Water Resources Research

Use of spatially distributed stream stage recorders to augment rain gages by identifying locations of thunderstorm precipitation and distinguishing rain from snow

Authors


Abstract

[1] Precipitation measurements in complex terrain are difficult to obtain, particularly during convective thunderstorm precipitation or mixed rain and snow. In such cases, stream stage recorders distributed across many small, headwater streams can bridge the gap between point measurements of precipitation and larger-scale total basin discharge. We use case studies from two steep basins with shallow soils in the Sierra Nevada, California to illustrate how distributed stream sensors can help identify the location, timing, and intensity of spatially variable precipitation.

1. Introduction

[2] Precipitation varies markedly in regions of complex terrain, and standard precipitation gages, due to undercatch and siting logistics [e.g., Sieck et al., 2007], are often unrepresentative of regional rain rates and distributions. Rain gages are generally sparsely distributed and are concentrated at lower elevations. Scanning radars can help detect spatial precipitation distributions between rain gages, but radar views are often blocked by terrain in mountainous regions. Thus, convective precipitation over high peaks may be entirely missed by conventional monitoring techniques.

[3] Distributed stream level recorders can be used to augment precipitation gages and radars to gain further insight into spatial patterns of precipitation. In mountainous terrain stream stage usually increases monotonically with increasing discharge. Thus, while stage information cannot reveal the precise quantity of water moving down a given stream, it can be used to identify the timing of peak and minimum flows, and the times when flow volumes are increasing or decreasing. In basins with a fast response time (e.g., steep slopes and shallow soils), this timing alone can identify where and when precipitation has occurred. In the Sierra Nevada and Rocky Mountain ranges of the western United States, most river basins draining areas on the order of 1000 km2 or less have response times of less than a day, as evidenced by clear diurnal cycles each spring associated with daily fluctuations in snowmelt [Lundquist and Cayan, 2002]. In these areas, stream stage records are useful to identify convective precipitation (e.g., summer thunderstorms) and in mixed rain and snow events, as detailed below. A dense array of precipitation gages could achieve similar objectives, but costs exceed $100 each without consideration of installation, or operational requirements (e.g., monthly visits to clear debris). Because stream recorders can cover more area with fewer sensors, incorporating them into the measurement strategy could potentially lower cost.

2. Summer Thunderstorms

[4] As snowmelt becomes a less reliable source of water in late summer, summer precipitation will become increasingly important in controlling late summer soil moisture and minimum flows in mountain streams, both of which are important for ecology and summer water supplies. Focusing on the western United States, Hamlet et al. [2007a, 2007b] found that modeled late-season soil moisture depends more on summer precipitation than on temperature or the spring snowpack. Thus, monitoring and predicting summer precipitation, which most often falls during thunderstorms at high elevations, is important to predict drought severity, fire danger, erosion, and species composition in these regions.

[5] The Merced and Tuolumne Rivers of Yosemite National Park, California (Figure 1) have been the focus of many hydrologic studies and are well instrumented compared to other Sierra Basins (see Dettinger et al. [2004], Lundquist et al. [2004, 2005], and Lundquist and Cayan [2007] for further basin and instrument details). About 90% of the area is underlain by intrusive rocks (chiefly granodiorite), which erode slowly and allow less infiltration than other rock types [Huber, 1987]. Mean soil depths are about 1 m (based on Yosemite National Park soil surveys), and mean slopes in the region are about 18° (calculated from a 10-m resolution DEM), with exposed bedrock and steeper slopes in the headwater regions. These features allow the rivers to respond quickly to precipitation input.

Figure 1.

Locations of thunderstorms on 23 July 2003, as indicated by the striped subbasins and lightning strikes. Dashed outlines identify monitored subbasins of the Tuolumne and Merced rivers in Yosemite National Park, California. Thirteen operating precipitation gages are marked by P and a symbol representing the operating organization. Note that P4, the Bridgeport precipitation gage, is located just north of the mapped area.

[6] Even with a relatively dense precipitation gage network, the occurrence, location, and intensity of precipitation during summer thunderstorms are difficult to monitor. The locations of lightning strikes (Figure 1) are detected by the National Lightning Detection Network (NLDN) [Orville, 2008], but the amount of precipitation associated with those lightning strikes must be determined independently. For example, on the afternoon of 23 July 2003, over 150 lightning strikes were detected within the Merced River basin between noon and 2:00 P.M. Pacific daylight time (PDT), with the majority of strikes occurring near 1:00 P.M. PDT. Water levels in several headwater streams of the Merced River rose by 2:00 P.M. PDT, 23 July (Figures 1 and 2), whereas gages monitoring larger catchments downstream peaked at 1:00 P.M. PDT (Merced at Echo), 4:00 P.M. PDT (Merced at Happy Isles), and 8:00 P.M. PDT (Merced at Pohono) the following day, 24 July. On the basis of discharge measured at the USGS gage at Happy Isles (Figure 2c), over 600,000 m3 of water moved through the basin during the 48 h following noon, 23 July. Normalized by basin area, this would be an average of 0.7 mm over the Pohono Basin and 1.3 mm over the Happy Isles Basin. At the same time, only small diurnal fluctuations in streamflow due to snowmelt were observed in the Tuolumne River drainage to the north (Figures 1 and 2), and a trace amount of precipitation (<1 mm) was recorded in only one of the fifteen operating rain gages located nearby (Figures 1 and 2). On 24 July, water levels rose in several streams in the Tuolumne and Merced watersheds, precipitation (ranging from less than 1 mm to more than 12 mm) was recorded in 4 rain gages, and most lightning strikes (not shown) occurred west of the basins where stream levels rose. On 25 July, most lightning strikes (not shown) occurred east of the park, precipitation (ranging from <1 to 4.5 mm) was recorded in 2 rain gages, and only Rafferty Creek (stream 4 on the map, Figure 1) exhibited a rise in water level. Of all 3 days, the largest stream response was on 23 July, which could not be expected from precipitation measurements alone. Such complicated patterns were observed throughout the summer thunderstorm season.

Figure 2.

Stream level records from (a) four subbasins of the Tuolumne River, (b) five subbasins of the Merced River, and (c) two U.S. Geological Survey (USGS) Merced River gages for 22–25 July 2003. Vertical dashed line represents the time of most lightning strikes shown in Figure 1. Note that axis date labels are centered at 12:00 P.M., with unlabeled tick marks representing midnight. Slight fluctuations in Budd and Rafferty Creek on 22–23 July are due to diurnal variations in snowmelt. (d) Daily precipitation records from 7 of 13 precipitation sensors near the basins. The six sensors not listed in the legend all reported no precipitation throughout the shown period. Locations of all gages are labeled by numbers on Figure 1.

[7] The small headwater streams (e.g., stream 6, the Merced Peak Fork and stream 7 the Lyell Confluence) responded immediately after lightning strikes on 23 July and exhibited two distinct peaks, likely reflecting two cells of heavy precipitation. This signal was damped and delayed as it moved further downstream, so that the standard USGS stations alone provided less information about the storm's location, timing, and local intensity. For example, a rise in stage in the same two headwater streams on the prior day, 22 July 2003, was damped by the time the pulse reached the Merced River at Echo (stream 9) and undetectable in the Happy Isles and Pohono U.S. Geological Survey (USGS) gages (Figures 1 and 2). The distributed stream stage recorders in low-order catchments added information to traditional precipitation gages and USGS streamflow records and allowed for better monitoring of the frequency and spatial distribution of summer thunderstorm precipitation. This is consistent with the progressive changes in hydrograph shape and losses of information at larger basin scales observed at a series of stations along the Sleepers River near Danville, Vermont [Dunne and Leopold, 1978, Figure 10–3].

[8] In this application small basins serve as the “rain catch domain” and provide more extensive coverage than point gages alone. Although rises in stream stage only occur when and where enough rain fell to wet the catchment and produce a hydrologic response, these larger storms are often significant to river chemistry and ecology. For example, similar storm events following a dry period have been linked to episodic stream acidification [Bishop et al., 1990; Peters, 1994], due to flushing of accumulated dry deposition, and such acidification has been linked to a decrease of species richness in aquatic invertebrates [Vinson and Hawkins, 1998] and to difficulties in fish reproduction [Beamish, 1976]. The location(s) where heavy rain fell can be pinpointed to within the boundaries of the gaged subbasins (Figure 1). Observations from numerous stage recorders in small, first-order streams, can be used to pinpoint where and when it rained, and to approximate the rain intensity. This information, while qualitative, is important to understand localized processes including erosion [Wondzell and King, 2003], chemical flushing of nutrients [Band et al., 2001], and ecological responses [Bynum and Smith, 2001].

3. Rain Versus Snow

[9] Distributed stream sensor networks in mountainous terrain can also help identify regions receiving rain versus snow. For a given storm, one of the greatest difficulties in flood prediction in complex terrain involves determining which percentage of a river basin receives precipitation as rain (rapid runoff) and which percentage receives precipitation as snow (delayed runoff). Standard precipitation gages and snow sensors have difficulty distinguishing between solid and liquid precipitation, such that only direct observations [U. S. Army Corps of Engineers, 1956], optical disdrometers [Yuter et al., 2006], or colocated snow pillow, snow depth, and heated precipitation sensors [Lundquist et al., 2008] can identify precipitation type. All of these options are expensive and only record precipitation at a point.

[10] Distinguishing rain from snow is particularly important in maritime basins spanning a wide range of elevations, such as the North Fork (NF) American River basin in the Sierra Nevada, California. The NF American River has average soil depths of about 1 m [Knowles, 2000], an average slope of 12° (calculated from a 30-m resolution DEM), and has been extensively studied (for more details on basin and instrument characteristics, see Dettinger et al. [2004], Jeton et al. [1996], Shamir and Georgakakos [2006], and Lundquist et al. [2008]).

[11] Pressure sensors were deployed in three small subbasins of the NF American River as part of NOAA's Hydrometeorological Testbed (HMT) project [Ralph et al., 2005], selected such that each drains a narrow range of elevations (Figure 3). Thus, their varying responses to different storms provide a measure of where rain contributes to runoff during each storm and where snowmelt contributes runoff in the spring, providing an independent measure of where and when precipitation contributes to runoff (Figure 4).

Figure 3.

Map and fraction of basin area below each elevation for three monitored subbasins and for the entire North Fork (NF) American Basin, California.

Figure 4.

Water year 2005–2006: (a) discharge at the North Fork stream gage, (b) stream stage at three subbasins at different elevations, (c) zoomed view of late December/early January, with discharge (thick black line) plotted on the left axis and stage plotted on the right axis, and (d) same as Figure 4c but zoomed to March–April. Vertical dashed lines in Figures 4a and 4b identify the periods shown in Figures 4c and 4d.

[12] Water year 2006 (October 2005 through September 2006) was characterized by a series of warm, high rainfall storms in late December/early January, and a series of much colder storms in March (Figure 4a). The December 2005 to January 2006 storms were characterized by high melting levels, and all three subbasins, at altitudes of approximately 500, 1500, and 2000 m, contributed to runoff (Figure 4b). Specifically, stage rose at Onion Creek, the highest subbasin, during the 20 December and 22 December storms, but not during the cooler 26 December storm. Stage rose again during the 28 December and 1 January storms (Figure 4c). In contrast, in March, Onion Creek did not contribute directly to discharge; snow accumulated and contributed to runoff later in the spring (Figure 4b). Onion Creek showed no response from 25 March to 2 April and only a small rise on 4 April (Figure 4d). Although water depth above the sensor at Colfax Creek reached the same level during the 25 March and 4 April events, the total NF basin discharge was three times larger during the 4 April event, due to warm rain falling at higher elevations, resulting in a larger area contributing to runoff. In late April and May, the shape of the total basin hydrograph (Figure 4a) reflected contributions from East Fork and Onion Creek but not Colfax Creek (Figure 4b); it is likely that snowmelt provided runoff at this time.

[13] In addition to identifying the different source areas for the total basin discharge, the monitored subbasins illustrate how the times when water levels begin rising and the times when they reach their peak vary between storms and locations (Figure 5). For example, during one of the first storms of the season, water levels began rising at 8:00 P.M. PST on 30 November at Colfax Creek, at 9:30 P.M. PST on 30 November at East Fork and Onion Creek, and at 2:00 P.M. PST on 1 December 2005 at the main NF gage (Figure 5a), with a lag time of over 15 h between the small subbasins and the main river. On 26–27 December 2005 (Figure 5b), the lag time between the three tributary basins and at the main NF gage was less than 10 h. On 24 March 2006 (Figure 5c), the lag between the Colfax and East Fork gages and the main NF gage was about 3 h. Onion Creek catchment received snow that day, and there was no stream rise. The relative timing of when the water level at each gage location reached its peak also varied widely between the three storms. These variations could be due to changes in soil moisture and groundwater levels through the season, to spatial variations in the timing of precipitation input, or to variations in the distribution of rain and snow across the basins.

Figure 5.

Discharge (thick black line) for the NF American River gage and stage at three subbasins for three 48-h periods, with one representing 1:00 A.M. PST on (a) 1 December 2005, (b) 27 December 2005, and (c) 24 March 2006. Vertical dashed lines indicate the time when the stream level began rising for a duration of 3 h or more. Vertical dashed-dotted lines indicate the time peak water level was observed at each site for the given storm. The discharge values on the axis refer to the NF USGS gage. The stage measurements have been shifted and stretched to ease comparison of the timing of changes in rising and falling water levels.

4. Conclusions

[14] These examples illustrate how stream stage can help to identify where, when, and what type of precipitation occurred in complex mountainous terrain. These sensors can be deployed quickly and inexpensively and can provide an independent check of spatial patterns in runoff, which are often hard to model because of the scarcity of reliable precipitation data. Stream sensors distributed across numerous small subbasins also provide intermediate-scale measurements of precipitation distributions for highly localized storms, bridging the gap between point measurements of precipitation (e.g., traditional rain gages) and the larger-scale catchment hydrologic response (e.g., traditional USGS measurements of discharge). These measurements are best suited for basins similar to the example basins with regards to steep slopes (average >10°), shallow soils (average ≈1 m), and fast response times (average <1 day). Fortunately, these characteristics apply to many mountain basins in the western United States, which are regions lacking in radar and rain gage coverage but sensitive to summer convective precipitation and winter mixed rain and snow events.

[15] To monitor thunderstorm precipitation, gages should be distributed in small, low-order streams draining mountain peaks where thunderstorm activity (identified by lightning ground strikes) is commonly observed. There is a trade-off between the number of sensors required and the size of the subbasins monitored, where more sensors monitoring smaller subbasins better pinpoint regions and timeframes with intense versus insignificant rainfall. If broadcast in real time, these data could help pinpoint locations that may be subject to forest fires, substantial erosion, or nutrient flushing. Because of travel delays in the streamflow pulse, these data could also provide several hours of warning regarding streamflow rises at locations downstream.

[16] To monitor rain versus snow, sensors should be placed in small tributary streams that drain a narrow range of elevations. The narrower the elevation range, the more precisely the gage can identify at which elevation snow changes to rain and contributes to runoff. Alternatively, a dense array of heated rain gages and snow depth sensors could distinguish spatial regions of rain versus snow, but the deployment logistics, power/electricity requirements, and cost of such a system would be greater than for stream stage recorders distributed over the same area. Detailed instructions of how to deploy stream stage recorders are included in the auxiliary material.

Acknowledgments

[17] Thank you to Steve Burges, Bob Westfall, Scott Tyler, John Selker, and three anonymous reviewers for providing comments and editorial guidance. Thank you to Dan Cayan, Mike Dettinger, Julia Dettinger, Jim Wells, Larry Riddle, and Fred Lott, who helped with Yosemite stream deployments and anchor designs. Thank you to the dedicated and professional staff of the Resource Management and Science and Wilderness divisions of the Yosemite National Park Service, particularly Jim Roche, Mark Fincher, and Joe Meyer, who helped with adapting deployment strategies to meet Wilderness Act and NEPA standards. Thank you to Clark King, Bob Zamora, Dave Kingsmill, Randall Osterhuber, Marty Ralph, and the NOAA HMT program for discussions on American River measurements. This work was supported by a Canon National Parks Science Scholarship, by the National Science Foundation under grant CBET-0729838, and by the National Oceanographic and Atmospheric Administration under award NA17RJ1232. Lightning strike data were supplied through collaboration with Dan Cayan, Jan Van Wagtendonk, and the California Applications Program.

Ancillary