USING GEOGRAPHIC INFORMATION SYSTEMS TO PREDICT ARSENIC CONCENTRATIONS IN GROUNDWATER OF SAN LUIS VALLEY, COLORADO

Consumption of inorganic arsenic in drinking water at high levels has been associated with chronic diseases. Risk is less clear at lower levels of arsenic, in part due to difficulties in estimating exposure. Herein we characterize spatial and temporal variability of arsenic concentrations, and develop models for predicting aquifer arsenic concentrations in the San Luis Valley, Colorado, an area of moderately-elevated arsenic in drinking water This study included historical water samples with total arsenic concentrations from 595 unique well locations. Spearman correlation coefficients were calculated between arsenic concentrations within the same well over three sets of time periods. Five kriging models were developed to predict groundwater arsenic concentrations; a separate validation dataset (n=51 wells) was used to identify the model with strongest predictability. Findings indicate that arsenic concentrations are temporally stable (r=0.88; 95% CI, 0.83-0.92 for samples collected from the same well 15-25 years apart) and the spatial model created using ordinary kriging best predicted arsenic concentrations (r=0.72 between predicted and observed validation data). These findings illustrate the value of geostatistical modeling of arsenic and suggest the San Luis Valley is a good region for conducting epidemiologic studies because of the ability to accurately predict variation in groundwater arsenic concentrations.