Development of regression equations to estimate flow durations, low-flow frequencies, and mean flows at ungaged stream sites in Connecticut using data through water year 2022

Elizabeth A. Ahearn; Gardner C. Bent

doi:10.3133/sir20255027

Home/Archive/USGS

USGS · sir20255027

Development of regression equations to estimate flow durations, low-flow frequencies, and mean flows at ungaged stream sites in Connecticut using data through water year 2022

Elizabeth A. Ahearn·Gardner C. Bent

View original source ↗Read source document ↗Export metadata ↓

Abstract

To aid Federal and State regulatory agencies in the effective management of water resources, the U.S. Geological Survey, in cooperation with the Connecticut Department of Energy and Environmental Protection and the Connecticut Department of Transportation, updated flow statistics for 118 streamgages and developed 47 regression equations to estimate selected flow duration, low flow, and mean flow statistics for the entire State of Connecticut, for the following: 1-, 5-, 10-, 25-, 50-, 75-, 90-, 99-percent flow durations; 7-day, 10-year low-flow frequency and 30-day, 2-year low-flow frequency; and mean flow, spring mean flow, and harmonic mean flow. In addition, regression equations were developed for monthly and seasonal flow durations, ranging from 25 to 99 percent for aquatic biological processes of salmonid spawning (November), overwinter (December–February), clupeid spawning (May), resident spawning (June), and rearing and growth (July–October) periods, and for flow durations ranging from 1 to 99 percent for the habitat forming (March–April) period. Statistics were derived from daily mean streamflow data collected from streamgages with at least 10 years of data through water year 2022 in southern New England and eastern New York. Forty streamgages in Connecticut and adjacent areas of neighboring States were used in the regression analysis. Regression methods of weighted least squares and generalized least squares were used to derive the final coefficients and measures of uncertainty for the regression equations. The equations used to estimate selected streamflow statistics were developed by relating the flow statistics to different basin characteristics (physical, land cover, and climatic) at the 40 streamgages. Nine basin characteristics served as the explanatory variables in the statewide regression equations: drainage area, percentage of area with coarse-grained stratified deposits, stream density, mean basin slope, mean basin elevation, percentage of area with hydrologic soil group A, mean monthly precipitation for November, mean seasonal precipitation in the winter (December, January, and February), and mean annual temperature. The root mean square error of the 47 equations ranged from 7.9 to 121.9 percent, with an average of 27.9 percent. The equations estimate flows most accurately near the mean (50-percent flow duration), become less accurate for low flows, and are the least accurate for extreme low flows. The root mean square error for the 50-percent flow duration is 15.1 percent, with an average of 17.6 percent across the six periods. The extreme low flow statistics of 7-day, 10-year low-flow frequency, 99-percent flow duration, and 99-percent rearing and growth period flow durations have root mean square errors of 121.9, 105.1, and 121.9 percent, respectively. The adjusted coefficient of determination of the 47 equations ranged from 73.4 to 99.5 percent, with an average of 95.1 percent.

Explore related subjects

Scientific Investigations Report Report

Source-reported bounding extent: 40.985171° to 42.050587° latitude; -73.727775° to -71.786994° longitude. This indicates report coverage, not an exact sampling location. View area on OpenStreetMap.

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Elizabeth A. Ahearn, Gardner C. Bent. 2025-07-23. Development of regression equations to estimate flow durations, low-flow frequencies, and mean flows at ungaged stream sites in Connecticut using data through water year 2022. https://doi.org/10.3133/sir20255027

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related USGS reports

USGS2026

Karst hydrologic processes in carbonate fensters of the western Great Smoky Mountains, Tennessee, 2017–18

The geology of the Great Smoky Mountains in Tennessee and North Carolina is dominated by metamorphic strata. Along the western extent of the range, however, several carbonate fensters expose the Ordovician Jonesboro Limestone along the western edge of the Great Smoky Fault. The fensters (Cades Cove, Tuckaleechee Cove, and Wear Cove) create opportunities for allogenic recharge at the contact between the surrounding, relatively insoluble, strata and the underlying soluble formations, resulting in the development of karst systems. Karst environments have a high degree of connectivity between surface streams and groundwater systems; thus, karst aquifers are inherently vulnerable to contamination from surface activities. Karst environments also provide a specific environment for sensitive biota. Protection of the unique resources of karst environments requires an understanding of the behavior of karst aquifers, knowledge of karst recharge areas, and the mapping of groundwater flow paths. A study began in 2017 to characterize the hydrologic behavior of the karst systems in Great Smoky Mountains National Park through hydrologic monitoring, dye tracing, and seepage investigations. Hydrologic monitoring equipment was installed at several karst hydrogeologic features: two in-cave sites, two major springs, Abrams Creek in Cades Cove, and a karst depression wetland. Dye tracing was conducted to determine resurgences for the cave systems, to delineate spring recharge areas, and to estimate karst groundwater velocities. Seepage investigations along Abrams Creek helped determine if streams were sinking into the Jonesboro Limestone, either at the contact with insoluble strata or as streams flowed across the carbonate fenster. Study results indicate considerable differences in groundwater behavior between the conduit-dominated flow systems in Tuckaleechee Cove compared to the lower-gradient, colluvium-covered karst systems in Cades Cove. Study results also indicate that contaminants in the area of Bull sink and Whiteoak Sink could be rapidly transported through karst groundwater towards springs in Tuckaleechee Cove. Karst resources in Cades Cove are most likely to be affected by local events or activities taking place within the cove itself or in upper Abrams Creek upstream of the Great Smoky Fault. This study provides information on the vulnerability and uniqueness of the karst systems in the western Great Smoky Mountains, which are characterized by complex geology, high-gradient subterranean streams, deep caves, and carbonate fensters.

Tennessee↗

USGS2026

Analysis of groundwater geochemistry in the Lansing, Michigan, area

Groundwater resources in the Lansing, Michigan, area are protected by a wellhead protection plan, but previous studies have indicated that groundwater in this area may be susceptible to surface contamination. The U.S. Geological Survey, in cooperation with the Lansing Board of Water & Light, analyzed geochemical and aquifer characteristics to assess the potential for contamination of groundwater resources. Recharge age categories were determined from tritium concentrations for some wells, and by using ternary diagrams and principal component analysis for other wells. These age classifications, combined with hydrogeologic and geochemical data, were analyzed by using statistical methods to identify trends that may indicate areas of increased groundwater susceptibility. Wells with modern recharge tended to have higher sodium, chloride, and chloride:bromide ratios than did wells with mixed recharge, suggesting greater surface connectivity and potential influence from anthropogenic sources such as road salt. In contrast, wells in the mixed age category exhibited more stable geochemistry and lower constituent concentrations. Classification and regression tree models identified aquifer geometry—particularly the thickness of the uppermost bedrock unit and presence of confining layers—as key predictors of recharge age and geochemical indicators. The trends identified from these analyses could help to reveal locations in the Lansing area that are among the most susceptible to contamination.

Michigan↗

USGS2026

Evaluating the number of realizations (train-test splits) necessary to ensure robust estimates of conventional hydrothermal resource favorability using Monte Carlo cross-validation, the Great Basin, Western USA

Supervised machine learning approaches used to produce favorability maps in resource assessments commonly partition an input dataset into training and testing subsets (that is, through train-test splits). However, if a single split is not representative of underlying resource conditions, the resulting favorability map may be biased, particularly with sparse or heterogeneous datasets. To mitigate the potential for unrepresentative favorability maps, recent works by U.S. Geological Survey researchers have used Monte Carlo cross-validation to implement a large number ( n) of randomized train-test splits to produce n realizations. The realizations are then aggregated using cell-wise median predictions to produce the resulting final favorability map, recognizing that the median will remove extreme values that might have been estimated from a comparatively few unrepresentative realizations. Theoretically, the median prediction at every cell converges to a single estimate as n approaches infinity, and the question addressed herein is: How many realizations are necessary to get a robust estimate (that is, an estimate unaffected by additional realizations)? Therefore, convergence metrics are described to identify the number of train-test splits and realizations needed to make a reliable map using the data and supervised machine learning workflow for conventional hydrothermal resource assessment of the Great Basin of the Western United States. For this study, robust prediction occurs with n approximately equal to 600 realizations, and smaller ensembles (for example, n is approximately equal to 200 or even 100) may be suitable during preliminary or exploratory analyses. Although the resulting threshold numbers will vary depending on the data and modeling approach used, the convergence metrics proposed and applied herein are applicable to other supervised machine learning studies that use Monte Carlo cross-validation.

California, Idaho, Nevada, Oregon, Utah↗