Mapping fluoride and arsenic hot spots

New models predict where the highest levels of natural fluoride and arsenic occur in groundwater.

Geochemists and statisticians have joined forces to map the presence of naturally occurring fluoride and arsenic on a global scale. Though far from perfect, the new probability maps, published in ES&T (DOI: 10.1021/es702859e; 10.1021/es071958y), have the potential to provide red flags on contaminated drinking-water sources, particularly in developing countries where on-the-ground data are lacking.

In high enough quantities, arsenic and fluoride can have detrimental effects on humans. Chronic arsenic exposure over decades triggers skin diseases, liver damage, and skin and lung cancer. Groundwater with levels of arsenic above the 10 micrograms per liter guideline from the World Health Organization has created such health problems in Vietnam, Bangladesh, and other places with geologically similar terrain (Environ. Sci. Technol. 2007, 41, 2074).

Fluoride, which may prevent dental decay at levels below 1 milligram per liter (mg/L), is added to drinking-water supplies in many developed countries to protect people’s teeth. But too much fluoride can cause discolored or malformed teeth, bone diseases, neurological effects, and other health problems. A recent assessment by the National Research Council suggested that daily maximum exposure guidelines, set by the U.S. EPA at 4 mg/L, should be revisited.

To help communities predict where they could sink wells to avoid groundwater with naturally high fluoride and arsenic concentrations, the Water Resource Quality group of the Swiss Federal Institute of Aquatic Science and Technology (Eawag), led by Annette Johnson gathered as much information as they could find on rock types, fault zones, topography, precipitation, soil pH, and other pertinent characteristics across the planet. They then assembled the data in geographic information system (GIS) models that used proxies for the presence of fluoride and arsenic from geologic sources.

Geology, evapotranspiration, and soil pH were key elements of the modeling. The team delineated eight “process regions” for fluoride according to climate and rock types. For arsenic, the team divided the world into “reducing” and “high-pH/oxidizing” regions, reflecting processes that mobilize arsenic, based on soil pH and water pathways.

To model the elements’ concentrations, the researchers relied on stepwise regression and fuzzy-logic equations that are part of neural networks. They tested the model results with a small amount of real-world data. For some places, such as the East African Rift Valley and Senegal, where the volcanic terrain and faults heavily influence fluoride levels in the groundwater, the models were quite accurate. However, truly validating the predictions requires additional field testing, the team members emphasize.

“The distinction between oxidizing and reducing aquifers on the basis of surface parameters—that’s novel” for arsenic, says Lex van Geen of the Lamont-Doherty Earth Observatory at Columbia University. “It’s pretty amazing to me that they can describe about two-thirds of the variance in the data just from surface parameters.” Still, he says, the third dimension poses a problem for taking into account what happens to arsenic in wells deeper than 20–30 meters.

“Subsurface geology is critically important,” comments George Breit of the U.S. Geological Survey (USGS). Values that describe soil or rock types at the surface could be masking subsurface water flow and deep geologic sources. But getting those data is extremely difficult, especially on a global scale.

The global-scale modeling, particularly for fluoride, “may be a very good first cut, but it missed some things,” Breit adds. For example, EPA and USGS data show elevated fluoride concentrations in the southeastern U.S. and high arsenic levels in New England groundwater—but neither hot spot is predicted by the new maps. However, Breit says, the new results are a “demonstration of the coming power” of GIS methods and particularly of neural networks, a statistical method still in its early stages.

“I don’t see it as a final product,” says Johnson, whose team plans to evaluate the models with on-the-ground observations in places such as China, Kenya, and the U.S., all of which have different settings and data resources. She and her colleagues also remain adamant that modeling cannot replace direct tests of a water source for fluoride and arsenic levels.

Noting that groundwater used for drinking must always be tested, Donna Myers, chief of the USGS National Water-Quality Assessment Program, says that “the ability to predict [levels in] areas that are unsampled [is] an improvement over knowing nothing at all.” Naturally occurring fluoride has been overshadowed in the U.S. by other elements of concern in groundwater, such as arsenic, radon, or uranium, she comments. With more detail, predictive maps “would provide much more information that would be useful to EPA” for policy and planning purposes, as well as to nongovernmental organizations that may eventually reap the benefits of having such data available to them on the ground. —NAOMI LUBICK