Investigation of recent decadal-scale cyclical fluctuations in salinity in the lower Colorado river

Abstract

Beginning in the late 1970s, 10- to 15-year cyclical oscillations in salinity were observed at lower Colorado River monitoring sites, moving upstream from the international border with Mexico, above Imperial Dam, below Hoover Dam, and at Lees Ferry. The cause of these cyclical trends in salinity was unknown. These salinity cycles complicate the U.S. Bureau of Reclamation's (Reclamation) responsibility for managing salinity in the river for delivery of water to Mexico to meet treaty obligations. This study develops a conceptual model of the salinity cycles from time-series water quality, streamflow, and precipitation data in both the lower and upper Colorado River Basins in order to provide Reclamation the ability to understand, anticipate, and manage future salinity cycles in the lower river. Compared with the Lees Ferry record, both maximum and minimum salinity levels increase downstream by about 25% at Hoover Dam, by about 49% at Imperial Dam, and by about 69% at the northern international boundary with Mexico. In the upper basin, cyclical salinity trends are evident at the outflow of three major tributaries, where salinity is also noted to be inversely related to streamflow. Time series trends in precipitation within the catchments of the three upper basin tributaries indicate cyclical periods with above normal precipitation and periods with below normal precipitation. Periods of greater than normal precipitation in the contributing areas correspond with declines in salinity at the catchment monitoring sites and periods of less than normal precipitation correspond with rising salinity at the sites. Based on the conceptual model developed in this investigation, a multiple linear regression model was developed using a stepwise variable selection procedure to simulate salinity in Lake Powell inflow. Important variables in the explanation of salinity entering Lake Powell include flow from the three upper basin tributaries, seasonality, and mean precipitation in the upper basin, among others. The root mean square error of prediction for the MLR model was 31.48 mg/L (5.7%).