on-site metering) at every borehole (abstraction point) before distribution and any loss through leaky connections. This has yielded positive results with recorded abstraction in 2017 being close to estimates derived from approximations based on the borehole permit records available since the late 1970s (Fig. 6a), suggesting that the WRA abstraction record is nearing completion. The combination of both indicate rapidly increasing trends. Estimated groundwater abstraction (the most realistic estimate from borehole permit records and assumed averaged abstraction value of 20 m3/day per borehole) suggests an (almost perfect; R2 = 0.9985) increasing cubic trend in groundwater use throughout the 1977–2017 period, from about 14,500 m3/day in late 1970s to 150,000 m3/day in 2017; i.e. an increase by a factor 10 in 40 years. Within the most recent period 2008–2019, for which permit applications for new boreholes is available from the WRA, the number of applications are increasing overall (Fig. 6b) consistent with the observed increasing abstraction. The number of permit applications however show a clear correlation with climatic conditions (Fig. 6b), whereby dry years are associated with, or immediately followed by, an increased number of applications.
The NAS estimated abstraction is following a trend similar to the NCC population data as reported in the 2018 “Revision of the UN World Urbanization Prospects” (United Nations 2018; Fig. 7a). Both are highly correlated with a R2 of 0.996 (Fig. 7b). This confirms the key role of NAS groundwater in meeting the city’s population increasing water demand and suggests that groundwater demand may reach 300,000 m3/day by 2035. When combining the UN population data and projections for 1950–2035 with the population projection for 2100 (Hoornweg and Pope 2014), the population growth trend is almost perfectly cubic (R2 = 0.99996). This further suggests that, assuming that groundwater demand continues to follow the Nairobi demographic trend; demand will reach over 1.7 million m3/day by 2100. This correlation between NAS estimated abstraction and NCC demographics was used to project future groundwater abstraction in the groundwater model (scenario GWDS; section ‘Future scenarios’).
Groundwater level time series
The 10-year records of monthly groundwater monitoring in the WRA observation wells (Fig. 3b) are shown in Fig. 8. There are clear long-term decreasing trends for all eight wells that were not being substantially influenced by local high frequency pumping variations. Linear trend analysis showed high significance (R2 ranging 0.55–0.90) for wells the least affected by pumping schedules (Hillcrest, Kabansora, KICC, St Lawrence and Trufoods). The long-term, linearly decreasing, trend is a clear indication of regional depletion of the NAS. Groundwater levels from the monitored boreholes within NCC are shown to be declining at a rate ranging between 5–2 m/year. Across the monitored wells, the mean decline rate is 0.43 m/year (4.3 m/decade) since 2007. The decline had a slight recovery around 2010 for most stations due to reduced pumping activities of the wells as recorded from WRA well monitoring forms. The sharp punctual anomalies are the result of unstable recovery periods before water levels are captured as all the boreholes reported here are production wells but with agreement between the owners and WRA, borehole owners are expected to stop the pumping 24 h before monitoring. It is possible that this agreement is not being fully implemented; hence, the observed sharp variations in the observed trends.
Spatio-temporal water-table evolution
Piezometric maps for different years (1950, 1960, 1970, 1990 and 2017) showed that over time, piezometric contours tend to shift towards the opposite direction of the groundwater flow, i.e. westward (Fig. 9), which again indicates a general decline in groundwater levels. Contour maps for years 1990 and 2015 were well constrained as groundwater observation data are well distributed spatially. The effect of pumping is clearly observed in NCC, characterised by a general depression of the piezometric surface. It is evident that the potential lines in early years (1950, 1960, and 1970) are relatively smooth demonstrating the natural groundwater gradient of the region with groundwater flowing from west to east. The 1990 and 2015 potentiometric surface map however produced more distorted potential lines, resulting from more local drawdown due to borehole abstraction. More specifically, the distorted potential lines are observed within the boundary of NCC where borehole density is highest and increasing at an alarming rate.
Overall, throughout the whole observation period, interpolated maps showed a general piezometric inflexion highlighting convergent groundwater flow in the lower part of the study area where the main permanent rivers are situated, which confirms that groundwater is discharging into the surface-water drainage network in this part of the catchment. The relative coarse resolution of interpolated maps, however, does not allow for accurately depicting the interactions between surface and groundwater at the scale of individual streams and rivers.
Maps of change in water-table depth between consecutive dates and throughout the whole observation period show some localised areas with rising water levels likely indicating increased recharge; however, the declining water levels indicate an overwhelming general groundwater depletion (Fig. 10). This depletion is particularly clear, and ubiquitous, between 1990 and 2015, with a general decline of more than 50 m in places; the period during which abstraction has been multiplied by about 3.5 (from about 40,000 to 140,000 m3/day; Fig. 6). In addition, the average groundwater decline since 1950 below Nairobi City Council is summarised in Table 4.
Table 4 Quantification of mean groundwater level change (m) underneath the Nairobi City County (NCC) area indicating continuously falling levels (depletion). SD standard deviation Groundwater numerical modelling and future abstraction scenarios
Model calibration results
During trial-and-error model calibration, values of hydraulic conductivities were adjusted consecutively within the range of reported values for each geological unit (Fig. 11a,b). All the calibrated geological units’ hydraulic conductivity values, with the exception of the Athi sediments, fall within the 25–75% percentile range of data collated from borehole completion reports for the respective aquifer units, including for the most extensively tapped Kapiti formation for which the calibrated K is close to the median observed K. The calibrated K for the Athi sediments falls within the total observed range but near the upper end of it. This is due to the observed K values predominantly reflecting the lower, less permeable parts of the Athi sediments present in the Nairobi City area in which most boreholes tapping this unit are drilled (Fig. 2). Regarding storage properties, from an applied initial value of 4.1 × 10−3 for both specific storage Ss and specific yield Sy, a uniform Ss value 1 × 10−5, and Sy values of 0.01 for lower aquifer formations and 0.1 for upper aquifer formations, along with semiconfining conditions below the Athi sediments (Fig. 11c), finally produced the best fit to head observations (Fig. 11d) which was deemed reasonable. The overall satisfactory model fit, including for recent years in most parts of high abstraction areas of the NAS midlands, further supports the initial assumption made on the average abstraction rate per borehole and the resulting total groundwater abstraction and trend for the NAS.
Long-term simulation results
Figure 12 plots the aquifer water budget (fluxes) evolution over the historical and future time periods, using the best model (two-layer Sy 0.01–0.1) as reference, and the two alternative min-max models (uniform Sy of 0.01 and 0.1 respectively) to illustrate model output uncertainties associated with uncertainty in storage values. In all models, the groundwater discharge in rivers (computed as drains), i.e. baseflow (Fig. 12b), shows a decline over time concurrent with increasing groundwater abstraction (Fig. 12a). This suggests that aquifer depletion is likely to cause a decrease in baseflow resulting in a detrimental impact on environmental flows and groundwater dependent ecosystems. As abstraction increases, aquifer storage and water levels keep declining, which propagate laterally, triggering a decline in groundwater discharge to the drainage network. In future scenarios, the impact is severe under the GWDS groundwater-dependent scenario as water needs are met by continued increasing groundwater abstraction. If the trend continues, perennial drainage networks and springs may change to ephemeral waterbodies mainly depending on surface runoff from intense rains and local, low storage perched aquifers. This detrimental effect is less pronounced with the CWSS scenario where groundwater abstraction is capped from 2018 onwards and complemented by remote water transfer to meet increased water needs. This scenario slows the rate of declining baseflows and stabilizes the rate of change in aquifer storage. In this scenario, leaking pipes further contribute to recharge, because more water is brought from outside the NAS through pipelines embedded beneath the near surface.
In more detail, modelled flow and groundwater level outputs show that their relative uncertainty, resulting from uncertainties in aquifer storage values (Sy), increases with time, starting from low uncertainty in 1940 when the system is assumed to be in quasi steady state, to highest uncertainty in 2120 at the end of the GWDS scenario. Interestingly, in the GWDS scenario, from years 2035–2040, the model is not able to support the applied abstraction (projected cubic growth) with many of the shallowest wells, especially in the NCC area, losing productivity and/or drying up as a result of lowering water tables. This is reflected in Fig. 12a by the deviation of model (actual) abstraction from the projected (applied) abstraction, which rather follows a logistic growth, as typical from systems with limited resources. This deviation is higher with lower aquifer storage capacity, i.e. supported abstraction is higher with Sy 0.1, than with two-layer Sy 0.1–0.01, which is also higher than with Sy 0.01. For the CWSS scenario, uncertainty decreased again after 2018 due to the maintenance of constant abstraction resulting in a progressive return to quasi steady state. For both scenarios, the max Sy value (0.1) enables the aquifer to best support the increasing abstraction through change in storage, which results in a lower impact on baseflow. In contrast, with the min Sy value (0.01), the decrease in baseflow is more severe due to reduced release of water from aquifer storage. In the best model which has decreasing Sy with depth (0.1 in upper portions vs 0.01 in lower portions) corresponding to intermediate conditions between uniform max-min Sy values, the simulated total decrease in baseflow since the beginning of large-scale groundwater development in the 1950s is 9%, with a net storage loss of 1.5 billion m3. By 2120 in the GWDS scenario, baseflow further decreases by almost half (49%) of its value in 1940, with a net storage loss of 10.5 billion m3. In the CWSS scenario, baseflow stabilises to a value of 12% decrease since 1940 and storage recovers (net storage returns to close to zero) aided by the increasing recharge resulting from leaking pipes associated with increasing outsourced water supply.
In terms of groundwater levels, both scenarios display a declining average water table as shown in Fig. 11c with slower decline being experienced when conjunctive water supply (capped groundwater abstraction and imported water supplies) is initiated. The 2018 average groundwater levels dropped by about 4 ± 1 m since 1940, which corresponds to about 5 ± 0.1 cm/year.
Spatial changes in simulated groundwater levels for the best model (two-layer Sy) and respective scenarios (GWDS and CWSS) are further presented in Fig. 13, which illustrates a regional rather than localised depletion, as consistently demonstrated by observations shown in Fig. 10. However, a greater depletion is experienced in the west and central region of the Greater NCC area. In the central region this is evidently due to the high population density with greater groundwater demand and high borehole density resulting in over-exploitation, which is projected to get worse, with most areas having no piped water supply (Karen, Rongai, and the neighbouring town of Kikuyu). The depletion of the western region is the hydrodynamic effect of the central depletion that lowers upgradient groundwater levels due to the decline of the piezometric base level in the central region. In 2120, simulated depletion is much higher for scenario GWDS than CWSS, by up to 24 ± 3 m of water-table decline on average for the former, i.e. six times this of 2018, as compared to 6 ± 1 m for the latter (Fig. 10). This means a further 20 and 2 m decline as compared to 2018 for GWDS and CWSS scenarios, respectively. When looking in more detail at the spatial distribution of groundwater declines, however (Fig. 13), the total water-table drop below and immediately upgradient of Nairobi County reaches a max estimate of 200 m (~1 m/year, continuing) in 2120, under scenario GWDS. For recent times (2018), the model estimates an average value of decline of around 30 m (max simulated value 46 m) across NCC tapped areas. This is broadly consistent with well observations within the NCC (36 m in average in 2015, Table 4), although suggesting, on the conservative side, a possible overall underestimation of depletion by the model.
As a result of both regional and local depletion, most springs, especially for scenario GWDS in the west are expected to experience decreased water discharge or may even cease flowing during dry seasons. The recharge from leaking supply pipes is contributing positively to groundwater storage capacity around the NCC area even though it is overshadowed by abstraction. This observation encourages groundwater artificial recharge as a way of restoring declining NAS groundwater levels.