Unraveling the mysteries of the large watershed black box: Implications for the streamflow response to climate and landscape perturbations



[1] Spatial and temporal trends in stream chemistry have been used to provide insights into the scale dependencies of streamflow generation processes in small catchments. However, these scale dependencies have not been thoroughly investigated at large watershed scales (defined as drainage areas greater than 1000 km2). Quantifying these scale dependencies is critical to understanding how large watersheds will respond to future perturbations; e.g., the long-term streamflow response to climate change and/or changes in land-cover and land-use. Here we investigate the spatial and temporal scaling relationships of all dominant streamflow generation processes in a large alpine watershed in the southern Rocky Mountains of Colorado. Observations in the watershed indicate that dominant streamflow processes are spatially and temporally variable. The relative strengths of dominant streamflow mechanisms vary as a function of internal watershed structure (i.e., spatial variability in topographic relief, soil development, groundwater flowpath development, and stream network structure) and external forcing such as timing and character of precipitation. This behavior coupled with previous observations that streamflow from the watershed contained a significant component of basin-scale groundwater, suggests that similar large watersheds may have internal buffering, at least initially, against rapid change.

1. Introduction

[2] Understanding the processes that control streamflow generation at the large watershed scale is fundamental to understanding how streamflow will respond to climatic and anthropogenic perturbations. While the characterization of these processes in small catchments has been well documented [Beven, 2006], extending these findings to larger watershed scales is very difficult [Sivapalan, 2003]. Furthermore, field characterization of these processes in large watersheds remains lacking despite an increasing urgency to understand process behavior at larger scales [Palmer, 2009; Singleton and Moran, 2010]. Due to lack of an adequate observational basis, streamflow generation processes in large watersheds have traditionally been described using “black-box” approaches [Black, 1996]. Black-box analogies are not sufficiently adequate for understanding impacts to streamflow from future changes associated with climatic and anthropogenic perturbations. Fortunately, the processes controlling streamflow generation have direct and quantifiable effects on stream chemistry [Bricker and Jones, 1995]. In other words, the various flowpaths contributing to streamflow have evolved along distinctive geochemical evolutionary pathways in transit to the stream, and this effect makes it possible to deduce dominant streamflow generation processes across multiple scales in large watersheds [Pinder and Jones, 1969].

[3] A reductionist perspective to understanding process behavior in large watersheds is to investigate small-scale process behavior in hopes of extending that behavior to larger scales. One approach is to view the streamflow response from large watersheds as an aggregation of responses from smaller landscape elements (i.e., the black box contains a multitude of hillslopes and subcatchments perhaps operating independently of each other). Any trends observed in stream chemistry will, therefore, be viewed as arising from the aggregation of the small-scale responses [Sivapalan, 2003]. This perspective implies that the relative importance of streamflow generation processes is not scale-dependent [e.g.,Hrachowitz et al., 2010] and by extension, that mixing processes in the stream network and/or hyporheic zone may control any emergent trends in stream chemistry [e.g., Shaman et al., 2004]. Although these processes are clearly important, this approach ignores the possibility that there are processes that operate across traditional spatial (hillslope, hyporheic, and subcatchment) and temporal scales [Reid, 1998]. This perspective also suggests that if, for example, fast runoff processes are dominant controls on streamflow generation across multiple scales, then linear and perhaps rapid responses to perturbations will be observed in streamflow responses from large watersheds [Singleton and Moran, 2010].

[4] Alternatively, large watersheds can be viewed as single elements whose streamflow response is more than simply an aggregation of responses from smaller landscape elements. In other words, the black box contains a multitude of different processes operating across many different spatial and temporal scales and openly interacting with each other outside the stream network or hyporheic zone. This perspective promotes a more fully four-dimensional view of large watersheds in which interactions can occur between very different processes characterized by very different geochemical pathways on widely ranging spatial and temporal scales and internal residence times. For example, recent work in a 1600 km2alpine watershed indicated that basin-scale groundwater contributions controlled the structure that emerged from the stream chemistry as scale increased. The small-scale processes primarily imposed seasonal noise on that structure [Frisbee et al., 2011]. Linear and/or rapid streamflow responses to perturbations may not always be observed in a fully three-dimensional flow field that allows for spatial and temporal variability in process behavior and provides additional internal buffering.

[5] Here a distinction must be made between observations indicating an earlier onset of runoff during snowmelt [e.g., Cayan et al., 2001; Moore et al., 2007] and long-term streamflow responses (i.e., the processes that sustain surface flow over the entire annual hydrograph). Fast runoff processes including surface runoff, very shallow subsurface runoff, and preferential flow in the soil may be dominant streamflow generation mechanisms during snowmelt in mountainous watersheds [Wilcox et al., 1997; Suecker et al., 2000; McNamara et al., 2005; Liu et al., 2008]. As a consequence, changes in the character of meteoric inputs, snowpack depth, duration of snowcover, etc. may be rapidly conveyed to the stream making them useful predictors for climate change. While these impacts have been documented, the vulnerability of long-term streamflow generation to perturbations at the large watershed scale has not been addressed.

[6] In this paper, we are building upon the work recently presented by Frisbee et al. [2011] by expanding our investigation into the spatiotemporal scale dependencies of streamflow generation processes. In order to accomplish this, we used endmember mixing analysis (EMMA) [Hooper, 2003] on four years of chemistry and stable isotope data in the Saguache Creek watershed (Figure 1a). The Saguache Creek watershed is a large (approximately 1600 km2), mountainous watershed located in the San Juan Mountains of southwestern Colorado (38°5′14″N and 106°8′29″W). A full site description can be found in the work of Frisbee et al. [2011]. The objective of this research was to determine if, and more fundamentally how, all dominant streamflow generation processes change as a function of spatial and temporal scale in this large watershed. The spatial and temporal variability of these processes provides fundamental information on the scaling of streamflow generation mechanisms in large watersheds and more importantly, has implications for our understanding of how streamflow will respond to future change.

Figure 1.

(a) Map of Saguache Creek watershed. Headwater subwatersheds: Saguache Creek Middle Fork (SCMF, 89 km2) and Saguache Creek South Fork (SCSF, 81 km2). Tributary subwatersheds: Hodding Creek (HC, 78 km2), Middle Creek (MC, 138 km2), and Sheep Creek (ShpC, 186 km2). Longitudinal stream sites: Saguache Creek Curtis Ranch (SCCR, 538 km2), Saguache Creek Upper Crossing (SC1, 692 km2), Saguache Creek Lower Crossing (SC2, 1083 km2), and Saguache Creek Hill Ranch (SCHR, 1410 km2). (b) Annual hydrograph for Saguache Creek measured near SCHR and based upon average daily streamflow data from 1927 to 2010 (average daily streamflow minus 1 standard deviation is shown as dotted line and average daily streamflow plus 1 standard deviation is shown as dashed line). Orange box encompasses the “Fall/Winter” season, blue box encompasses the “Snowmelt Freshet”, and the grey box encompasses the “Summer Rainfall” season. Blue bar illustrates the typical sampling season.

2. Methods

2.1. Identifying Streamflow Generation Mechanisms From Component Chemistry

[7] Grab samples of streamflow were collected monthly from nested headwater and tributary subwatersheds and at increments in accumulated drainage area working longitudinally down the main stream channel; representing drainage areas ranging from 78 to 1410 km2 (Figure 1a). Samples of potential endmembers including meteoric water, surface runoff, soil-water, and groundwater were sampled on the same spatial and temporal scales. Full geochemical and isotopic data and collection methodologies are provided byFrisbee [2010]. Components of streamflow were broadly classified into four major streamflow generation mechanisms post-EMMA analysis:fast runoff processes, unsaturated flow, groundwater, and network routing. Fast runoff processes included near-channel runoff, overland flow processes, and shallow flow through forest litter and unconsolidated talus. Fast runoff exhibited minimal geochemical evolution associated with limited contact between water and soil. Unsaturated flow exhibited intermediate geochemical evolution associated with enhanced contact between water and soil. We used soil-water samples that were thought to be representative of the geochemical evolution associated with matrix flow since this endmember will, in most cases, be more geochemically evolved than preferential flow in the soil. Groundwater exhibited significant geochemical evolution associated with longer residence times in the bedrock aquifer. Network routing exhibited the preservation of tributary inputs in the stream network. In other words, the most plausible explanation for the appearance of geochemically dilute components in streamflow during relatively quiescent periods was the integration and subsequent routing of meteoric events into the stream network. Details on the EMMA analysis and the figures for the mixing subspaces and streamflow separations for the headwater subwatersheds and main channel sites are given byFrisbee et al. [2011]. Mixing subspaces and streamflow separations for tributary subwatersheds are provided in the auxiliary material (Figure S1) since these are not provided by Frisbee et al. [2011].

[8] The accuracy of the source partitioning provided by EMMA was quantified in two different ways. First, we examined the relationships between actual and re-created chemical compositions of stream water samples. The re-created chemical composition of stream water samples was determined using only the conservative tracers and the number of endmembers that had been identified by the diagnostic tools of mixing models. Furthermore, the selected endmembers had been screened using calculations of the distance from their original chemical composition and theU-space (mixing subspace) projections of the endmembers (i.e., short distances indicate better fits [Christophersen and Hooper, 1992; Liu et al., 2008]). The correlation of actual and re-created streamflow chemistry, therefore, provides a quantitative means to assess the validity of the conceptual models (a well-posed model produces a p < 0.05 indicating a statistically significant relationship [Liu et al., 2008]). Thirty-six out of 38 stream water chemistry reconstructions resulted in statistically significant relationships, indicating that the streamflow partitioning was appropriate (see Table S1 in theauxiliary material for data). Second, we assessed the source partitioning provided by EMMA using an “artificial” stream water composed of known components and known contributions of each component. EMMA was capable of correctly identifying 100 percent of the components responsible for streamflow generation and identifying 100 percent of the contributions of each component in each individual “artificial” sample. These data provide confidence in the source partitioning.

[9] The temporal variability of these components was examined by partitioning the streamflow samples into three distinct temporal periods: 1) snowmelt freshest typically occurring from April to July, 2) a summer rainfall season that overlaps the latter part of the snowmelt recession in July and extends through September, and 3) a fall and winter period that occurs from October through March in which streamflow is lowest (Figure 1b).

3. Results: Deconstructing the Black Box

3.1. Controls on Streamflow Processes During the Snowmelt Freshet

[10] Fast runoff processes were a dominant control on streamflow generation in the steep, high-elevation headwater subwatersheds and in the smallest tributary subwatershed, HC (Figure 2a). As scale increased, the dominance of fast runoff processes decreased and other components became increasingly important. For example, streamflow in the larger tributary subwatersheds, MC and ShpC, and at the SCCR site contained a significant component of unsaturated flow. The strength of the unsaturated flow component is likely attributable to increased soil development and changes in the character and timing of snowpack accumulation and snowmelt at larger scales since precipitation is strongly correlated with elevation in the San Juan Mountains. As accumulated drainage area increases, the main channel of Saguache Creek incorporates more flow and solute input from tributary subwatersheds. These contributions are reflected in the strength of the network routing component in streamflow at SC1 and SC2 (Figure 2a). Thus, during the snowmelt season, the appearance of the network routing component is presumably due to the integration of snowmelt runoff from local sources as well as snowmelt pulses from tributaries. The groundwater component was quite variable across all scales.

Figure 2.

(a) Components responsible for streamflow during the Snowmelt Freshet, (b) during the Summer Rainfall season, and (c) during the Fall/Winter season. Drainage area (shown in parentheses in km2) increases from left to right.

3.2. Controls on Streamflow Processes During the Summer Rainfall Season

[11] The relative strength of the groundwater component increased at all scales (Figure 2b). The increased groundwater contributions were balanced by decreases in the contributions from fast runoff processes and unsaturated flow across all scales. In fact, the groundwater component in streamflow increased considerably with increasing scale from SCCR (538 km2) to SCHR (1410 km2). This range in accumulated drainage area is significant because it appears that interactions with groundwater flowpath distributions containing increasingly longer flowpaths, hence waters of greater geochemical evolution, are controlling the continued geochemical evolution of streamflow (Figures 2b and 3a; see Frisbee et al. [2011]for discussion). Interactions between the stream and these longer, basin-scale flowpaths are likely not possible at smaller scales such as headwater subwatersheds.

Figure 3.

Conceptual model for the generation and geochemical evolution of streamflow in Saguache Creek showing fast runoff processes (small blue arrows), network routing (large, open black arrows), and groundwater (thin black flowpaths – flowpaths do not cross; this is an artifact of representing 3D flow in a 2D space). (a) Scale dependency of streamflow generation processes, (b) scale dependency of groundwater components, and (c) scale dependency on controls of stream chemistry.

3.3. Controls on Streamflow Processes During the Fall/Winter Season

[12] In general, the relative strength of the groundwater component increased significantly at all scales (except for SC2 and SCHR) and this was balanced by continued decreases in the contributions from fast runoff processes (Figure 2c). Interestingly, contributions from the network routing component increase with increasing scale from SC1 to SCHR perhaps indicating the passage of dilute components incorporated into the network prior to the onset of winter and/or infrequent melting at the lower elevation reaches during the winter. Most streams are bridged with ice by late November and as a consequence, it is difficult to supply fast runoff to streams with the exception of infrequent melting of ice and snow accumulated around the stream. Furthermore, the average daily temperature of this watershed during the winter ranges from 0 to −20°C and as a consequence, there is little, if any, liquid water available for surface runoff.

4. Discussion: Reconstructing the Black Box

[13] These analyses reveal temporal and spatial variability in the components of streamflow illustrating the scale dependencies of streamflow generation processes. These data can be used to reconstruct the large watershed black box and develop a fully four-dimensional conceptual model for large watersheds (Figure 3). Fast runoff processes are important across multiple scales. However, the relative importance of fast runoff decreases with increasing scale and seasonally with the transition from summer to fall/winter. In general, fast runoff processes are more important where soil development is weak and terrain is steep; features characteristic of the headwater subwatersheds (Figure 3a). Fast runoff does not allow significant contact time with soil or rock; therefore, these components tend to be geochemically unevolved (Figure 3c). Streamflow responses to perturbations from similar headwater subwatersheds may be rapid due to the relative dominance of fast runoff processes.

[14] Soils generally show greater development as elevation and topographic relief decrease. The tributary subwatersheds typically show a greater range in elevation and topographic relief than the headwater subwatersheds. Thus, soils become more developed in the tributary subwatersheds. EMMA results indicated that streamflow generation in tributary subwatersheds contains a large component of unsaturated flow (i.e., there is a coincident shift toward greater control on streamflow generation by flow through the soil; Figure 3a). This behavior was supported by field observations indicating a coincident decrease in the occurrence of surface runoff during snowmelt and storm events in the lower-elevation and lower-relief regions of the tributary subwatersheds. Unsaturated flow is more geochemically evolved than fast runoff processes and can add a significant chemical load to streamflow (Figure 3c). Matrix flow in the soil may provide additional buffering to rapid changes while preferential flow through the soil will likely not provide significant buffering.

[15] Groundwater is important at all scales and in general, its importance increases with increasing scale and with time since the passage of the snowmelt freshet (Figure 3a). Groundwater evolves geochemically with scale. Therefore, low-order streams may be sourcing short residence time, kinetically-limited groundwater from local flow systems while high-order streams may be sourcing components of longer residence time, transport-limited groundwater from regional flow systems (Figure 3b). The longer response times of deep, basin-scale groundwater systems may provide significant internal buffering, at least initially, against perturbations [Erskine and Papaioannou, 1997; Singleton and Moran, 2010] and consequently, streamflow that is dominated by groundwater contributions may not respond rapidly to these changes.

[16] One of the few other studies at a similar range of scales, that of Hrachowitz et al. [2010], did not observe significant variability in dominant runoff mechanisms as a function of scale in Scottish catchments ranging in size from 20 to 1700 km2. This may be attributed to relative differences in internal catchment structure (i.e., site specific properties such as soil development, characteristics of local bedrock, vegetation interactions, etc.), regional precipitation characteristics, and perhaps differences in methodologies. For example, elevation gradients strongly influence precipitation in the large, mountainous watersheds of the American West often producing pronounced hypsometric relationships in the remaining water-balance components. These relationships often generate feedbacks to the geomorphological development of these watersheds (e.g., spatial variability in soil development) depending on the permeability and weatherability of the local bedrock. These strong gradients are not present in the Scottish catchments; therefore, strong spatiotemporal relationships may not always be observed.

5. Conclusions and Importance of New Paradigms

[17] The findings of Hrachowitz et al. [2010] and Frisbee et al. [2011] provide differing perspectives which essentially can be treated as endmembers in the range of conceptual models for large watersheds. In the case of Hrachowitz et al. [2010], the relative importance of streamflow generation processes may not vary with scale. In extension, if, for example, fast runoff processes are dominant across scales, then the streamflow response to perturbations will likewise be rapid and adaptive strategies will consequently be limited. In contrast, our observations indicate that the relative importance of streamflow generation processes is spatially and temporally variable. In this case, the streamflow response to perturbations may be buffered as compared to the scale-invariant conceptual model, especially if streamflow contains significant contributions from sources other than fast runoff processes [Frisbee et al., 2011]. Consequently, there may be greater possibilities for adaptive management.

[18] Much more effort is needed to evaluate the existing models and test and develop new conceptual models that capture both small and large scale processes that control the streamflow response of large watersheds. Ultimately and perhaps most importantly, new paradigms will promote a more holistic understanding of streamflow generation processes in large watersheds and how these processes are coupled to the geochemical and geomorphologic evolution of large watersheds. This will improve our understanding of how streamflow from large watersheds will respond to natural and anthropogenic perturbations.


[19] Funding for this research was provided by the SAHRA (Sustainability of semi-Arid Hydrology and Riparian Areas) Science and Technology Center of the National Science Foundation (NSF agreement EAR-9876800) and by the New Mexico Water Resources Research Institute in the form of a student water research grant. Assistance from Fengjing Liu was funded by the USDA-NIFA's Evans-Allen grant. We thank the many field assistants from New Mexico Tech and we are grateful for the assistance from Enrique R. Vivoni (Arizona State University), Jan M.H. Hendrickx (New Mexico Tech), Art W. White (USGS), and Peter Lipman (USGS). Steve Sanchez, Saguache BLM/USFS Field Office, and local ranchers provided support with field installations in the Saguache Creek watershed. The comments and suggestions from Markus Hrachowitz and one anonymous reviewer greatly improved this paper.