Predicting mean annual and mean monthly streamflow in Colorado ungauged basins

Abstract

River managers often need estimates of streamflow for ungauged streams. These estimates can be used in water rights acquisitions, in-stream flow management, habitat assessment, water quality planning, and stream hazard identification. This publication describes new regression models for predicting mean annual and mean monthly streamflow in Colorado. Unlike previous regional regression studies, the new models incorporate snow persistence (SP), the fraction of time a watershed remains snow covered. Models were developed using streamflow data from 131 watersheds with drainage areas <1,500 km 2 , no transbasin diversions, and <10% urban area. In addition to SP, topographic, climate, geologic, and hydrologic region variables were used in model predictions. All new models had very good performance, with <6% absolute bias and stronger performance compared to current regional regression models in StreamStats. The mean annual model had the strongest performance, with Nash-Sutcliffe efficiency coefficient (NSE) of 0.93 and <2% absolute bias. Mean monthly models had best performance during snowmelt runoff months (May-Jul; NSE ≥0.79; absolute % bias ≤ 4) and weaker performance during low flow months (Aug-Apr; NSE ≥ 0.59; % bias ≤ 5). Tests of the mean annual model using decadal average streamflow from 1910s to 2000s show very good performance (NSE > 0.75), but predictions were biased low by 14-28% in wetter decades. All equations and coefficients needed to run the models are presented in the publication appendix, and the associated data release includes the spatial data and model code, which can be applied using R or within an R-based Shiny web app.

On the hydrological difference between catchments above and below the intermittent‐persistent snow transition

“…Scatter plots of annual discharge ( Q , mm, log‐scale) versus (a) precipitation ( P , mm), (b) snow persistence (%), (c) aridity index ( P /PET) and (d) soil moisture given as volumetric water content (%) at 5 cm depth, for each of the study catchments. 'Other' sites refer to nearby catchments with streamflow data from USGS or the Colorado Division of Water Resources (Eurich et al, 2021, Supporting information Table S1). Grey error bars in (a) indicate a 50% under‐catch range for precipitation on the x ‐axis and a range of ± 50% for Q …”

“…SP was determined as the fraction of the MODIS 8‐Day images with snow present from January 1 to July 3, a period that 'brackets the extent of peak snow accumulation to complete snow ablation in most parts of the western United States' (Moore et al, 2015). We did not use any fractional snow‐covered area products (fSCA) because SP integrates snow cover over space (watershed area) and time; at these aggregated scales, previous analyses successfully related the MODIS 8‐day binary maximum snow‐covered extent to streamflow responses (Eurich et al, 2021; Hammond et al, 2018; Sexstone et al, 2020).…”

“…We conducted the same analysis for 23 other gauged watersheds in the study region (Eurich et al, 2021). The gauges of these watersheds ranged from 1957 to 3422 m.…”

Groundwater‐Mediated Memory of Past Climate Controls Water Yield in Snowmelt‐Dominated Catchments

“…Potential mechanisms underlying the increase of streamflow with higher antecedent groundwater storage include, but are not limited to, displacement of stored groundwater and the activation of faster, shallower flowpaths in higher transmissivity shallow soils when deeper storage is full, or fill‐and‐spill of subsurface reservoirs (Kampf et al., 2020; McDonnell, 2003; Tromp‐van Meerveld & McDonnell, 2006). Consistent with the importance of groundwater storage on snowmelt‐derived streamflow, statistical streamflow models that do not explicitly include antecedent storage and groundwater interactions capture average conditions well, but may be less accurate in forecasting both high and low flows (e.g., Eurich et al., 2021).…”

High Resolution SnowModel Simulations Reveal Future Elevation‐Dependent Snow Loss and Earlier, Flashier Surface Water Input for the Upper Colorado River Basin

“…In reporting the results of our analyses, we use three elevation categories to synthesize patterns in SWE and SWI change by elevation: "high," >3,300 m; "mid," 2,300 m < elevation ≤ 3,300 m; "low," ≤2,300 m. These elevation categories were chosen because they broadly reflect the satellite-derived boundaries of the present intermittent and seasonal snow zones in the study region, zones that reflect similar snow patterns (Moore et al, 2015) and similar patterns of soil moisture and streamflow response to snowmelt (Hammond et al, 2018;Harrison et al, 2021). These elevation categories are also consistent with regional regression equation elevation ranges for streamflow prediction in ungaged areas (Capesius & Stephens, 2009;Eurich et al, 2021).…”