Statistical approaches to analysing trends in groundwater quality

A trend is the underlying rate of change and is often used to distinguish a long-term tendency from erratic short-term fluctuations (noise). Groundwater quality varies with time over various timescales from daily seasonal and annual, superimposed on each other. Trend estimation is complicated by cyclic behaviour, step changes and data censoring as well as short-term fluctuations. It is also dependent on the dataset characteristics, the sampling frequency, monitoring period and regularity of sampling. The increasing quality of the dataset allows more sophisticated tests. Drivers for trend assessment include quantification of trend reversal under the Water Framework Directive, prediction of peak concentrations to enable water utilities to meet drinking water regulations, the impact of climate change and environmental impact assessments. A semi-automated linear regression methodology has been developed to process irregular water quality timeseries using the ‘R’ statistical software. This was used to analyse a large dataset of groundwater nitrate data from England to define past trends and to make estimates of future concentrations. Tests were included for lack of linearity, outliers, seasonality and breaks in the trend. The method provides annotated graph with quantitative estimates of trends and provides warnings of possible departures from the underlying assumptions. The method was not useful for series with step changes and excursions. 21% of the series analysed showed a significant improvement in the overall fit when such a break was included and half of these indicated an increase in trend with time. An assessment of seasonality in nitrate concentrations was also made by including a term for the month of sampling in the regression model. Significant seasonality was found in about one third of the series. In 2000, 34% of sites analysed exceeded the 50 mg/L standard. If present trends continue, 41% of groundwater sources could exceed the standard by 2015. Nitrate concentrations can be regressed to covariables, such was water level, to allow the prediction of both trends and seasonal peak concentrations, but cannot be used to predict the impact of changes. Long-term regular monitoring is the key to successful trend estimation. Purpose-built monitoring boreholes suffer less from operational disturbances