**3. Results**

Microbiological and physical–chemical quality parameters of 252 water sources spread all over Guinea-Bissau were evaluated. All the studied wells were hand-dug, shallow, without proper wall isolation, often located at the vicinity of latrines and waste dumps (<30 m), and most of them lacked any well cover or fence to prevent contamination. Regarding the method of water collection, the majority of the wells were fitted with a bucket (*n* = 122), 72 with a manual pump, 6 with a solar/electric pump, and 13 with a mixed version combining pump and bucket. Of the analysed boreholes, 4 were fitted with a distribution system with faucets, 23 with a solar/electric pump, 4 with manual pumps, and 4 with a mixed version combining pump and bucket (Table S1).

Microbiological and physical–chemical analysis results are summarized in Table 1 and Figure S2. Water temperature averaged 29.1 ◦C (22.9–35.3 ◦C) year-round, with only two wells sampled in each season below the 25 ◦C recommended maximum temperature for drinking water, according to UK standards [20]. The water conductivity was low to moderate (median 163 µS/cm), with oxygen concentrations averaging 5.5 mg/L (average oxygen saturation 59%). The water was acidic to very acidic, averaging pH 5.3. Averages were similar between seasons, being higher in boreholes (pH 6.6 vs. pH 5.2 in wells). Overall, 83% of the water sources surveyed exhibited pH below the EU parametric value for drinking water, representing 89% and 49% of the wells and boreholes studied, respectively. Acceptable standard values for colour and turbidity were exceeded in 56% and 40% of the sampled water sources, respectively. As expected, extremely high values of these parameters were recorded in the wet season in the shallow wells (Table 1). Additionally, higher values were obtained when a bucket was used to withdraw the water (Figure S2). The nitrate, nitrite, and ammonium concentrations were below the EU and WHO parametric values for drinking water in most wells, although a value slightly above the acceptable parametric threshold was registered in the wet season. The highest concentrations of nitrate and ammonium were observed in wells and associated with the use of buckets (Figure S2). Overall, heavy metal (Al, As, Cr, Cu, and Fe), and cyanide concentrations were under the parametric value for drinking water. However, a higher number of surveyed water sources revealed non-compliance with the standard parametric values related to the wet season, wells, and when using a bucket to collect water (Table 1 and Figure S2).

**Table 1.** Minimum and maximum values for water quality parameters seasonally assayed, according to water source type. In italics are the percentage of sites above the parametric values for drinking water, and the respective number of water sources surveyed. EU—European Union parametric values for drinking water [19], WHO—Word Health Organization guidelines values for drinking water [18].


a—Acceptable for consumption (<5 PtCo).

The obtained results showed that the drinking water available to the population was grossly polluted with faecal material. The majority of the water sources sampled (83% and 77% for FC and IE, respectively), failed to meet the microbiological quality standards for drinking water, as recommended by the WHO and the EU. Values averaged 4217 and 800 CFU/100 mL for FC and IE, respectively (Table 1). Overall, faecal contamination levels were significantly higher during the wet season (Tukey's HSD test, *p* < 0.05), with values as high as 376, 850 CFU/100 mL for FC, and 33,000 CFU/100 mL for IE. Additionally, significantly higher (Tukey's HSD test, *p* < 0.05) faecal contamination levels were associated with wells and the use of a bucket for water collection. The lowest levels of contamination were observed in water samples collected from boreholes associated with electric pumps with small water distribution systems fitted with faucets (Figure 2).

with small water distribution systems fitted with faucets (Figure 2).

drinking water, as recommended by the WHO and the EU. Values averaged 4217 and 800 CFU/100 mL for FC and IE, respectively (Table 1). Overall, faecal contamination levels were significantly higher during the wet season (Tukey's HSD test, p < 0.05), with values as high as 376, 850 CFU/100 mL for FC, and 33,000 CFU/100 mL for IE. Additionally, significantly higher (Tukey's HSD test, p < 0.05) faecal contamination levels were associated with wells and the use of a bucket for water collection. The lowest levels of contamination were observed in water samples collected from boreholes associated with electric pumps

Figure 2. Variation in the microbiological water quality parameters—(a) faecal coliforms and (b) intestinal enterococci—in the water sources across Guinea-Bissau between (1) seasons, (2) source type, and (3) method of water collection. Diamond shapes represent the mean. Circles represent **Figure 2.** Variation in the microbiological water quality parameters—(**a**) faecal coliforms and (**b**) intestinal enterococci—in the water sources across Guinea-Bissau between (1) seasons, (2) source type, and (3) method of water collection. Diamond shapes represent the mean. Circles represent outliers.

FC and IE were significantly correlated (r = 0.85, p < 0.05), as expected. Significant positive correlations (p < 0.05) were observed between both faecal indicators and colour (rFC = 0.32, rIE = 0.30), turbidity (rFC = 0.50, rIE = 0.49), ammonium (rFC = 0.32, rIE = 0.30), nitrate (rFC = 0.34, rIE = 0.35), nitrite (rFC = 0.33, rIE = 0.30), aluminium (rFC = 0.12, rIE = 0.12), chromium (rFC = 0.20, rIE = 0.14), copper (rFC = 0.15, rIE = 0.20), cyanide (rFC = 0.10). and iron (rFC = 0.22, rIE = 0.26). Turbidity and colour, associated with downpours during the wet season, showed positive significant correlations (p < 0.05) with the nitrogen species, and the ma-FC and IE were significantly correlated (r = 0.85, *p* < 0.05), as expected. Significant positive correlations (*p* < 0.05) were observed between both faecal indicators and colour (rFC = 0.32, rIE = 0.30), turbidity (rFC = 0.50, rIE = 0.49), ammonium (rFC = 0.32, rIE = 0.30), nitrate (rFC = 0.34, rIE = 0.35), nitrite (rFC = 0.33, rIE = 0.30), aluminium (rFC = 0.12, rIE = 0.12), chromium (rFC = 0.20, rIE = 0.14), copper (rFC = 0.15, rIE = 0.20), cyanide (rFC = 0.10). and iron (rFC = 0.22, rIE = 0.26). Turbidity and colour, associated with downpours during the wet season, showed positive significant correlations (*p* < 0.05) with the nitrogen species, and the majority of metal concentrations (Figure S3).

jority of metal concentrations (Figure S3). Hand-dug wells represented the majority of the water sources sampled, with the highest contamination levels; therefore, the relationship between environmental factors and microbiological indicators was further explored using a regression analysis approach. Hand-dug wells represented the majority of the water sources sampled, with the highest contamination levels; therefore, the relationship between environmental factors and microbiological indicators was further explored using a regression analysis approach. As a result, the fitted BRT models for the microbiological parameters FC and IE were similar in performance, although the FC model performed slightly better, with an explained deviance of 49% and a cross-validated correlation of 0.7 (Table 2).

outliers.


**Table 2.** Predictive performance of the final models developed for the microbiological indicators quantified in wells in Guinea-Bissau. Final settings: bag fraction—0.5; tree complexity—1; no. folds—10; learning rate—0.001. folds—10; learning rate—0.001. Faecal Coliforms Intestinal Enterococci

Water 2022, 14, x FOR PEER REVIEW 7 of 15

deviance of 49% and a cross-validated correlation of 0.7 (Table 2).

As a result, the fitted BRT models for the microbiological parameters FC and IE were similar in performance, although the FC model performed slightly better, with an explained

Table 2. Predictive performance of the final models developed for the microbiological indicators quantified in wells in Guinea-Bissau. Final settings: bag fraction—0.5; tree complexity—1; no.

The partial responses for FC and IE for each predictor variable and variable contributions are shown in Figures 3 and 4, respectively. Both models retained the same eight predictor variables. For both models, the two most influential variables were turbidity and the method used to collect water, accounting for 57.9% and 55% of the explained deviance for FC and IE, respectively. The levels of faecal contamination increased with higher turbidity and the use of a bucket to collect water. The contributions of the remaining six variables differed between models, but the effects on the response were similar. The concentrations of faecal indicators were higher in the wet season, although precipitation was a more influential variable for the FC model. Latitude was also an influential variable in both models, with contamination increasing with increasing latitudes (towards northern regions). The levels of contamination also increased with higher concentrations of nitrate and pH. Faecal indicators concentration decreased as temperature rose. The effects of copper were different for the two models: FC concentration showed a slight decrease with higher levels of the metal, whereas IE concentration peaked and then decreased as metal levels increased. butions are shown in Figures 3 and 4, respectively. Both models retained the same eight predictor variables. For both models, the two most influential variables were turbidity and the method used to collect water, accounting for 57.9% and 55% of the explained deviance for FC and IE, respectively. The levels of faecal contamination increased with higher turbidity and the use of a bucket to collect water. The contributions of the remaining six variables differed between models, but the effects on the response were similar. The concentrations of faecal indicators were higher in the wet season, although precipitation was a more influential variable for the FC model. Latitude was also an influential variable in both models, with contamination increasing with increasing latitudes (towards northern regions). The levels of contamination also increased with higher concentrations of nitrate and pH. Faecal indicators concentration decreased as temperature rose. The effects of copper were different for the two models: FC concentration showed a slight decrease with higher levels of the metal, whereas IE concentration peaked and then decreased as metal levels increased.

Figure 3. Boosted regression tree (BRT) partial dependence plots showing the effect of each predictor variable on faecal coliforms (FC) of the water of wells in Guinea-Bissau: Turb—turbidity; Method—method used for water collection; Prec—precipitation; NO3—nitrate; Temp—temperature; Lat—latitude; Cu—copper. At each plot, the fitted function shows the relationship between the response variable (y−axis), and the predictor variable (x−axis), holding the values of all other variables at their mean. The relative contribution (%) of each predictor variable for the BRT model is shown in brackets. **Figure 3.** Boosted regression tree (BRT) partial dependence plots showing the effect of each predictor variable on faecal coliforms (FC) of the water of wells in Guinea-Bissau: Turb—turbidity; Method method used for water collection; Prec—precipitation; NO3—nitrate; Temp—temperature; Lat latitude; Cu—copper. At each plot, the fitted function shows the relationship between the response variable (*y*−axis), and the predictor variable (*x*−axis), holding the values of all other variables at their mean. The relative contribution (%) of each predictor variable for the BRT model is shown in brackets.

Figure 4. Boosted regression tree (BRT) partial dependence plots showing the effect of each predictor variable on intestinal enterococci (IE) of the water of wells in Guinea-Bissau: Turb—turbidity; Method—method used for water collection; NO3—nitrate; Lat—latitude; Prec—precipitation; Cu copper; Temp—temperature. At each plot, the fitted function shows the relationship between the response variable (y−axis), and the predictor variable (x−axis), holding the values of all other variables at their mean. The relative contribution (%) of each predictor variable for the BRT model is shown in brackets. **Figure 4.** Boosted regression tree (BRT) partial dependence plots showing the effect of each predictor variable on intestinal enterococci (IE) of the water of wells in Guinea-Bissau: Turb—turbidity; Method—method used for water collection; NO3—nitrate; Lat—latitude; Prec—precipitation; Cu copper; Temp—temperature. At each plot, the fitted function shows the relationship between the response variable (*y*−axis), and the predictor variable (*x*−axis), holding the values of all other variables at their mean. The relative contribution (%) of each predictor variable for the BRT model is shown in brackets.
