3.3. Health risk map
A health risk map was made using the optimal cut-off value of 0.47 (Fig. 5). It identifies numerous densely populated regions where people rely on groundwater associated with high fluoride. In total, over 13 million people are estimated to be potentially affected by fluoride contamination in groundwater, which is 6.0% of the total population in Pakistan. However, this number would be even larger if the population in regions with probabilities under the cut-off of 0.47 would be taken into consideration. With the increasing population in Pakistan (population growth rate around 2% in the year 2020) (World Bank, 2020), the problem may become more severe in the future if the reliance on groundwater remains high. Large at-risk populations are found in northern Punjab, Islamabad, and Khyber Pakhtunkhwa. Furthermore, the fluoride risk map (Fig. 5) indicates that residents in the cities of Lahore, Sargodha, Depalpur, Peshawar, Bannu, Karachi, Quetta, and others are at high risk, which is confirmed by the high prevalence of fluorosis in some of these cities (Ahmad et al., 2020; Mohsin et al., 2014; Rahman et al., 2018; Sami et al., 2016). Conversely, the total number of people at risk in the Thar Desert (SE Sindh), which has high probabilities of fluoride contamination, is far smaller owing to its sparse population, yet residents there are still under high risk.
The presented probability and health risk maps (Fig. 3, Fig. 5) raise awareness about fluoride contamination and its adverse health impacts in Pakistan. Furthermore, they can help authorities in prioritizing areas for implementing mitigation measures. These could include monitoring programs for drinking water wells or fluoride removal, e.g. adsorption treatment (Bhatnagar et al., 2011) or membrane separation processes (Waghmare and Arfin, 2015), and improving health management systems. Compared to the previous nationwide representation of fluoride at the sub-tehsil-scale (Khan et al., 2002), the novel maps presented here have a 3–4 order of magnitude higher spatial resolution (250 m), are based on much larger new datasets, and predict the probability of high groundwater fluoride for areas where data are lacking. Also in relation to a recent study by Khattak et al. that contains clusters of many groundwater fluoride measurements across much of Punjab (Khattak et al., 2021), the new maps identify hotspots, e.g. in the Sargodha Division, that that study did not uncover.
This study, which presents a large new dataset of fluoride in groundwater across Pakistan, combined with geospatial modeling and risk mapping using various environmental predictors, highlights several regions where exposure to high fluoride levels pose a significant public health risk. Hot spots include the Thal Desert in Punjab (Sargodha Division), the Thar Desert in Sindh, and the Sulaiman Mountains in the western part of the country. Analysis of the importance of the predictor variables and their correlation with fluoride show that high fluoride concentrations in groundwater benefit from arid climatic conditions with high temperatures and evapotranspiration, the presence of fluoride-bearing minerals (e.g. carbonate sedimentary rock), and the presence of calcisols.
Knowing the countrywide groundwater fluoride risk and affected populations shall be helpful for authorities and water resource managers in identifying fluoride-contaminated wells and mitigating the risk for residents. All groundwater wells in areas with a high probability (e.g., above the cut-off value of 0.47) should be tested, for instance, in the Thar Desert and the Sargodha Division (especially the Bhakkar, Mianwali, and Khushab districts in the upper Thal Desert). Particular attention should also be paid to risk areas with a high population density such as Lahore, Sargodha, Depalpur, Peshawar, and Bannu. Mitigation measures include monitoring, provision of alternative sources of drinking water, fluoride removal treatment, and awareness-raising campaigns. These maps are not a replacement for actual groundwater testing but indicate hazard and risk for drinking water use.
Future work could consider additional groundwater contaminants, e.g. uranium, nitrate, pesticides or salinity in order to obtain a more comprehensive understanding of the safety of groundwater. Model accuracy could be further improved by incorporating additional data and other predictor variables, such as hydrological parameters, if available.
The fluoride prediction map was compared with a similar fluoride map of India (Podgorski et al., 2018) (Fig. S8). High probabilities of fluoride exceeding 1.5 mg/L in northern Punjab align very well with those across the border in Indian Punjab. Similarly in the Thar Desert, which forms a natural boundary between Pakistan and India, the predictions on either side of the border are compatible. Whereas these sections are well constrained by fluoride data, border areas in between match less well, which may be due in part to a lack of measurements in southern Pakistani Punjab.
Since much of Pakistan has a warm, dry climate and people consequently drink more water, a prediction model was also created for fluoride concentrations exceeding 1.0 mg/L (Fig. S7). However, with only 543 of the 5483 measurements falling in the range of 1.0–1.5 mg/L, the high-hazard regions of the prediction maps for 1.5 mg/L (Fig. 3) and 1.0 mg/L (Fig. S7) are largely similar. For example, higher probabilities for 1.0 mg/L were found for central Pakistan near the borders of Punjab, Khyber Pakhtunkhwa, and Balochistan. The selection of variables by RFE was also similar for both models (1.0 mg/L and 1.5 mg/L).
3.2.3. Predictor importance
Some of the most important predictors are the climate parameters (Fig. 2a). The correlations in Table S4 also confirm that drier climate conditions favor fluoride release. High importance was also found for the elevation variable, which is strongly negatively correlated with temperature.
The two continuous variables of soil parameters, fraction of coarse soil fragments and nitrogen fraction, also have high importance measures. The fraction of coarse fragments is high in the Sulaiman Mountains of eastern Balochistan and western Punjab and Sindh and are composed of mixed or carbonate sedimentary rock that may contain fluoride-bearing minerals (Fig. 4). On the other hand, nitrogen fraction is associated closely with the presence of forested mountains in the north, which has lower temperatures, and where the sparse measurements generally show low fluoride concentrations. The “calcisols” binary soil predictor, which is associated with substantial accumulations of lime, is connected to the presence of high fluoride concentrations, as the precipitation of calcite removes calcium from dissolution and results in higher fluoride concentrations (Banerjee, 2015).
*See original article at for more maps, https://www.sciencedirect.com/science/article/pii/S0048969722031552?via%3Dihub