Estimating Locations of Perennial Streams in Idaho Using a Generalized Least-Squares Regression Model of 7-Day, 2-Year Low Flows

Abstract

Many State and Federal agencies use information regarding the locations of streams having intermittent or perennial flow when making management and regulatory decisions. For example, the application of some Idaho water quality standards depends on whether streams are intermittent. Idaho Administrative Code defines an intermittent stream as one having a 7-day, 2-year low flow (7Q2) less than 0.1 ft3/s. However, there is a general recognition that the cartographic representation of perennial/intermittent status of streams on U.S. Geological Survey (USGS) topographic maps is not as accurate or consistent as desirable from one map to another, which makes broad management and regulatory assessments difficult and inconsistent. To help resolve this problem, the USGS has developed a methodology for predicting the locations of perennial streams based on regional generalized least-squares (GLS) regression equations for Idaho streams for the 7Q2 low-flow statistic. Using these regression equations, the 7Q2 streamflow may be estimated for naturally flowing streams in most areas in Idaho. The use of these equations in conjunction with a geographic information system (GIS) technique known as weighted flow accumulation allows for an automated and continuous estimation of 7Q2 streamflow at all points along stream reaches. The USGS has developed a GIS-based map of the locations of streams in Idaho with perennial flow based on a 7Q2 of 0.1 ft3/s and a transition zone of plus or minus 1 standard error. Idaho State cooperators plan to use this information to make regulatory and water-quality management decisions. Originally, 7Q2 equations were developed for eight regions of similar hydrologic characteristics in the study area, using long-term data from 234 streamflow-gaging stations. Equations in five of the regions were revised based on spatial patterns observed in the initial perennial streams map and unrealistic behavior of the equations in extrapolation. The standard errors of prediction for the final equations ranged from a minimum of +75.0 to -42.9 percent in the central part of the study area to a maximum of +277 to -73.5 percent in the southern part of the study area. The equations are applicable only to unregulated, naturally-flowing streams and may produce unreliable results outside the range of explanatory variables used for equation development. Extrapolation outside the range of available data was necessary, however, to predict perennial flow initiation points and transition zones along stream reaches. The map of perennial streams was evaluated by comparing predicted stream classifications with four independent datasets, including field observations by other government agencies. Overall, 81 percent of the comparison data points agreed with the USGS perennial streams model. Regions with the highest number of disagreements had a high percentage of mountainous and forested area with potential mountain front recharge zones, and regions with the highest agreements had a high percentage of low gradient, low elevation area. As a whole, the USGS model predicted a higher number of perennial streams than predictions made with the independent datasets. Some disagreements were due to poor site location coordinates, timing of the comparison site visits during unusually wet or dry years, discrepancies in classification criteria, and variable ground water contributions to flow in some areas. The Idaho Department of Environmental Quality Beneficial Use Reconnaissance Program (BURP) dataset is considered the most representative dataset for comparison because it covered a range of climate conditions and the number of sites visited were consistent from year to year during the study period. Eighty-five percent of BURP comparison data points agreed with the USGS perennial streams model. Although site-specific flow data may be needed to correctly classify streams in some areas, this information rarely is available and is not always practical to o