Evaluation of multiple regression models using spatial variables to predict nitrate concentrations in volcanic aquifers

Multiple linear regression of spatial variables including land use, soil type, and topography was applied to predict nitrate concentration and evaluate major factors affecting nitrate occurrence in springs and wells in the southern and northern areas of Jeju volcanic island, Korea. Three types of contributing area surrogates (CAS), namely circle, semicircle, and wedge, were employed to calculate the spatial variables. The regression results showed R² of 0.81–0.84 for springs and 0.74–0.77 for wells; R² values for wedge and semicircular CAS were more than 10% higher than those for circular CAS. The R² of spring models was significantly affected by both the shape and size of CAS, with optimal radii of 150–250 m and 300–400 m in the southern and northern areas, respectively, corresponding to thinner upper basaltic aquifers, and implying shorter flow paths in the southern area. The most influential variables in springs were orchards and soil types related to agriculture including silty loam and silty clay loam, indicating that nitrate levels are strongly affected by N fertilization in cultivated areas. In contrast, wells showed much less sensitivity to both shapes and sizes of CAS, with less contribution of land use and soil type to the regression, which could be attributed to a mix of multiple aquifer zones and widely different factors in the installation and operation of wells. Field parameters of electrical conductivity (EC) and pH increased the R² up to 10%, suggesting that these can be useful when regression with spatial variables yields a lower R². The optimal spatial scales for prediction of nitrate concentration and spatial variables that significantly contribute to nitrate contamination can provide relevant criteria for establishing groundwater management policies, considering the increasing anthropogenic land‐use trends on the island, where groundwater is highly sensitive to changes in spatial variables. Copyright © 2015 John Wiley & Sons, Ltd.