Affordable Access

Predicting Electrical Conductivity in the South Delta using Multivariate Regression

Center for Watershed Sciences, UC Davis
Center for Watershed Sciences John Muir Institute of the Environment


SWRCB Contract 06-447-300 A thorough statistical analysis was performed on a set of data provided by the Water Board to determine the strength of the results presented in the Draft Technical Report for Alternative San Joaquin River Flow and Southern Delta Salinity Objectives (SWRCB 2011). The Water Board had performed a single variable regression on each of three southern Delta locations against the salinity at Vernalis. The locations were Brandt Bridge on the San Joaquin River and Union Island and Tracy on the Old River with R-squared values from the regressions of 0.93, 0.96 and 0.75 respectively. The approach used here was to first duplicate the regression results in the Water Board report through linear regressions on monthly data and then to examine several possibilities to determine if there was a better relationship possible. The methods included multivariate regressions on the daily and monthly-averaged data, multivariate regressions using a transformation of the independent variable that suggested a non-liner relationship (flow at Vernalis), and several multivariate regressions with several data subsets. The subsets included those with and without agricultural barriers, with and without the fish barrier, with and without all barriers, during the winter and summer irrigation periods and finally on an individual monthly basis. A final investigation was performed to see if there was any meteorological or tidal aspect to the peak EC signals at Old River at Tracy. The simple linear regressions at Brandt and at union Island are quite strong and stand on their own. There was no improvement seen at these locations for any of the advanced calculations. The simple linear regression for Old River at Tracy is actually quite strong (0.76 R-squared), but does show some scatter about the prediction. Multivariate regressions do not improve the prediction unless subsets of the data are used. There is minor improvement in the predictions when either some or all of the barriers are not in place and when individual months are considered. While some months coincide with those without some or all barriers, exceptions occur when the barriers are always out from December \textendash March. The regression is poorest in November, just before the all barriers out period, and improvement occurs all through the 4-month period when barriers are out. While there is evidence that some recent spikes in salinity at Tracy occurred at the end of Spring tides when the estuary is draining, there was no corresponding regression evidence that this is the case.

Report this publication


Seen <100 times