Rating stability, and frequency and magnitude of shifts, for streamgages in Virginia through water year 2013

The U.S. Geological Survey, in cooperation with the Virginia Department of Environmental Quality, has quantified several measures of rating stability and the frequency and magnitude of changes to ratings through time for 174 real-time continuous streamgages active in Virginia as of September 30, 2013. Generalized additive models (GAMs) were fitted through all available flow measurements for all the streamgages in Virginia’s real-time network as of September 30, 2013, with at least 20 flow measurements with positive flow values. For each measurement with a positive flow value, residuals from the GAM curve were calculated. Time series of these residuals were used to identify major changes to the control (the stream feature or features which control the relation between stage and flow); the measurements in the periods of equilibrium between major changes were assigned to rating families. Of the 127 rating families that were identified as being distinct at sites, documented explanations were found for 67 of them. The most common reasons for the control to change enough to warrant a new rating family are moving the streamgage (28 times), floods (26 times), and construction activities (13 times). Provisional flow data from any streamgage that has recently experienced a major flood, regardless of historical stability, are more uncertain than usual until post-flood evidence emerges that the rating is stable, or if the rating has changed, until it is known to be well defined.

A direct comparison between provisional flow data (those data originally displayed on the web in near-real time) and flow data approved for publication following subsequent flow measurements and review could not be made because provisional flow data have not been archived. As a substitute, alternative flow (AltFlow) tables were constructed for periods with complete records of shifts and ratings. Alternate flows consist of $Q$_same, the flow value from the shifted rating table used to compute the daily flow value at the time of the most recent flow measurement that corresponds to the gage height of each day’s daily flow value, and $Q$_prev, the flow value from the shifted rating table in effect at the time of the previous flow measurement that corresponds to the gage height of each day’s daily flow value.

Several metrics that summarize AltFlow tables were computed and evaluated; particular importance was given to how well the metrics agreed with the descriptive stability class developed from interviews with hydrographers. Of these stability metrics, at least four were determined to be meaningful and to represent different aspects of control stability that might be relevant to water managers: total root mean square error between log-transformed $Q$_prev and $Q$_same, percentage of days when the difference between $Q$_prev and $Q$_same is greater than (>) $Q$_same, the sum of absolute AltFlow error divided by total flow, and percentage of days with zero difference between $Q$_prev and $Q$_same.

Three other meaningful metrics of control stability or provisional flow-data quality were computed: R² (coefficient of determination) of GAMs from the flow measurements, percentage of total estimated days, and percentage of estimated days in the winter. Correlations among metrics varied, indicating they responded to different aspects of control stability. Relations among the various stability metrics and quantitative basin and site characteristics were weak. Although quantitative relations between stability metrics and basin and site characteristics were all weak, some common patterns still emerged. Controls and ratings on large streams [>500-square mile (mi²) drainage area] and at high elevations (>1,000 feet) were more likely to be stable than controls and ratings on small streams (less than (<) 100-mi² drainage area) and at low elevations (<500 feet). There were exceptions to both generalizations, and streamgages that were intermediate in both characteristics varied widely in stability.

Typical timing of record computation changed during water years 1991–2013. From 1991 through 2001, the median number of days between the start date of the shift and the date it was created fluctuated between about 240 and about 300 days and decreased by about 4 months from 2001 to 2002. Only in 2012 and 2013 did one-half of the shifts have a delay of about 60 days between start date and final modification.