Use of boosted regression trees to quantify cumulative instream flow resulting from curtailment of irrigation in the Sprague River basin, Oregon

A boosted regression trees (BRT) approach was used to estimate the amount by which streamflow is increased when irrigation is regulated (curtailed) upstream of a streamgage on the Sprague River in southern-central Oregon. The BRT approach differs from most other approaches that require baseline conditions for comparison, where those baseline conditions are determined from past observations by searching for hydrologically similar years when irrigation was not regulated. Such baseline conditions are always imperfect estimates of the true baseline conditions. The BRT approach instead estimates unique baseline conditions for any year in which irrigation is regulated by calculating the baseline condition based on measurements of precipitation and weather observations that determine evapotranspiration, and other measurements that are proxies for the effects of climate and regional groundwater pumping on water-table elevation, using a model that has been trained in years of no regulation. The amount by which streamflow is increased by regulation is then calculated by subtracting the estimated baseline conditions from the measured streamflow. The approach is challenged by the fact that the streamflow increase may be a small fraction of the total streamflow; nonetheless, during 2 years in which regulation was started early and was implemented consistently through the season, the increased flow made up about one third of the flow past the streamgage during the regulation period. An advantage of this approach is that with rigorous model testing with holdout data, the threshold for detecting streamflow increase and intervals around the estimates of increase at a desired level of confidence can be quantified. The model relies on datasets that are readily available and updated continuously and therefore can be used operationally to inform resource management.