Large Snowmelt Versus Rainfall Events in the Mountains 10.1002/2014JD022753 Steven R

PUBLICATIONS

Journal of Geophysical Research: Atmospheres

RESEARCH ARTICLE Large snowmelt versus rainfall events in the mountains 10.1002/2014JD022753 Steven R. Fassnacht1,2 and Rosemary M. Records3

Key Points: 1ESS-Watershed Science, Colorado State University, Fort Collins, Colorado, USA, 2Cooperative Institute for Research in the • Large daily snowmelt equals or 3 exceeds precipitation in the mountains Atmosphere, Fort Collins, Colorado, USA, Department of Geosciences, Colorado State University, Fort Collins, Colorado, USA • Daily precipitation equals or exceeds rainfall in the mountains Abstract While snow is the dominant precipitation type in mountain regions, estimates of rainfall are used for design, even though snowmelt provides most of the runoff. Daily data were used to estimate the 10 and Supporting Information: • Readme 100 year, 24 h snowmelt, precipitation, and rainfall events at 90 Snow Telemetry stations across the Southern • Table S1 Rocky Mountains. Three probability distributions were compared, and the Pearson type III distribution yielded the most conservative estimates. Precipitation was on average 33% and 28% more than rainfall for Correspondence to: the 10 and 100 year events. Snowfall exceeded rainfall at most of the stations and was on average 53% and S. R. Fassnacht, [email protected] 38% more for the 10 and 100 year events. On average, snowmelt was 15% and 8.9% more than precipitation. Where snow accumulation is substantial, it is recommended that snowmelt be considered in conjunction with rainfall and precipitation frequencies to develop flood frequencies. Citation: Fassnacht, S. R., and R. M. Records (2015), Large snowmelt versus rainfall events in the mountains, J. Geophys. Res. 1. Introduction Atmos., 120, 2375–2381, doi:10.1002/ 2014JD022753. Extreme rainfall events are considered to be more intense than snowmelt and depth duration frequencies of rainfall or precipitation [e.g., Perica et al., 2013] and are used to predict and model floods, especially in Received 22 OCT 2014 watersheds with poor streamflow records [U.S. Soil Conservation Service, 1973]. However, in cold climates and Accepted 20 FEB 2015 at high elevations, seasonal snow accumulation magnitude and melt rate can also influence flood hazards Accepted article online 26 FEB 2015 Published online 28 MAR 2015 [Hirschboeck et al., 2000]. In the United States, some examples of high-damage snowmelt-related floods occurred on the Red River, North Dakota and Minnesota in 1997 (U.S. Geological Survey, A History of Flooding in the Red River Basin, U.S. Department of the Interior, General Information Product 55, 2007, available at http://pubs.usgs.gov/gip/2007/55/pdf/finalWebGIP55.pdf) costing US$3.5 billion in damage [Shelby, 2004] and in the Appalachian region between 1993 and 2003, with over 15 separate floods causing >US$50,000 each in damages [Graybeal and Leathers, 2006]. In areas where floods occur from more than one hydrologic process (e.g., from both rainfall and snowmelt), it may not be appropriate to group all peak flow data into a single statistical population for flood frequency analysis [Waylen and Woo, 1982]. The government standard on the precipitation frequency estimates recently published includes both rainfall and precipitation frequency estimates but not snowmelt estimates for the state of Colorado [Perica et al., 2013]. However, previous paleohydrology and streamflow studies have shown that floods at higher elevations in this region are caused by snowmelt, not rain [e.g., Jarrett, 1990; England et al., 2010]. Above a threshold estimated at about 2300 m, there is a large decrease in the magnitude of extreme rainfall [Jarrett, 1990]; intensities of snowmelt approach those of rainfall [Payton and Brendecke, 1985], and there is an abrupt transition from rainfall- to melt-dominated stream peak flows, although rainfall flood peaks are larger on a per unit area basis [Jarrett, 1990; England et al., 2010]. Incomplete understanding of these elevation thresholds and their influence on regional hydrology could have a direct effect on engineering designs and land use planning in the mountains. Should infrequent snowmelt events prove to be large, using rainfall or precipitation frequencies alone from high-elevation stations to assess flood probability might underestimate the magnitude of potential runoff events. To our knowledge, the magnitude of extreme snowmelt has not been compared to extreme rainfall or precipitation events in the Southern Rocky Mountain region of Wyoming, Colorado, and New Mexico. For flood frequency assessments, comparison of snowmelt to rainfall amounts is more relevant than comparisons of snowmelt to precipitation, because the latter can include snowfall which is not immediately available for runoff generation. However, we also analyze precipitation frequencies to allow our results to be readily compared to existing precipitation frequency guidelines [e.g., Perica et al., 2013]. In this paper, we quantify the magnitude of precipitation, rainfall, and snowmelt amounts for 24 h durations and 10 and 100 year return periods at higher-elevation locations across the Southern Rocky Mountain region

Figure 1. Ratio of (a) the 100year, 24h snowmelt to the 100year, 24h precipitation and (b) the 100year, 24h rainfall for 90 SNOTEL stations across the Southern Rocky Mountains, USA. Data points are derived using the Pearson type III distributions for 1982–2013 (described in text). Values <1.0 show that snowmelt is less than precipitation (or rainfall); values ≥1.0 show that snowmelt exceeds precipitation (or rainfall). Stations that show approximately the same ratio between Figures 1a and 1b indicate that the 100 year, 24 h precipitation is similar to the 100 year, 24 h rainfall. Stations that show a smaller ratio (warmer color) in Figure 1b than in Figure 1a indicate that the 100year, 24h rainfall exceeds the 100 year, 24 h precipitation.

to (1) compare the relative magnitudes of rainfall to precipitation events, (2) compare the relative magnitudes of snowmelt to precipitation events, and (3) compare the relative magnitudes of snowmelt to rainfall events. We are interested in water added to the system from melt and/or rain that has the potential to contribute to overland ﬂow, groundwater recharge, and/or streamﬂow.

2. Methods We used the daily time series of Snow Telemetry (SNOTEL) precipitation and snow water equivalent (SWE) data from 90 stations in the Southern Rocky Mountains, USA, for the years 1982–2013. All stations had at least 26 years of record and most a complete 32 years. Stations ranged from ~2300 to 3500 m in elevation, averaged ~3000 m, and were located on both western (48 stations) and eastern sides of the Continental Divide (42 stations) (Figure 1 and Table S1 in the supporting information). At each station and for each water year (October of the previous year through September), we calculated the following to derive three annual time series: (1) the maximum daily precipitation, (2) the maximum daily rainfall (precipitation when SWE equaled 0), and (3) the maximum daily snowmelt (decrease in SWE). Since stationarity of the data is an underlying assumption of most frequency analyses [Khaliq et al., 2006], we tested each of the annual time series for trends with the nonparametric Mann-Kendall test at the 5% significance level. Where trends were significant, we identified the slope of the trend (Sen’s slope). For annual time series where there was a significant trend for a particular station, we detrended each value in the series using the Sen’s slope and maintained the time series average. We then calculated the skew coefficient for each station for each of the three detrended annual time series. We used the Pearson type III distribution to evaluate the magnitude of precipitation, rainfall, and snowmelt events for 10 and 100 year (24 h) return periods using the skew to determine the K factor for each time series [Chow, 1951; Hoffmann et al., 1981; Interagency Advisory Committee on Water Data, 1982]. Durations of up to 24 h are of most interest for projects designed to peak flows [Perica et al., 2013]. The L moment method [Hosking and Wallis, 1997] was not used since it is utilized for regional analysis, while the present study

examined each station individually. Since the focus of this study was to compare the magnitude of snowmelt to precipitation and to rainfall, we also evaluated the effect of using different distributions (the Log-Pearson type III and the Gumbel) on the estimated magnitude of precipitation, rainfall, and snowmelt events as well as the magnitude of event ratios. The results presented here are for the Pearson type III distribution except where noted.

3. Results and Discussion Figure 2. The 10 and 100 year, 24 h rainfall versus precipitation for 90 SNOTEL stations across the Southern Rocky Mountains, USA. Data points are from There were significant decreasing Pearson type III distributions for 1982–2013 (described in text). trends in annual maximum daily snowmelt and rainfall at 31% and 6% of SNOTEL stations, respectively, and significant decreasing or increasing trends in annual maximum daily precipitation at 8% of the stations. Trends were small (À14 to À6.4 mm/decade for snowmelt, À4.8 to À2.7 mm/decade for rainfall, and À6.0 to +5.1 mm/decade for total precipitation). Detrending had little effect on the estimated events; averaged among the 90 stations, the original snowmelt annual time series 10 and 100 year events were 0.07 mm and 1.55 mm more than the detrended values, and coefficients of determination between original and detrended values were greater than 0.97 for both of the 10 and 100 year snowmelt events. Effects of detrending on rainfall and total precipitation were smaller than for snowmelt. The 10 and 100 year, 24 h precipitations were on average 33% greater (ranging from 4% to 163% more) and 28% greater (6% less to 201% more), respectively, than corresponding rainfalls (Figure 2 and Table S1 in the supporting information), indicating that many of the extreme precipitation events in the Southern Rocky Mountains are snowfall. All 10 year precipitation events were greater than rainfall events, but for 10 of the 90 stations, the 100 year, 24 h rainfall events were equal or larger in magnitude than precipitation; the latter differences are due to the statistical nature of this analysis. Jarrett and Tomlinson [2000] noted that some of the largest 24 h precipitation events in western Colorado were from snowfall. Perica et al. [2013] found that differences between precipitation and rainfall frequency estimates were nontrivial but small above approximately 1200 m in Colorado. This contrast with our findings could be due to differences in distribution fitting (e.g., Perica et al. used the three-parameter generalized extreme value distribution, with regionalized parameters) and to distinct methods used to identify snowfall events and snow water equivalent. Precipitation phase in their study was determined based on a 1.1°C temperature threshold, and SWE was estimated from recorded snow depths assuming a constant 10% fresh snow density. Snowfall densities are variable (2 to 25%) [Fassnacht, 2011], as are temperature thresholds for distinguishing snow and rain at higher elevations in Colorado (from 0°C for all snow to 9°C for all rain) [Fassnacht et al., 2013]. A 50% probability of snow at 4.5°C may be a more appropriate threshold in these areas [Fassnacht et al., 2013], which would result in lower rainfall estimates than those reported by Perica et al.[2013]. In this study, precipitation was classified as rainfall based on other observations (SWE values decreasing or zero), and SWE values were calculated by measuring the pressure exerted on snow pillows rather than estimated from a snowfall depth. As this approach does not require the assumption of constant thresholds for precipitation phase or snow density, we believe it may be more appropriate for the Southern Rocky Mountains. In general, the 10 year snowmelt was greater than the corresponding rainfall (or precipitation) at more stations and by a larger average percentage than for 100 year events at the same station. Snowmelt exceeded rainfall by more than 10% at 75 and 55 stations for the 10 and 100 year events, respectively. For these stations, snowmelt exceeded rainfall by 4.3–69 mm for the 10 year event and 6.0–83 mm for the 100 year event. Rainfall exceeded snowmelt by more than 10% at 9 and 21 stations (Figures 1b and 3b), corresponding

Figure 3. The 10 and 100 year, 24 h snowmelt versus (a) precipitation and (b) rainfall for 90 SNOTEL stations across the Southern Rocky Mountains, USA. Data points are from Pearson type III distributions for 1982–2013 (described in text).

to differences of 4.7–25 mm for the 10 year event and 7.9–63 mm for the 100 year event. On average, magnitudes of snowmelt were 53% and 38% more than rainfall (10 and 100 year events), with snowmelt to rainfall ratios ranging from 0.58 to 3.69 and from 0.44 to 3.75 for the 10 and 100 year events. The largest ratios occurred at the Fremont Pass station, where snowmelt exceeded rainfall by 77 and 112 mm, respectively. Snowmelt exceeded precipitation at 48 and 32 stations, while precipitation exceeded snowmelt at 22 and 31 stations (Figures 1a and 3a). On average, magnitudes of snowmelt were 15% and 8.9% more than precipitation, with snowmelt to precipitation ratios ranging from 0.55 to 3.09 and from 0.43 to 3.30. The magnitude of 10 and 100 year rainfall, snowmelt, and precipitation events, and the absolute differences between events, are summarized in Table 1 and presented for each station in Table S1 in the supporting information. Several other studies at individual locations in Colorado have shown large snowmelt events to be similar or greater in magnitude or in contribution to floods when compared to rainfall at elevations higher than 2300 m. Payton and Brendecke [1985] found that the 100 year intensities of observed rainfall exceeded those of modeled snowmelt at shorter intervals in the Boulder watershed, but that for longer durations (>6 h), snowmelt intensities approached those of rainfall. Jarrett and Costa [1988] noted that above the 2300 m elevation threshold in the Big Thompson River basin, rain does not contribute to flood potential and rainfall discharges per unit drainage area are very small. Kampf and Richer [2014] found that the majority of modeled runoff in the Cache La Poudre River watershed came from elevations higher than 2900 m, where there was persistent seasonal snow accumulation. The 100 year, 24 h snowmelt was at least 10% greater than precipitation at 40% of stations west of the Continental Divide and at 31% of the stations east of the Continental Divide (Figure 1a). The ratios of snowmelt to precipitation at both the northern and southern extents of the study area were generally smaller than in the central Southern Rocky Mountain region. Snowmelt was 2 times greater or more than the magnitude of precipitation at three of the 90 stations (Figure 1a), while the converse was observed at only one station. When the 100 year, 24 h snowmelt was compared to rainfall, the contrast between western and eastern sides was more pronounced, with melt exceeding rainfall by at least 10% at 73% of stations west of the Continental Divide but at only 48% of eastern stations (Figure 1b). Snowmelt was 2 times greater or more than the magnitude of rainfall at 15 of the 90 stations; these 15 were nearly equally distributed between the western side of the Continental Divide (7 stations) and the eastern side (8 stations) and were mostly located in the northern and central mountains of Colorado (Figure 1b). Only the Quemazon, New Mexico, station had twice as much rainfall (and precipitation) as snowmelt. Previous work has shown that variability in accumulated precipitation and 1 April SWE is markedly higher in watersheds with easterly or southerly aspects; the latter is an expression of the watershed’s exposure to moist airflow and capacity for precipitation development [Changnon et al., 1991]. Jarrett and Tomlinson [2000] noted that the observed maximum 24 h rainfalls in southwestern Colorado were somewhat greater than those in northwestern Colorado but much less than rainfalls of shorter durations (6 h) in eastern Colorado.

Serreze et al. [2001] found that SNOTEL stations in the western United States with large annual snowfall amounts tended to also have large individual 2.30 Rain

À snowfall events and that large snowfall and precipitation events at a particular

Precipitation- station were not necessarily apparent at surrounding stations (low coherence) in Colorado, Wyoming, and Utah. They attributed the latter phenomena to strong localized topographic controls. 135

2.65 14.8 Since a key focus of this paper is to illustrate the ratio of snowmelt to precipitation À À and rainfall, the ratios estimated using three probability distributions were Snowmelt Precipitation- compared. There were limited differences in the magnitude of the computed 100 year, 24 h events and even less difference in the 10 year, 24 h events for the three probability distributions evaluated. The 10 year, 24 h events have occurred 62.6

Rain within the period of record of most stations used within the analysis, while the À

Snowmelt- magnitudes of most 100 year, 24 h events are extrapolated from the data; therefore, smaller differences among distributions should be expected for the 10 year event than for the 100 year event results reported below. There are several sources of uncertainty in the present analysis. Here we consider the uncertainty introduced by the deﬁnition of precipitation phase (rain or snowfall), the occurrence of snowmelt, and the choice of distribution in frequency analysis. SNOTEL stations measure total precipitation without distinguishing its phase. In this study, we assumed that precipitation was rainfall when SWE equaled zero (there was no snow present) and otherwise occurred as snowfall. It is possible that rainfall could also occur during snowmelt (when SWE was decreasing), in which case, this approach could underestimate rainfall amounts. Additionally, SNOTEL does not make direct measurements of snowmelt—instead, we have calculated melt amounts as the change in SWE over a 24 h period. However, if both snowfall and snowmelt occur during this period, the total amount of water lost from the snowpack Rain Precipitation Rain Snowmelt could potentially be underestimated. To evaluate the effect of these assumptions Precipitation- on reported results, we calculated the 10 and 100 year, 24 h events again with precipitation as rain occurring when SWE was decreasing and when SWE equaled zero and deﬁning snowmelt as the change in SWE plus precipitation 6.0 13.0 79.1 64.3 81.7 17.4 79.5over 1.30 a 24 h period. 38.5 30.9 39.1 À À Snowmelt

Precipitation- The change in both classifications had only a small effect on the magnitude of the 10 and 100 year, 24 h events under the Pearson type III distribution. Rainfall classified as precipitation when SWE equaled zero was exactly the same as fi 24.2 rainfall classi ed as precipitation when SWE decreased and when SWE equaled Rain À zero, with the exception of a single site. At this station, the 100 year, 24 h events Snowmelt- 10 Year, 24 h Event (mm)differed by 0.1 mm. 100 Year, 24 h Event (mm) On average, snowmelt calculated as the change in SWE plus precipitation was 0.7 mm greater than snowmelt calculated as the change in SWE only for both the 10 and 100 year, 24 h events (Table 1). The 10 year, 24 h event under the revised snowmelt calculations was greater at 85 of the stations but by a small amount (≤3.7 mm); similarly, differences were lower at four stations but only by a very small amount (≤0.2 mm) (Table S1 in the supporting information). For the 100 year, 24 h event, the range of differences were more; the revised snowmelt calculations yielded higher snowmelt estimates (≤7.1 mm increase) at 52 stations and lower estimates (≤1.0 mm decrease) at 37 stations (Table S1

Precipitation Rain Snowmelt in the supporting information).

Summary of the 10 and 100 Year, 24 h Events Including the Differences The differences were generally small for both rainfall and snowmelt and mostly less than the reported precision of the data (2.54 mm). It should be noted that adding precipitation to the change in SWE likely overestimates the input of water AverageSDMinimumMaximum 54.5 29.1 94.6 14.2 41.6 25.2 64.8 60.5 9.69 33.6 114 15.9 18.9 85.1 18.7 27.1 16.9 58.6 9.90 149 24.1 126 21.0 25.9 181 32.9 140 31.1 62.1 18.5 98.3 Table 1. to the hydrological system, as the change in SWE is a decrease in mass which

includes potential liquid water from rain leaving the system on the same day. In fact, if only snowmelt amounts are desired, it may be useful to subtract precipitation from the change in SWE; since the phase of precipitation is not known, the precipitation could be snow accumulating rather than rain leaving the snowpack. Another potential source of uncertainty is the choice of probability distribution. In this study, we evaluated three distributions that are commonly used in hydrologic frequency analysis: Pearson type III, Log-Pearson type III, and Gumbel. Other probability distributions could also be used, such as generalized extreme value, gamma, and generalized normal, and could result in differences in the reported results. Our objective was to compare the relative magnitudes of large rain, snowmelt, and precipitation events. The true underlying distribution will never be known, and differences between quantile estimators resulting from alternative distributions are typically much less than the uncertainty in the estimators themselves [Stedinger and Grifﬁs, 2008]. The uncertainty in the results associated with the choice of probability distribution is generally small. On average, ratios of 100 year, 24 h snowmelt to total precipitation (snowmelt to rainfall) events estimated with the Log-Pearson type III were 2.6% (2.0%) lower than estimates under the Pearson type III distribution and 3.6% (4.1%) higher under the Gumbel distribution. On average, for the 100 year events, the Gumbel distribution yielded snowmelt amounts 11.4% greater than those estimated using the Pearson type III distribution, while precipitation and rainfall estimations were a 7.89% and 7.67% greater, respectively. The Log-Pearson type III distribution only increased 100 year, 24 h snowmelt estimates by 0.53% overestimates under the Pearson type III distribution, but precipitation estimates were 3.47% greater (3.07% for rain).

Acknowledgments Additional factors besides melt or rainfall depth are also important when translating depth duration frequencies The SNOTEL daily data are available to flood frequencies. Hirschboeck et al. [2000] noted that “scale is a key component of flood causality because the from the National Water and Climate fl ” Center of the Natural Resources way precipitation is delivered in space and time affects the type of ood and its accompanying hazards. At Conservation Service at . Individual snowmelt dominated [Jarrett, 1990]. Snowmelt or rainfall intensities vary throughout a 24 h period, with melt stations are listed in the supporting information. The elevation data used in rates generally peaking in late afternoon and ceasing later at night, i.e., only occurring for about 12 h each day. Figure 1 were obtained from the U.S. Geological Survey National Elevation data set at (last access 23 January 2014) (data set: NED; multiple data set The 10 year, 24 h precipitation, rainfall, and snowmelt events for 90 mountain locations in the Southern Rocky names based on raster locations). The Mountains averaged 54.5, 41.6, and 60.5 mm, while the 100 year events averaged 79.1, 64.3, and 81.7 mm. Continental Divide and state and Snowmelt yielded the largest magnitude events at any station (114 and 181 mm for the 10 and 100 year national boundaries shown in Figure 1 are from the U.S. Geological Survey return periods). Precipitation events were larger than rainfall, illustrating the importance of large snowfall National Atlas shapefiles (data sets: events in the region. On average, snowmelt events were larger than precipitation and rainfall events. condivl020, data set name: Continental Divide of the United States; data set: While this study was conducted for the Southern Rocky Mountains and the results reflect one region’s statesp020, data set name: State hydrometeorology, these results have broader implications for other areas with significant snow accumulation. Boundaries of the United States; and data set: bound_p, data set name: North In such regions, we suggest that snowmelt as well as rainfall and precipitation frequencies are likely important American Atlas-Political Boundaries. for developing accurate flood frequencies. Furthermore, at high elevations, using frequencies of extreme All data sets were downloaded from rainfall or precipitation alone may underestimate the water potentially available for runoff generation. (last access 4 June 2014 for the first two data sets and last access 25 November 2014 for the national boundaries). Two anonymous reviewers References provided useful, constructive feedback Changnon, D., T. B. Mckee, and N. J. Doesken (1991), Hydroclimatic variability in the Rocky Mountains, Water Resour. Bull. Am. Water Resour. on the original manuscript. Glenn Assoc., 27(5), 733–743. Patterson and Graham Sexstone of Chow, V. T. (1951), A general formula for hydrologic frequency analysis, Eos Trans. AGU, 32, 231–237. Colorado State University provided England, J. F., J. E. Godaire, R. E. Klinger, T. R. Bauer, and P. Y. Julien (2010), Paleohydrologic bounds and extreme flood frequency of the upper assistance with the literature review Arkansas River, Colorado, USA, Geomorphology, 124,1–16, doi:10.1016/j.geomorph.2010.07.021. and preliminary geospatial analysis, Fassnacht, S. R. (2011), Snow density, in Encyclopedia of Snow, Ice, and Glaciers, Encycl. of Earth Sci. Ser., edited by V. P. Singh, P. Singh, and U. K. respectively. Interesting discussions Haritashya, p. 502, Springer, Dordrecht, Netherlands. with Douglas Laraby of Winter Park Resort Fassnacht,S.R.,N.B.H.Venable,J.Khishigbayar,andM.L.Cherry(2013),Theprobabilityofprecipitationassnowderivedfromdaily provided insight into the relevance of air temperature for high elevation areas of Colorado, United States, in Cold and Mountain Region Hydrological Systems Under snowmelt versus rainfall in the mountains Climate Change: Towards Improved Projections (Proceedings of Symposium H02, IAHS-IAPSO-IASPEI Assembly),vol.360,pp.65–70, of Colorado. All their input is appreciated. IAHS, Gothenburg, Sweden. Funding was provided by NASA award Graybeal, D. Y., and D. J. Leathers (2006), Snowmelt-related flood risk in Appalachia: First estimates from a historical snow climatology, J. Appl. NNX11AQ66G. Meteorol. Climatol., 45, 178–193.

Hirschboeck, K. K., L. L. Ely, and R. A. Maddox (2000), Hydroclimatology of meteorologic floods, in Inland Flood Hazards: Human, Riparian, and Aquatic Communities, edited by E. E. Wohl, pp. 39–72, Cambridge Univ. Press, Cambridge, U. K. Hoffmann, W., P. Baltes, J. Fischer, S. Kaufmann and U. Dirkschneider (1981), Burning, Chapter 6 in Breaker, Brain, Hamburg, Germany. Hosking, J. R. M., and J. R. Wallis (1997), Regional Frequency Analysis: An Approach Based on L Moments, Cambridge Univ. Press, Cambridge, U. K. Interagency Advisory Committee on Water Data (1982), Guidelines for Determining Flood-Flow Frequency: Bulletin 17B of the Hydrology Subcommittee, 183 pp., Office of Water Data Coordination, U.S. Geol. Surv., Reston, Va. Jarrett, D., and M. Tomlinson (2000), Regional interdisciplinary paleoflood approach to assess extreme flood potential, Water Resour. Res., 36(10), 2957–2984, doi:10.1029/2000WR900098. Jarrett, R. D. (1990), Paleohydrologic techniques used to define the spatial occurrence of floods, Geomorphology, 3, 181–195. Jarrett, R. D., and J. E. Costa (1988), Evaluation of the flood hydrology in the Colorado Front Range using precipitation, streamflow, and paleoflood data for the Big Thompson River basin Water Resources Investigations Rep. 87-4117, U.S. Geol. Surv., Denver. Kampf, S. K., and E. E. Richer (2014), Estimating source regions for snowmelt runoff in a Rocky Mountain basin: Tests of a data-based conceptual modeling approach, Hydrol. Process., 28, 2237–2250, doi:10.1002/hyp.9751. Khaliq, M. N., T. B. M. J. Ouarda, J. C. Ondo, P. Gachon, and B. Bobée (2006), Frequency analysis of a sequence of dependent and/or nonstationary hydro-meteorological observations: A review, J. Hydrol., 329(3–4), 534–552, doi:10.1016/j.jhydrol.2006.03.004. Payton, E. A., and C. M. Brendecke (1985), Rainfall and snowmelt frequency in an alpine watershed, in Proceedings of the 53rd Annual Western Snow Conference, Long-Term Ecological Research Program, Boulder, Colo. Perica, S., D. Martin, S. Pavlovic, I. Roy, M. St. Laurent, C. Trypaluk, D. Unruh, M. Yekta, and G. Bonnin (2013), Precipitation-Frequency Atlas of the United States, NOAA Atlas 14, vol. 8, version 2.0, U.S. Dep. of Commer., National Oceanic and Atmospheric Administration, National Weather Service, Silver Spring, Md. Serreze, M. C., M. P. Clark, and A. Frei (2001), Characteristics of large snowfall events in the montane western United States as examined using snowpack telemetry (SNOTEL) data, Water Resour. Res., 37(3), 675–688, doi:10.1029/2000WR900307. Shelby, A. (2004), Red River Rising: The Anatomy of a Flood and the Survival of an American City, Borealis Books, St. Paul, Minn. Stedinger, J. R., and V. W. Griffis (2008), Flood frequency analysis in the United States: Time to update, J. Hydrol. Eng., 13, 199–204. U.S. Soil Conservation Service (1973), A Method for Estimating Volume and Rate of Runoff in Small Watersheds, SC-TP-149, U.S. Department of Agriculture Soil Conservation Service, U.S. Gov. Print. Off., Washington, D. C. Waylen, P., and M. Woo (1982), Prediction of annual floods generated by mixed processes, Water Resour. Res., 18(4), 1283–1286, doi:10.1029/ WR018i004p01283.