Groundwater Arsenic Contamination Throughout China

See allHide authors and affiliations

Science  23 Aug 2013:
Vol. 341, Issue 6148, pp. 866-868
DOI: 10.1126/science.1237484

Arsenic and Populace

The solubility of arsenic in groundwater aquifers is controlled by a number of hydrologic and geochemical factors. In rural communities that rely on groundwater for drinking water, the risk from exposure may pose a public health threat, especially when groundwater pumping can increase arsenic solubility. In an effort to provide a focused assessment of risk to arsenic exposure from groundwater, Rodríguez-Lado et al. (p. 866; see the Perspective by Michael) constructed a geostatistical model that incorporates a number of factors that control arsenic solubility across China. Most of the risk centers in a few provinces—Xinjiang, Inner Mongolia, Henan, Shandong, and Jiangsu—but the total population exposed to arsenic levels above 10 micrograms per liter could be upwards of 19 million people.


Arsenic-contaminated groundwater used for drinking in China is a health threat that was first recognized in the 1960s. However, because of the sheer size of the country, millions of groundwater wells remain to be tested in order to determine the magnitude of the problem. We developed a statistical risk model that classifies safe and unsafe areas with respect to geogenic arsenic contamination in China, using the threshold of 10 micrograms per liter, the World Health Organization guideline and current Chinese standard for drinking water. We estimate that 19.6 million people are at risk of being affected by the consumption of arsenic-contaminated groundwater. Although the results must be confirmed with additional field measurements, our risk model identifies numerous arsenic-affected areas and highlights the potential magnitude of this health threat in China.

China faces groundwater quality problems of enormous proportions from both industrial and natural sources (1). Populations at risk of exposure to excessive levels of arsenic, a natural groundwater contaminant, have been emerging since the 1960s (27). With the exception of Guizhou province, where endemic arsenicosis is mainly due to coal burning (7, 8), most cases of water-related arsenic poisoning have occurred in arid regions of the northern provinces, probably because of the scarcity of clean water for cooking and drinking from sources other than groundwater.

Apart from geothermal and mining environments, two main environmental conditions are known to be linked to natural arsenic enrichment in groundwater systems (9): (i) aerobic alkaline environments in closed basins in arid and semiarid regions, where high pH leads to alkaline desorption of arsenic from mineral oxides, and (ii) aquifers under strongly reducing conditions, where the release of arsenic is related to reductive dissolution of arsenic-bearing iron (hydr)oxides in sediments. Numerous studies in China have related arsenic-enriched groundwater to reducing conditions (1016), particularly in arid regions, where such conditions in alluvial and lacustrine aquifer sediments are often concomitant with high alkalinity and/or salinity. In other areas, such as Taiwan, arsenic release has also been associated with anoxic aquifers (17). The few studies reporting arsenic measurements in oxic aquifers in China (1820) indicate that arsenic concentrations in oxic groundwaters are mostly below 10 μg liter−1.

Between 2001 and 2005, the Chinese National Survey Program, conducted by the Chinese Ministry of Health, tested some 445,000 wells in 20,517 villages of 292 counties (12% of all counties in China) for arsenic contamination. In almost 5% of wells, arsenic levels were higher than the former Chinese standard of 50 μg liter−1, and about 10,000 individuals were found to be affected by arsenicosis in known and suspected endemic areas (21). Screening of wells has continued, and it has been estimated that 5.6 million people are exposed to high concentrations of arsenic in drinking water (>50 μg liter−1) and that some 14.7 million are exposed to arsenic concentrations of >10 μg liter−1 (22). Because of the sheer size of China, it will take several decades to complete the screening of millions of wells to determine the spatial occurrence and magnitude of arsenic contamination throughout the country.

Here, we present a predictive model that uses survey data and geological and hydrogeochemical parameters as proxies of processes that affect arsenic solubility in groundwater aquifers to identify safe and unsafe areas with respect to geogenic groundwater arsenic contamination. The model was validated with an independent data set and a cut-off probability value of 0.46 (see supplementary materials), revealing fairly good agreement (Cohen’s kappa = 0.51), with 77% of the samples correctly classified in the high- and low-risk classes. The model sensitivity (ability to correctly classify samples with arsenic concentrations of >10 μg liter−1) and specificity (ability to correctly classify samples with arsenic concentrations of ≤10 μg liter−1) were 83% and 75%, respectively, which indicates good performance for the prediction of both safe and unsafe areas with respect to arsenic. The model uncertainty is predominantly low (fig. S5) but exceeds 0.5 in the sparsely inhabited sand deserts.

Previous studies in Southeast Asia have identified a relatively small number of geological and hydrogeochemical parameters as significant spatial predictors that can be used to characterize the regional distribution of high arsenic (2326). Holocene sediments, for example, have been found to be strongly associated with arsenic contamination. Such coincidence can also be observed in areas reported to suffer from arsenic contamination in China (Fig. 1). We initially considered 16 environmental proxies, in the form of 30 arc sec (~1 km2) digital raster maps, as potential factors for identifying areas affected by geogenic arsenic hazard. Only eight proxies were significantly linked to the regional distribution of high arsenic occurrence and were used, together with 2668 georeferenced arsenic measurements, to develop an ensemble model with binomial logistic regression as the base classifier (see supplementary materials). The model (Fig. 2A) forecasts that the area at risk of groundwater arsenic contamination (>10 μg liter−1) may encompass more than 580,000 km2, and it pinpoints elevated arsenic concentrations in regions where endemic arsenic poisoning has already been detected (2730). Large areas such as Tarim basin (Xinjiang), Ejina basin (Inner Mongolia), Heihe basin (Gansu), Qaidam basin (Qinghai), the Northeastern plain (Inner Mongolia, Jilin, and Liaoning), and the North China plain (Henan and Shandong) were identified as being potentially affected, although these results must be confirmed with additional field measurements.

Fig. 1 Location of known and potential arsenic-affected basins in China.

Areas with high levels of arsenic (As) are generally characterized by Holocene sediments (green), where large basins may be affected.

Fig. 2 Arsenic predictive map of China.

(A) Modeled probability of arsenic concentrations exceeding the 10 μg liter−1 threshold in groundwaters. Boxes indicate regions shown in (B). (B) Basins used to describe the model performance.

The mean model coefficients (table S1) indicate a positive correlation of Holocene sediments, soil salinity, fine subsoil texture, topographic wetness index (TWI; see supplementary materials), and density of rivers with high arsenic, and a negative contribution of slope, distance to rivers, and gravity. Holocene sediments, soil salinity, subsoil texture, and TWI were significant in most of the ensemble members, highlighting their relative importance in predicting the occurrence of groundwater arsenic in relation to the remaining variables. This is consistent with the conditions found in most studies performed in high-risk areas in China, such as Xinjiang province (model sensitivity = 85%) and Hetao-Huhhot basin (model sensitivity = 95%), where high arsenic tends to be associated with the anoxic environments of poorly drained aquifers in young sediments with flat topography and high salinity (Fig. 2B) (16). The model also performed well for oxic aquifers, such as Minqin basin, Chahaertan oasis, and Liao-Ho basin (Fig. 2B, areas 3 and 4). Surveys in these areas (21, 22) revealed that these aquifers bear low arsenic concentrations, as correctly predicted by our model with an accuracy ranging from 83 to 98%.

We categorized areas on the probability map as low-risk (P ≤ 0.46) and high-risk (P > 0.46), and cross-referenced the map with a population distribution map for the year 2000. This yielded an estimate of 19,580,000 people in China who live in high-risk areas, mainly in Xinjiang, Inner Mongolia, Henan, Shandong, and Jiangsu provinces (Fig. 3). This may be an overestimation of the actual population at risk, because treated and piped water may be used in some places, but water use statistics are not available. However, most of the arsenic-affected areas in China correspond to arid or semiarid regions, where groundwater is the predominant source of drinking water.

Fig. 3 Estimated Chinese population at potential risk of exposure to arsenic concentrations of >10 μg liter−1.

Calculation is based on the model results and population density estimates.

Our approach complements traditional groundwater quality surveys, which are expensive and time-consuming. The model requires only a small number of geospatial parameters to provide a preliminary assessment of affected areas, reducing the area that needs to be screened. It may also be appropriate for use in other parts of the world, especially in arid regions such as northwestern Argentina and Chile, where high arsenic concentrations have been reported, or in countries such as Mongolia, Kazakhstan, and Kyrgyzstan, where risk assessments for groundwater arsenic contamination have not yet been performed.

Supplementary Materials

Materials and Methods

Figs. S1 to S5

Tables S1 to S7

References (31–50)

References and Notes

  1. Acknowledgments: Supported by the Sino-Swiss Science and Technology Cooperation Program of the Swiss State Secretariat for Education and Research (project IZLCZ2 123971), the National Science and Technology Pillar Program of China during the 11th Five-Year Plan Period (contract grant no. 2006BAI06B04), and the External Cooperation Program of the Chinese Academy of Sciences. We thank G.-B. Jiang and J. Shi of the Research Center for Eco-Environmental Sciences, Beijing, for support and data collection in the initial phase of the project. L.R.-L. also acknowledges the Isidro Parga Pondal research program from the Xunta de Galicia (Spain) for current financial support. Additional data are available at
View Abstract

Navigate This Article