USGS Home
Contact USGS
Search USGS

U.S. GEOLOGICAL SURVEY
Scientific Investigations Report 2009–5015

Back to Table of Contents

Evaluation of Perennial Streams Model

The perennial streams model was evaluated by comparing classifications at various stream points on the map with four independent datasets from government agencies. A brief overview of these datasets is provided in table 3. Some stream classifications in the comparison datasets were based on single site visits and may not be representative of conditions throughout the year or over several years. Additionally, classification methods differed slightly among datasets. The locations of comparison points for each of the datasets are shown in figure 7.

The Beneficial Use Reconnaissance Program (BURP) dataset was provided by the IDEQ. The purpose of BURP is to help Idaho regulators meet the requirements of the Clean Water Act by collecting data for determining the existing uses and beneficial use support states of Idaho’s water bodies (Idaho Department of Environmental Quality, 2008). Field visits were conducted in summer and autumn each year to collect aquatic community samples and aquatic habitat information. Field crews measured flow at a site if it was not dry. Sites were coded intermittent if dry or measured flow was less than 0.1 ft³/s; otherwise, the site was coded as perennial.

The Environmental Monitoring and Assessment Program (EMAP) datasets were provided by the USEPA. EMAP was developed to monitor and assess the status and trends of national ecological resources with a focus on forecasting future risks to those resources (U.S. Environmental Protection Agency, 2008). EMAP datasets used in this comparison are “EMAP Office” and “EMAP Field”. USEPA personnel initially evaluated candidate stream sites using River Reach File 3 (RF3) (U.S. Environmental Protection Agency, 1994) and NHD maps and stream classifications. The EMAP Office dataset included stream sites that were coded as nonperennial on RF3/NHD maps and stream sites that were coded perennial but were not considered candidates for field visits because of limited access or other reasons. Streams in this dataset were then coded as perennial or intermittent using several office-based approaches, including examining maps and GIS coverages and talking to regional and state scientists with knowledge of the area. The EMAP Field dataset included sites that were coded as perennial on RF3/NHD maps and were deemed candidates for further field verification. After field visits, sites were coded perennial if there was enough water present to sample, even if flow was not continuous at the time of the visit. Otherwise, the site was coded intermittent. Field visits were made during the summer/autumn baseflow period, which was expected to be the minimum flow period for any given year.

The USGS Low Flow dataset includes stream sites that are measured periodically as part of the USGS Idaho Water Science Center partial record network used to supplement long-term streamflow-gaging records when generating peak and low flow statistics. Of the master partial record network database, sites were included in the comparison with the USGS perennial streams map if the site was dry, water was pooled, or measured flow was less than 0.1 ft³/s (to be comparable with IDEQ BURP sites). All sites in this dataset are considered intermittent because they were observed to go dry or nearly dry annually. No attempt was made by the field hydrographer making the streamflow measurement to classify the stream as intermittent or perennial.

The comparison datasets were overlaid on the USGS perennial streams model. Comparison sites were forced to snap to the nearest USGS modeled stream (perennial or transitional), and coding of the comparison site was compared with the coding of the nearest USGS modeled stream. Sites were considered “agreements” or “disagreements” with the model based on the rules presented in table 4. All sites that were initially coded “disagreements” were checked to verify whether the comparison point snapped to the correct stream. In many cases, the comparison point was on a NHD stream not modeled by the USGS (for example, a true intermittent stream) but snapped to the closest USGS-modeled perennial stream. Once the correct location of the comparison point was verified, these cases were considered agreements with the model, because the USGS model was intended to include only perennial and transitional stream reaches. Comparison data points were removed from the analysis if their coordinates appeared grossly incorrect and no accompanying description was available to determine the correct location of the site. In addition, about 20 percent of the sites that were coded as agreements were checked to ensure that the point snapped to the correct USGS-modeled stream. Codings were revised if necessary.

A classification table (table 5) shows a comparison between the USGS model and comparison datasets. Overall, 81 percent of the comparison data points agreed with the USGS model according to the agreement rules shown in table 4. In table 5, the classification categories that are considered “disagreements” are: Observed intermittent stream―Predicted perennial stream (417 sites) and Observed perennial stream―Not mapped (4 sites). A comparison of each dataset with the USGS perennial streams model is presented by region in table 6. The total number of sites available for comparison and the percentage of total sites that agreed with the USGS model are shown for each dataset.

As a whole, the USGS model compared well with comparison datasets. In most of the disagreements, the USGS model overestimated the amount of perennial streams, meaning that many streams were predicted perennial when comparison datasets classified them as intermittent. Some disagreements were expected due to varying quality of site location coordinates (latitude/longitude), timing of site visits during unusually wet or dry years, discrepancy between 1-day site visits and a 7-day low flow criterion, and localized contributions of ground water to flow in some areas.

The USGS perennial streams model was generated based on long-term streamflow-gaging records and, with the inclusion of a transition zone upstream of the perennial stream endpoint, was expected to represent a range of climatic conditions. However, if a comparison dataset was generated solely during an unusually wet or dry period, a high number of disagreements with the model are expected. To evaluate climatic representativeness, the average Palmer Drought Severity Index (PDSI) for Idaho was examined for each year in the comparison dataset. PDSI generally ranges from -6 (extreme drought) and +6 (extreme wet), centering on 0 as a “normal” condition (Palmer, 1965).

The EMAP Office dataset compared better with the USGS model than the EMAP Field dataset (table 6). This was expected because the EMAP Field dataset was based on a single site visit, and the EMAP Office dataset was generated from professional opinion based on overall long-term knowledge of the area and map resources. Additionally, the EMAP Field dataset site visits were conducted during 2000–2003, when the statewide PDSI ranged from -1 to -4 (Cook and Krusic, 2008), an unusually dry period, and a higher number of disagreements were expected.

The highest number of disagreements with the USGS model was observed in the USGS Low Flow Network dataset. Most disagreements probably are due to poor latitude/longitude coordinates during some periods in the dataset and the timing of site visits. Because the dataset spans about 6 decades, global positioning system (GPS) receivers were not available to measure site latitude and longitude for the earlier measurements. Many coordinates were pulled from maps, which may not have been as accurate as GPS, and many of these comparison points did not include a stream name or description. In addition, the accuracy of GPS coordinates from March 1990 to May 2000 was hampered by selective availability, an intentional GPS signal degradation feature implemented by the U.S. Department of Defense for national defense purposes (National Executive Committee for Space-Based Positioning, Navigation, and Timing, 2008). Selective availability introduced slowly changing random errors of as much as 100 m in publicly-available GPS navigation signals (U.S Department of Commerce Technology Administration, 2000). Eight comparison points were measured during this period, and if site coordinates were poor, the location as plotted may be wrong and may indeed be located on a non-modeled (intermittent) part of a stream or entirely on another stream.

Like other comparison datasets, stream classification in the USGS Low Flow Network dataset was made based on a single assessment, not a 7-day duration of flow, and no attempt was made by hydrographer to classify the stream when in the field. Stream classification in this dataset was made solely for comparison with the USGS perennial streams model. Also, most disagreements occurred during years when the statewide PDSI was negative, indicating dry and, in some years, drought conditions. Of the sites that were considered disagreements, 80 percent were visited during years when the PDSI was negative according to Cook and Krusic (2008). About 25 percent of all site visits in this dataset were made in 1977, when the statewide PDSI ranged from -6 to -2 (Cook and Krusic, 2008). These measurements were made during abnormally dry years and resulted in an intermittent classification when, during years of normal precipitation, the stream may flow year-round. Mean drainage area for the sites considered disagreements was relatively high (38.5 mi²) in comparison with sites considered agreements (7.4 mi²). Several opportunistic measurements were made at sites with drainage areas as large as 517 mi² during years of drought, resulting in disagreements with the model. Under normal hydrologic conditions, most of these sites were expected to agree with the model.

The BURP dataset represented a range of climate conditions and included the highest number of comparison sites. Statewide average PDSI ranged from -3.5 to 4.5 during the study period, 1993–2006 (Cook and Krusic, 2008). Eight years during the study period were considered “wet” years (positive PDSI) and six years were considered “dry” years (negative PDSI). The number of sites visited each year was fairly constant during the study period; therefore, the BURP dataset is considered the most representative dataset for comparison, and it agrees well with the model (85 percent agreement overall).

Overall, regions with the highest number of disagreements were 1, 5, 6, and 8, which have a high percentage of mountainous and forested area. These regions also include many areas of rapid transition from mountains to expansive valleys. In these areas, the USGS model over-predicted the number of perennial streams in comparison with the NHD and independent datasets. Highly variable ground water and surface water interactions, including streamflow losses, are documented in these regions in Donato (1998), Kahle and Bartolino (2007), and Skinner and others (2007). Streamflow losses to ground water along mountain fronts, where an interface can form in the surficial geology between residuum and more-permeable colluvium and alluvium, are well documented (Niswonger and others, 2005; Covino and McGlynn, 2007; Foster, 2008). Streams can change from perennial to intermittent at this interface, often called a mountain front recharge zone (Covino and McGlynn, 2007). The USGS model does not account for these site-specific hydrologic conditions and may not adequately represent true conditions in stream reaches where ground water gains and losses are highly variable.

Regions with the highest agreements were 2, 3, 4, and 7; all these regions except region 2 have a high percentage of low gradient, low elevation area and fewer mountain front recharge zones. However, most of the comparison sites in region 2 were in the lower gradient areas around Lake Coeur d’Alene and Lake Pend Oreille. These areas were expected to have more consistent ground water contributions to streamflow and less variable surficial geology, resulting in more consistent low flow patterns and a higher percentage of perennial streams. Surprisingly, the highest standard errors of prediction were for regions 3 and 7 in their respective regression equations; an explanation for these high standard errors is provided in the section titled “Model Revisions Based on Spatial Patterns”.

Back to Table of Contents