Upper washita river experimental watersheds: data screening procedure for data quality assurance

The presence of non-stationary conditions in long-term hydrologic observation networks is associated with natural and anthropogenic stressors or network operation problems. Detection and identification of network operation drivers is fundamental in hydrologic investigation due to changes in systematic errors that can exacerbate modeling results or bias research conclusions. We applied a data screening procedure to the USDA-ARS experimental watersheds data sets () in Oklahoma. Detection of statistically significant monotonic trends and changes in mean and variance were used to investigate non-stationary conditions with network operation drivers to assess the impact of changes in the amount of systematic error. Detection of spurious data, filling in missing data, and data screening procedures were applied to >1000 time series, and processed data were made publicly available. The SPELLmap application was used for data processing and statistical tests on watershed segregated data sets and temporally aggregated data. A test for independency (Anderson test), normality, monotonic trend (Spearman test), detection of change point (Pettitt test), and split record test ( and -tests) were used to assess non-stationary conditions. Statistically significant (95% confidence limit) monotonic trends and changes in mean and variance were detected for annual maximum air temperature, rainfall, relative humidity, and solar radiation and in maximum and minimum soil temperature time series. Network operation procedures such as change in calibration protocols and sensor upgrades as well as natural regional weather trends were suspected as driving the detection of statistically significant trends and changes in mean and variance. We concluded that a data screening procedure that identifies changes in systematic errors and detection of false non-stationary conditions in hydrologic problems is fundamental before any modeling applications.