How reliable are standard indicators of stationarity?

In modern hydrological practice large confidence is placed on modelling results that are used for planning and design. This is especially the case where the modelling results have been carefully verified against independent data. An underlying assumption of the calibration/verification process is that the whole data series is stationarity. Standard parametric and non-parametric tests are available for examining the stationary of hydrologic time series but it has been shown here that these may be inadequate for that purpose unless applied with care. Annual, seasonal, monthly and daily time series of precipitation and climate data were examined considering parts of the series formed using sequential windows. Seven standard parametric and non-parametric tests were applied to these relatively long series and while it was shown that some tests suggested that all series were stationary, most series were shown to be non-stationary in more than one of the tests, some of them at very high levels of significance. This apparently hidden non-stationarity could have very large effects on water resources modelling. These effects would have considerable influence in calibration and verification of models and in simulation of long series of water resources characteristics and could be especially important as the effects of climate change become more pervasive.