Investigating distribution of nitrate concentration using ensemble nonparametric quantile regression

Abstract Nitrate ( NO3−) pollution in groundwater is a major concern due to its negative health effects; therefore, accurately estimating and predicting the  NO3− concentration in groundwater is necessary. The  NO3− concentration distribution can be used to find less polluted areas, and these identified areas can be candidates for drinking water resources. We considered a total of 14,297  NO3− concentration observations in South Korea. Altitude, slope, land use, hydrogeological unit, and surface soil texture data were also collected to assess the covariates that affect  NO3− concentration levels. Sample quantiles display nonlinear patterns based on these covariates. Thus, we propose using an ensemble nonparametric quantile regression approach to determine the  NO3− concentration distribution. The proposed approach is a data-driven method that ensembles nonlinear quantile models to capture the complex relationships between quantiles of the response variable and the covariates while controlling the computational complexity, which are advantages over non-ensemble quantile regression methods. The validation study demonstrates that the proposed method exhibits a smaller loss value compared to that of the non-ensemble models considered for comparison. We investigated lower quantile maps of  NO3− concentration (5% and 10%), as we are interested in less polluted areas. The proposed model attempts to reach the sample quantile values while a non-ensemble nonlinear model with altitude and slope does not except land use is coniferous at low altitude. We created a proportional map with less polluted areas at the district level using the proposed approach. This provided us with the top five districts (Samcheok City, Geochang County, Yanggu County, Inje County, and Yeongwol County) with the highest proportions of less-polluted areas in South Korea.