Screening for long-term trends in groundwater nitrate monitoring data

Abstract

A large body of UK groundwater nitrate data has been analysed by linear regression to define past trends and estimate future concentrations. Robust regression was used. The datasets showed too many irregularities to justify more traditional time-series approaches such as ARIMA-type methods. Tests were included for lack of linearity, outliers, seasonality and a break in the trend (by piecewise linear regression). Of the series analysed, 21% showed a significant improvement in the overall fit when a break was included. Half of these indicated an increase in trend with time. Significant seasonality was found in about one-third of the series, with the largest nitrate concentrations usually found during winter months. Inclusion of nearby water-level data as an additional explanatory variable successfully accounted for much of this seasonality. Based on 309 datasets; from 191 distinct sites, nitrate concentrations were found to be rising at an average of 0.34 mg NO3 I-1 a-1. In 2000, 34% of the sites analysed exceeded the 50 mg I-1 EU drinking water standard. If present trends continue, 41% could exceed the standard by 2015. We explored an alternative to the previously proposed Water Framework Directive aggregation approach for estimating trends in whole ground-water bodies (the 'Grath' approach: spatially average then find the trend). We first determined the trends for single boreholes and then spatially averaged these. This approach preserves information about the spatial distribution of trends within the water body and is less sensitive to 'missing data'. © 2007 Geological Society of London.