On-site sanitation density and groundwater quality: evidence from remote sensing and in situ observations in the Thiaroye aquifer, Senegal

In rapidly urbanising low-income towns and cities, there remains an absence of scientific evidence and regulatory structures to sustain the quality and quantity of groundwater used for low-cost water supplies and to reconcile this with continued use of the subsurface for low-cost sanitation. Here, we analyse the relationship between the density of on-site sanitation and shallow groundwater quality in the Thiaroye aquifer of Quaternary sands in Dakar, Senegal. On-site sanitation was mapped using object-oriented classification and visual interpretation of high-resolution, optical satellite images and ground-truthing surveys. Groundwater quality was assessed over a three-year period (2017–2019) from a network of 61 sources comprising boreholes, dug-wells, hand tubewells and piezometers. More than 253,000 on-site sanitation facilities are identified over an area of 520 km with densities ranging from 1 to 70 per hectare. A moderate, statistically significant linear relationship (r1⁄4 0.55, p « 0.01) is found between the density of on-site sanitation facilities and nitrate concentrations in sampled groundwater sources. Groundwater contamination beyond the WHO drinking-water guideline value (50 mg/L) occurs where densities of on-site sanitation facilities exceed 4 (±4) per hectare, a threshold commonly surpassed in peri-urban areas underlain by the Thiaroye aquifer of Dakar.


INTRODUCTION
Groundwater is a climate-resilient source of freshwater in sub-Saharan Africa (Cuthbert et al. ), playing a vital role in sustaining improved access to safe water in pursuit of United Nations Sustainable Development Goal 6water and sanitation for all by 2030. The strategic importance of groundwater in urban areas lies additionally in: (1) its lower vulnerability to contamination relative to surface waters reducing treatment costs, and (2)  Research to date has focused on protecting groundwater sources from faecal contamination in African towns and cities, and is dominated by local-scale analyses evaluating setback (safe) distances between 'source', on-site sanitation facility, and 'receptor', an individual groundwater-fed waterpoint (e.g. ARGOSS ; Howard et al. Here, we conduct the first city-scale analysis in sub-Saharan Africa of the relationship between the density of on-site sanitation (generally unimproved) and shallow groundwater quality within an unconsolidated Quaternary sand aquifer in the Thiaroye area of Dakar (Senegal). To map the density of on-site sanitation facilities over an area of ∼520 km 2 , we apply an object-oriented classification and visual interpretation of high-resolution, optical satellite imagery (i.e. Quickbird, Geoeye, Orbview), supported by extensive ground-truthing surveys. The density of on-site sanitation facilities (generally unimproved) is then compared to observed concentrations of faecal contamination from solutes (nitrate) and bacteria (Escherichia coli) sampled from 61 shallow ground-water sources over three years (2017-2019) to assess their association (dependency).
Key intended outcomes of this city-scale analysis include an improved understanding of the capacity of this unconsolidated sand aquifer to attenuate faecal effluent from on-site sanitation facilities and guidance concerning the carrying capacity of the aquifer to provide safe water and store faecal waste.

Study area
The study area lies within the Thiaroye shallow aquifer on per km 2 with an average household size of 6 individuals.
Sanitation provision in the study area derives from on-site systems that consist almost entirely of septic tanks; access to on-site sanitation is estimated to be ∼95% (ONAS ).
Faecal loading from septic-tank effluent has progressively contaminated the Thiaroye aquifer (Diédhiou ) from which abstraction for public water supplies has declined by 84% from 3.7 × 10 6 m 3 /year in 1968 to 5.  (Table 1). Through this approach we estimated the number of on-site sanitation facilities within a hectare (100 × 100 m) grid across the unsaturated zone of the shallow aquifer over a total area of 520 km 2 comprising 51,500 grids.
Classification of the Quickbird satellite imagery required preliminary processing (i.e. calibration, radiometric enhancement and geometric correction) and an object-oriented classification with segmentation process. Radiometric calibration of satellite imagery makes it possible to convert the digital signal recorded by the satellite into a physical variable such as radiance in order to discern objects and select learning samples. Radiometric enhancement stretches the histogram of the satellite image (coded to 8 bits) so that it extends across the whole range of radiometric responses (i.e. 1-255 patterns). An accuracy assessment was then undertaken to assess classification errors. For each of the 10 classes, 100 test pixels were created at points visually identified as the referenced class in the imagery and then compared with the corresponding location in the classified image for accuracy.
The overall performance of the classification gives very satisfactory results with an overall accuracy of 93% and a Kappa index of the order of 0.93 (Table 2).
The object-oriented classification was preceded by segmentation which allowed us to refine the object-oriented classification by reducing as much as possible spectral confusions. This process made it possible to map surfaces occupied by houses based not only on their spectral values but also on their morphological characteristics. As a result, the analysis was able to discriminate as best as possible areas occupied by the houses from areas not occupied by the houses. Ground-truthing field surveys were facilitated by consulting cadastral maps of each district community to obtain an overview of the average size of houses, usually 15 × 10 m, but occasionally 7.5 × 10 m. Field surveys were conducted along 10 well-distributed, validation grid transects (200 × 300 m) with the aid of GPS-marked boundaries.
During these, the location of encountered houses and type of on-site sanitation facilities (e.g. septic tanks, pit latrines) were recorded; 1,267 houses were visited to validate mapping of on-site sanitation facilities.  Linear regression was employed to test for possible association (dependency) between nitrate contamination

Mapping on-site sanitation facilities results
On-site sanitation mapping reveals the density of on-site sanitation facilities per hectare ( Figure 2). In total, the analysis identified 253,014 septic tanks over an area of 520 km 2 .
Densities of on-site sanitation vary between 1 and 70 septic tanks and pit latrines per hectare (100 × 100 m), with an average density of 20 per hectare. Cartographic results were validated from ground-truthing (Figure 3(a)) which consisted of a census of the number of houses located within the 10 validation grids distributed over the entire  (Table 3).

Groundwater quality
Mean nitrate concentrations recorded in 2017, 2018, and 2019 vary considerably in space across the Thiaroye aquifer    (Table S3).

Relationship between on-site sanitation density and shallow groundwater faecal contamination
Groundwater sources with very high nitrate concentrations and E. coli counts generally but not exclusively correspond to areas with very high on-site sanitation densities. In areas where the density of on-site sanitation is lower, there is an associated decrease in nitrate concentrations in sampled groundwater (Figure 5