Groundwater controls on vegetation composition and patterning in mountain meadows

Authors


Abstract

[1] Mountain meadows are groundwater-dependent ecosystems that are hot spots of biodiversity and productivity. In the Sierra Nevada mountains of California, these ecosystems rely on shallow groundwater to support their vegetation communities during the dry summer growing season in the region's Mediterranean montane climate. Vegetation composition in this environment is influenced by both (1) oxygen stress that occurs when portions of the root zone are saturated and anaerobic conditions limit root respiration and (2) water stress that occurs when the water table drops and the root zone becomes water limited. A spatially distributed watershed model that explicitly accounts for snowmelt processes was linked to a fine-resolution groundwater flow model of Tuolumne Meadows in Yosemite National Park, California, to simulate water table dynamics. This linked hydrologic model was calibrated to observations from a well observation network for 2006–2009. A vegetation survey was also conducted at the site in which the three dominant species were identified at more than 200 plots distributed across the meadow. Nonparametric multiplicative regression was performed to create and select the best models for predicting vegetation dominance on the basis of the simulated hydrologic regime. The hydrologic niches of three vegetation types representing wet, moist, and dry meadow vegetation communities were found to be best described using both (1) a sum exceedance value calculated as the integral of water table position above a depth threshold of oxygen stress and (2) a sum exceedance value calculated as the integral of water table position below a depth threshold of water stress. This linked hydrologic and vegetative modeling framework advances our ability to predict the propagation of human-induced climatic and land use or land cover changes through the hydrologic system to the ecosystem. The hydroecologic functioning of meadows provides an example of the extent to which cascading hydrologic processes at watershed, hillslope, and riparian zones and within channels are reflected in the composition and distribution of riparian vegetation.

1. Introduction

[2] Many mountain ranges, such as those of the American Cordillera, are characterized by wet winters and dry summers, which lead to ecosystems that depend on snowmelt and groundwater to provide moisture through the growing season. Riparian meadows in this environment are groundwater-dependent ecosystems [Allen-Diaz, 1991].Water table depth is a primary determinant of vegetation community, with a shallower water table regime supporting wet meadow vegetation and a deeper water table supporting xeric vegetation typical of dry meadows [Allen-Diaz, 1991; Berlow et al., 2002; Castelli et al., 2000; Chambers et al., 1999; Darrouzet-Nardi et al., 2006; Dwire et al., 2006; Dwire et al., 2004; Steed and DeWald, 2003]. These ecosystems are highly sensitive to changes in water table position, which can be driven by climate change [de Valpine and Harte, 2001], changes to channel morphology [Loheide and Booth, 2010; Loheide and Gorelick, 2007], or cumulative watershed impacts due to land use practices [Patterson and Cooper, 2007]. As a result, meadow conditions are a reflection of both local, fine-scale and cumulative, watershed-scale hydrologic processes.

[3] Climate change threatens meadows of the Sierra Nevada as the snowpack melts earlier and more winter precipitation falls in the form of rain, leading to a longer, drier growing season [Stewart et al., 2005], which may favor dry meadow vegetation at the expense of wet meadow vegetation. According to the Sierra Nevada Ecosystem Project [1996], the aquatic, riparian and meadow ecosystems are the most degraded of Sierra Nevadan habitats. Because of degradation from past human activity such as grazing, logging, road construction, and mining [Ffoliott et al., 2004; Ratliff, 1985; Trimble and Mendel, 1995], many meadows are in need of management and restoration [Baker et al., 2004; Chambers and Miller, 2004; Martin and Chambers, 2001, 2002; McKinstry et al., 2004; National Research Council, 2002; Wright and Chambers, 2002]. However, both science-based design of restoration practices and projections of meadow sustainability under future climatic conditions require improved capacity for predicting the composition and distribution of vegetation within meadow ecosystems under future scenarios.

[4] Streamflow regimes, snowmelt or precipitation patterns, hillslope hydrology, groundwater flow processes, and evapotranspirative water consumption all intersect in riparian areas in general, and meadows specifically, to create a dynamic hydrologic regime [Hammersmark et al., 2008; Loheide et al., 2009; Loheide and Gorelick, 2005]. This hydrologic regime influences the composition and distribution of vegetation across the riparian zone. Each individual species has a range of hydrologic conditions, or a hydrologic niche, which it can tolerate. When conditions are too wet, and the root zone is saturated for extended periods of time, anaerobic conditions may develop that reduce oxygen availability for root respiration. The effects of this anaerobiosis can significantly reduce productivity and cause plant mortality in species unadapted to oxygen-limited conditions. However, hydrophytes and wetland species have evolved to thrive in this environment via various mechanisms including development of aerenchyma [Jackson and Armstrong, 1999], which is a spongy tissue that allows exchange of gases between the root and shoot, and provides a competitive advantage to these plants in environments that experience periods of oxygen limitation. On the other hand, water-limited conditions can also cause stress differentially among various species [Feddes et al., 1978; Metselaar and van Lier, 2007]. A competitive advantage is obtained by those plants that have evolved to develop mechanisms for conserving and acquiring water. Examples of adaptations that favor species in water-limited conditions include reduced stomatal density, effective control of stomatal opening and closing, and heavy investment of resources in below ground biomass to access moisture from a large subsurface volume.

[5] Quantitative approaches for using hydrologic models to simulate groundwater flow regimes to predict vegetation composition and distribution in riparian areas were pioneered in systems affected by reservoir operations by Rains et al. [2004] and Springer et al. [1999]. This general approach was transferred to meadows by Loheide and Gorelick [2007], who used a three-dimensional, variably saturated groundwater flow model to investigate the effects of stream incision and the pond-and-plug restoration technique [Loheide and Gorelick, 2005] on the coupled groundwater and vegetation system. Vegetation was related to water table position using a vegetation threshold hydrograph approach [Loheide and Gorelick, 2007], which predicted vegetation community on a scale that graded from mesic to xeric. While this approach successfully simulated the occurrence of xeric swaths of vegetation along incised channels, a probabilistic prediction of vegetation occurrence is advantageous, particularly when a specific vegetation community or species is considered. Hammersmark et al. [2009] studied a similar meadow environment, and used statistical methods to determine that mean growing season water table depth explained the greatest variance in vegetation composition among many hydrologic summary variables that were tested (e.g., maximum water table depth and range). Hammersmark et al. [2009] then used generalized additive modeling to create a hydrologic niche model that related mean water table depth and range to likelihood of presence for 11 species and created probabilistic maps of vegetation distribution.

[6] The approach of Hammersmark et al. [2009] used the best hydrologic summary variables, mean water table depth and range in water table depth, as predictors of species occurrence. When combined, these two variables only indirectly address the two primary hydrologically driven factors that physiologically affect plant productivity: water stress and oxygen stress. Similarly, the approach of Loheide and Gorelick [2007], which uses a vegetation threshold hydrograph to describe the seasonal depth to water requirements for a given plant community, also indirectly relates vegetation composition to stress caused by conditions that are either too wet or too dry, but does not explicitly consider extreme hydrologic conditions that may exert stronger control on vegetation composition.

[7] The objective of this work is to develop a linked hydrologic-ecologic model that can explain and predict vegetation distribution in a manner that uses physically based hydrologic modeling and accounts for plant stress caused by water-limited and oxygen-limited conditions. Work by Silvertown et al. [1999, 2001] suggests that sum exceedance metrics, which are calculated as the integral of water table depth above an oxygen stress depth threshold and below a water stress depth threshold, can be used as indicators of oxygen- and water-limited conditions. The research presented here uses nonparametric multiplicative regression to determine the depth thresholds which provide the greatest power in predicting vegetation composition and to create empirical models of the hydrologic niches for wet, moist, and dry meadow vegetation communities. These hydrologic niches are then applied to physically based simulations of groundwater dynamics to probabilistically predict the distribution of meadow communities. By applying these relationships between vegetation composition and oxygen stress and water stress to water table dynamics we demonstrate that in high-elevation meadows it is possible to forgo computationally complex variably saturated groundwater flow models and use simulated depths to the water table as a surrogate to determine vegetation composition.

2. Site Description

[8] Tuolumne Meadows is located in the Sierra Nevada mountains of California, United States, within the boundary of Yosemite National Park, at an elevation of 2600 m (Figure 1). The Tuolumne River flows from east to west through the meadow and is in strong hydrologic connection with the meadow aquifer [Loheide and Lundquist, 2009; Loheide et al., 2009; Lowry et al., 2010]. In addition to the Tuolumne River, three ephemeral tributaries feed the meadow within the study area. The upstream watershed feeding the Tuolumne River at the outlet of the meadow has an area of approximately 234 km2, of which approximately 50 km2 of hillslope is tributary to the meadow itself. Mean annual precipitation is approximately 1000 mm, with over 80% of the precipitation falling as snow during the winter months (California Department of Water Resources station ID TUM). Peak snowmelt, which occurs in late spring, is the primary hydrologic driver of the meadow ecosystem. The meltwater is stored within the shallow meadow aquifer and sustains meadow vegetation during the dry summer growing season [Lowry and Loheide, 2010]. Tuolumne Meadows is instrumented with a network of monitoring wells, stream stage recorders and a weather station. Water level measurements were recorded at 55 monitoring wells along 5 transects within the meadow. Manual measurements were collected biweekly in each of the wells during the growing season in addition to hourly water level records at 18 wells that contain pressure transducers (Solinst Canada Ltd., Georgetown, Ontario). Hourly stream stage measurements were taken at five locations along the Tuolumne River and at a single location on each of the tributaries feeding into the Tuolumne River within the meadow study area. Stream discharge measurements were collected biweekly during the growing season from 2007 to 2009 at each of the stream stage sites using an acoustic Doppler sonar (SonTek ADV, San Diego, California) or following the wading methods of Rantz et al. [1982]. Meteorological records collected at Tuolumne Meadows and the nearby Dana Meadows include precipitation, snow water equivalent, air temperature, relative humidity, wind speed and solar radiation (California Department of Water Resources station IDs TUM and DAN).

Figure 1.

Tuolumne Meadows site map. The photograph is taken from the hillslope in the southwest edge of the meadow looking east.

[9] The National Park Service conducted a detailed vegetation survey in 2008 identifying the dominant meadow species at regularly spaced plots throughout the study area. These dominant species range from wet meadow plant communities along the Tuolumne River and in low-lying abandoned river meanders, to dry plant communities in the slightly elevated regions at the edge of the meadow where there is an increased depth to the water table (Figure 2). The indicator species for these communities are listed in Table 1 and are further discussed in section 3.4.

Figure 2.

Observed distribution of wet, moist, and dry meadow vegetation communities represented by the species listed in Table 1. The vegetation survey was conducted by the National Park Service in 2008.

Table 1. Species Used as Indicators of Wet, Moist, and Dry Vegetation Communities
Species NameCommon Name
Wet Community
Carex vesicariablister sedge
Salix eastwoodiaemountain willow
Aster alpigenusalpine aster
Carex subnigricansnearlyblack sedge
Dodecatheon alpinumalpine shooting star
Juncus balticusBaltic rush
Ptilagrostis kingiiSierra false needlegrass
Polygonum bistortoidesAmerican bistort
Horkelia fuscapinewoods horkelia
Trisetum spicatumnorthern oat grass
Stellaria longipeslongstalk starwort
 
Moist Community
Antennaria corymbosaflat-top pussietoes
Solidago multiadiataalpine goldenrod
Calamagrostis brewerishorthair reedgrass
Vaccinium caespitosumdwarf bilberry
Castilleja lemoniiLemmon's Indian paintbrush
 
Dry Community
Artemisia tridentatabig sagebrush
Pinus contortalodgepole pine
Carex rossiiRoss' sedge
Carex filifoliathreadleaf sedge

3. Methods

[10] Due to the interconnected nature of snowmelt, stream, and groundwater processes on meadow hydrology, it was necessary to link four independent models in order to simulate the drivers controlling the distribution of meadow vegetation (Figure 3). Watershed-scale processes, particularly snowmelt and hillslope hydrology, were simulated using the Distributed Hydrology Soil Vegetation Model (DHSVM) [Wigmosta et al., 1994]. The Hydrologic Engineering Centers River Analysis System (HEC-RAS) [Brunner, 2010] was used to develop relationships between stream discharge and stream stage at various locations along the channels; these relationships were applied to stream discharge records simulated by the watershed-scale hydrologic model to create time variable records of stream stage that were then used as boundary conditions for a two-dimensional model of groundwater flow (COMSOL). At the outer edge of the meadow, DHSVM simulated hillslope fluxes were applied as specified flux boundary conditions to the COMSOL groundwater flow model. In addition, the DHSVM meadow snowmelt was simulated as a source term representing groundwater recharge in the COMSOL groundwater flow model. Nonparametric multiplicative regression was performed using current vegetation composition and simulated water levels at 222 discrete locations within the meadow to create an empirical habitat model of the hydrologic niches of wet, moist and dry vegetation communities. The resulting habitat models were combined with the groundwater flow model to predict vegetation communities at all locations within the meadow.

Figure 3.

Model hierarchy for Tuolumne Meadows simulations. The watershed and hydraulic models supply boundary conditions to the groundwater flow model. Nonparametric multiplicative regression is performed on the plot level vegetation data to determine the hydrologic niches of each vegetation community. The resulting hydrologic niche and simulated hydrologic fluctuations from the groundwater flow model are used to create spatially continuous, probabilistic maps of vegetation community distribution.

3.1. Watershed Hydrologic Model

[11] DHSVM, which is described by Wigmosta et al. [1994, 2002], was used to simulate the energy and water balance for the entire watershed feeding Tuolumne Meadows (234 km2) using a 150 m grid size and three hour time step. Distributed vegetation, soil, and geology data were obtained from the National Park Service Archives. Meteorological data was collected at both Dana Meadows, which feeds the Tuolumne River upstream of Tuolumne Meadows, and Tuolumne Meadows (California Department of Water Resources station IDs DAN and TUM). Spatially distributed temperature and precipitation input for the model were estimated using a constant lapse rate of −6.5°C km−1 [Lundquist and Cayan, 2007] and monthly normal precipitation weighting from the Precipitation on Independent Slopes Model (PRISM) [Daly et al., 1994, 2002], respectively. Within each cell, an energy balance, based on the spatially distributed meteorological conditions, is calculated to determine snowmelt and ET rates. Precipitation and snowmelt applied at the land surface within DHSVM are routed through the watershed as both surface and groundwater flow. Routing is based on the maximum land surface gradient from a 30 m digital elevation model, and water is passed to one or more of the eight surrounding cells. Water that infiltrates and drains from the soil layers becomes groundwater within DHSVM and is then routed in the subsurface based on a maximum hydraulic gradient using a Darcian flow assumption. Both groundwater and surface water that reaches a stream cell is considered captured by the stream and is routed as streamflow through the watershed. The model was calibrated to observed stream discharge within the watershed. At the edge of Tuolumne Meadows, groundwater flows directly from the hillslope aquifer to the meadow aquifer, and overland flow is assumed to enter the aquifer as focused recharge when the hillslope flux reaches the high-permeability meadow sediments. Both hillslope groundwater flow and overland flow are simulated by DHSVM and are passed to the groundwater flow model where they are treated jointly as a specified flux boundary condition as detailed by Lowry et al. [2010].

3.2. Streamflow to Stream Stage

[12] Estimation of stream stage within the Tuolumne River and the three tributaries entering the meadow is based on rating curves generated using survey data of the channel geometry. Using a real time kinematic global position system (RTK-GPS), 6 transects were surveyed along the Tuolumne River in addition to 13, 14, and 17 transects along Budd, Delaney, and Unicorn creeks, respectively. At each transect, synthetic rating curves were generated with HEC-RAS by simulating stream stage for the specific cross-section geometry at stream discharge rates that spanned the entire range of expected flow conditions. The Manning's roughness coefficient for each transect was selected by using paired stream discharge and stage measurements as calibration targets. Calibrated rating curves were then used to predict stage as a function of time and cross-section location using the stream discharge predicted by the watershed-scale hydrologic model. Stream stage was predicted between the transects by linear interpolation, and these interpolated stream stage records were used as specified head boundary conditions in the groundwater flow model described below.

3.3. Groundwater Flow Model

[13] The groundwater flow system within Tuolumne Meadows is simulated using COMSOL Multiphysics (COMSOL, Inc. Burlington, Massachusetts), following the approach of Li et al. [2009], to solve the Boussinesq equation [Bear, 1972], which represents a transient, unconfined, two-dimensional groundwater flow system. Use of the simplified two-dimensional groundwater flow model is based on an assumption of essentially horizontal groundwater flow within the meadow, which is appropriate for an aquifer with a relatively large areal extent (>2000 m2) as compared to its thickness (<5 m). The model domain covers an area of approximately 2.2 km2, and a period from October 2002 to September 2009 was simulated. Input to the groundwater flow model included historical precipitation data, model calibrated streamflow (DHSVM) that was converted to stream stage using HEC-RAS-generated rating curves, simulated hillslope water fluxes (DHSVM), and simulated snowmelt resulting in recharge within the meadow (DHSVM). The groundwater flow model simulates hourly water table position across a variable-resolution finite element grid, with maximum computational time steps never exceeding 1 h. Initial groundwater conditions within the meadow are specified as fully saturated at the beginning of an initialization period which runs from October 2002 to May 2003 to reach suitable initial conditions for the simulation period which runs from May 2003 to September 2009. The flow model is calibrated to manual water level measurements taken from June 2006 to August 2009. These manual water level measurements were collected biweekly, during the 2006–2009 growing seasons, at each of the 55 active monitoring wells. Groundwater recharge is treated as a source term resulting from both rain and snowmelt, whereas overland flow, exfiltration, and evapotranspiration of groundwater are treated as sink terms. The generation of overland flow, which is represented as a sink term from the groundwater modeling domain, occurs at locations where the water table rises above the land surface or when precipitation falls on an already saturated area. Precipitation data is based on daily measurements within the meadow (California Department of Water Resources station ID TUM). Snowmelt within the boundaries of the meadow is based on the simulated melt from the DHSVM model. The FAO- 56 Penman-Monteith method [Allen et al., 1998] is used to calculate the potential evapotranspiration (PET) based on atmospheric conditions. A detailed explanation of the linking between the groundwater flow model and DHSVM, and additional information on sources and sinks to the groundwater flow model can be found in work by Lowry et al. [2010].

[14] The distribution of the hydraulic conductivity field within the model domain is based on the conceptual model of meadow stratigraphy described by Lowry et al. [2010]. The region closest to the Tuolumne River represents a high-conductivity alluvial region as a result of significant reworking of the sediment due to changes in the river's position through time. This region contains numerous abandoned channels consisting of coarse-grained sands and gravels. The outer regions, at the edge of the model, consist of lower conductivity colluvium transported from the hillslope. The hydraulic conductivity zones are based on lidar data from which subtle terraces and transitions in topographic texture revealed boundaries between lithologic zones. The groundwater flow model is calibrated by varying the hydraulic conductivity of these zones to minimize the difference between manual water levels and simulated water levels, which are reported in terms of root-mean-square error. Continuous water level data from pressure transducers in a limited number of wells is qualitatively used to ensure adequate simulation of the late growing season water level recession as the meadow dries out.

3.4. Vegetation Modeling

[15] A vegetation survey was conducted by the National Park Service at 222 sites within the meadow study area with a uniform grid spacing of approximately 100 m. At each survey point the percent cover of the three most dominant vegetation species in the plot was recorded. Twenty species were identified as indicators of wet, moist, and dry meadow vegetation communities (Table 1). The percent cover of the dominant species in each vegetation community was summed and is displayed as a pie graph for each plot in Figure 2. In some locations there are regions of bare ground and/or species that were not indicators of one of the three vegetation communities in Table 1 because they were either rare species or were commonly found in multiple communities (wet, moist, and dry). The percent cover of bare ground and nonindicator species are represented as a transparent section within the pie graphs (Figure 2). The percent cover data for each of the three vegetation communities were then transformed to presence or absence data indicating that one of the twenty species used as community indicators was dominant at the vegetation survey site.

[16] The dominant species grouped into the wet, moist and dry vegetation communities are evaluated based on the associated seasonal water table response at each location. These data create a matrix of dominant vegetation at each of the vegetation survey locations and a matrix of the corresponding water level response at the vegetation survey locations from the 2003–2009 simulated water levels. Relationships were determined between hydrologic indicators (i.e., metrics describing the water table regime) and vegetation communities (i.e., wet, moist, and dry) using nonparameteric multiplicative regression. This method quantifies the relationships between important predictors and the observed vegetation response. These relationships are determined using the multiplicative habitat modeling software HyperNiche2 [McCune and Mefford, 2008]. This method allows us to determine the estimated likelihood of dominance of observed wet, moist and dry vegetation community indicators based on hydrologic conditions.

[17] In this study we consider two sets of hydrologic indicators as potentially representative predictors. The first set of hydrologic indicator metrics is introduced to quantify the extent to which a hydrologic regime is too wet and causes oxygen stress in unadapted plants, and the second set of metrics quantifies the water stress imposed by dry conditions. These parameter sets are both based on a sum exceedance metric [Gowing et al., 2002; Silvertown et al., 2001; Silvertown et al., 1999] that represents both the amount of time the water level is beyond a specified depth threshold as well as the magnitude beyond the depth threshold. Graphically, this sum exceedance metric can be shown as the integral between the water level curve and the depth threshold (see shaded areas in Figure 4). The wet hydrologic indicator is determined relative to a depth below land surface representing the oxygen stress threshold. As a result of a shallow water table (i.e., above the oxygen stress depth threshold), oxygen within the rooting zones is low due to saturated conditions. Under these conditions, only water tolerant species can survive for extended periods. The dry hydrologic indicator is determined based on conditions of water stress. During periods where the water table drops, vegetation has limited access to groundwater, and there is limited soil moisture for root water uptake. In climates with minimal summer precipitation, these conditions produce water stress, resulting in environments that support only drought-tolerant species. The range of hydrologic conditions investigated in this study represents oxygen stress and water stress depth thresholds ranging from 0 to 200 cm below land surface. Oxygen stress and water stress metrics are calculated at 5 cm depth increments within their respective ranges plus an additional point at 2.5 cm oxygen stress. The period of time when oxygen stress and water stress control vegetation growth within the meadow is defined here as the hydroecologically significant period. This hydroecologically significant period is comparable to the growing season but is based on the average air temperature within the meadow. The hydroecologically significant period is defined based on air temperature in order to use both historic records as wells as predictive models of future climate to approximate the growing season within the meadow. The start of the hydroecologically significant period is initiated when there are 10 consecutive days with the average air temperature above 2°C (36°F) and concludes when the average air temperature is below 2°C (36°F) over four consecutive days. The hydroecologically significant period for plant growth runs from 11 May to 23 October, based on the average hydroecologically significant period start and end dates from 2003 to 2008 within Tuolumne Meadows.

Figure 4.

Hydroecologically significant period sum exceedance values based on oxygen stress and water stress depth thresholds.

[18] In order to identify the most significant predictors of vegetation communities within the meadow, nonparametric multiplicative regression is first used as a scoping tool. This is meant to narrow down the range of hydrologic predictors to the two parameters (one oxygen stress and one water stress metric) that best explain the observed variability in the vegetation data. Initially, the complete predictor matrix is applied to the habitat model with the full range of hydrologic parameters (i.e., all oxygen stress and water stress sum exceedance values based on all possible thresholds 0–200 cm depth) with a local mean-Gaussian model form. The local mean-Gaussian model form applies a weight based on a Gaussian probability function centered on the local mean, which reduces the chance of over fitting these observations. This allows identification of the hydrologic predictors that best describe the vegetation communities based on the goodness of fit. Goodness of fit is quantified using log β, which is calculated using a leave-one-out strategy as the log of the ratio of likelihood of the observed values predicted by the nonparametric multiplicative regression (posterior) model and that under a naïve (prior) model only based on average frequency of the data set [McCune and Mefford, 2008; Yost, 2008]. Although log β is primarily an indicator of the strength of the relationship between predictor and response variables, it is affected by both sample size and the ratio of presence or absence in vegetation plots [Jeffreys, 1935; Yost, 2008]. Jeffreys [1935] suggested the following guidelines for interpreting how log β values define goodness of fit: 0–0.5, weak; 0.5–1, substantial; 1–2, strong; >2, decisive.

[19] Using the log β values from each univariate model, we rank each hydrologic predictor for each of the three vegetation communities from highest to lowest log β in order to determine which hydrologic parameters best describe our vegetation communities. The ranking number for each predictor for the wet, moist and dry communities are then summed for each depth threshold in order to determine a new cumulative rank score called the sum rank of log β. This sum rank of log β describes a weighted measure of the overall best single hydrologic predictor of the wet, moist and dry vegetation communities that is capable of predicting a given vegetation community's probability of occurrence as a dominant community. As a result of the negative correlation between the hydrologic predictors representing wet and dry conditions, it is beneficial to choose a pair of predictors with low correlation and minimum sum rank of log β when determining the two best hydrologic predictors.

[20] Using only the two best hydrologic predictors, one representing wet conditions and one representing dry conditions, we perform another nonparametric multiplicative regression to determine each community's hydrologic niche (the likelihood of dominance of each of our vegetation communities across this limited parameter space). Results from this second regression, using the two best hydrologic predictors, are then evaluated based on the product of the wet, moist and dry communities log β values at a constant tolerance for each of the three communities. The models with the highest product log β (not including any models with negative log β) are used to predict vegetation patterning based on the simulated water levels within the meadow.

4. Results and Discussion

4.1. Groundwater Flow Calibration

[21] Calibration of the groundwater flow model, which is linked to both the watershed and channel hydraulics models, is evaluated based on the root-mean-square error (RMSE), comparing simulated water levels to observed water levels at each of the well locations. Observed water levels used for calibration are a result of manual water level measurements (see black circles in Figure 5) taken during the growing season from 2006 to 2009. Simulated water levels are also qualitatively evaluated based on seasonal variability in observed water levels using hourly pressure transducer records at a limited number of wells (see blue lines in Figure 5). Based on a total of 55 wells used for calibration, the average RMSE was 0.208 m, with a maximum RMSE of 0.412 m and a minimum RMSE of 0.068 m. These results, based on the RMSE, show on average our simulated water levels are within 0.21 m of the true water levels within the meadow for the growing seasons of 2006–2009.

Figure 5.

Distribution of root-mean-square error (m) of the calibrated groundwater flow model and selected comparisons of simulated and observed water levels in Tuolumne Meadows.

4.2. Predictors of Vegetation

[22] A wide range of hydrologic metrics characterizing oxygen stress and water stress using different thresholds were considered as potential predictors of wet, moist and dry meadow vegetation communities (Figure 6). Prior to selecting which hydrologic metric(s) would be the most suitable for predicting vegetation composition, we determined the degree to which metrics are correlated to one another (Figures 7a–7d). As expected, an oxygen or water stress sum exceedance value at a particular threshold depth is very highly and positively correlated with an oxygen or water stress sum exceedance value at a threshold that is only slightly deeper or shallower, but as the threshold depths diverge, this correlation weakens. In fact, both the correlation coefficients for the oxygen or water stress sum exceedance value range from unity when a metric with a given threshold is compared to itself (diagonals in Figures 7c and 7d), to ∼0.6–0.7 when the thresholds differ significantly (deep versus shallow). Conversely, a negative correlation is observed when oxygen stress (wet) metrics are compared with water stress (dry) metrics because a site experiencing high water stress is likely to experience low oxygen stress and vice versa (Figure 7a). In addition to having negative values, the wet/dry correlation coefficients have lower absolute values than either the wet/wet or dry/dry predictors. This indicates a potential advantage to using both a dry and wet metric to predict the range of vegetation communities when a pair of oxygen and water stress sum exceedance metrics that have only a weak or moderate correlation are selected. If all of the parameters are highly correlated, then there is little benefit from choosing one metric over another, and little can be gained from using more than one metric. Thus, we have chosen to only consider possible predictor pairs for which the absolute value of the correlation coefficient is less than 0.8.

Figure 6.

Strength of relationship between vegetation communities and hydrologic predictors using (a) water stress sum exceedance value and (b) oxygen stress sum exceedance values across a range of thresholds. The shaded bars, from dark to light, represent the predictive power (log β) of a given sum exceedance threshold for the dry, moist, and wet communities, respectively. The black line represents the sum rank of all three vegetation communities with the minimum value representing the best predictor.

Figure 7.

(a) Correlation coefficient for all sum exceedance thresholds, ranging from 0 to 200 cm for oxygen stress and water stress. (b) Cross section, taken along the dashed line in Figure 7a, of correlation coefficient at water stress threshold of 55 cm and the full range of oxygen stress depth thresholds quantifying the correlation at the local (20 cm) and global (175 cm) minimum of the sum log β. Correlation coefficient for all sum exceedance thresholds, ranging from 0 to 200 cm for (c) oxygen stress and (d) water stress. A correlation coefficient approaches 1 as distance between dry/dry or wet/wet sum exceedance thresholds become smaller. Negative values represent conditions when the variables are negatively correlated and occur when comparing oxygen stress and water stress metrics.

[23] The predictive power of water stress and oxygen stress metrics were quantified at each depth threshold individually using the log likelihood ratio (log β). Of the potential water stress metrics, the best predictor of wet vegetation is obtained at a water stress depth threshold of 55 cm (Figure 6a), while the best depth threshold for the water stress predictor of dry vegetation is 50 cm below land surface (Figure 6a). However, according to this analysis, the moist community is relatively insensitive to the water stress. Ranking each of the predictors from best to worst and taking the sum for all three vegetation community predictor ranks, it is found that the best predictor among all three communities for water stress is 55 cm below land surface (see minimum value of solid line in Figure 6a). This depth threshold was selected as the single metric for characterizing water stress based on the minimum rank; however, it is important to note that the log β is relatively high (>3.0) for both the wet and dry communities individually (Table 2). It is convenient to have a single depth threshold; however, based on the predictor ranks it is likely that a depth range from 40 to 65 cm (Figure 6a) would produce similar results in predicting communities affected by water stress (see correlation coefficient Figure 7d).

Table 2. The Log Likelihood Ratio of Hydrologic Niches for Each Vegetation Community and the Associated Tolerance Values
Vegetation Community (Response Variable)log BPredictor Variable 1aPredictor Variable 2a
  • a

    Tolerance is given in parentheses.

Dry3.52oxygen stress 0.20 m (6.06)water stress 0.55 m (45.2)
Moist0.61oxygen stress 0.20 m (6.06)water stress 0.55 m (45.2)
Wet3.92oxygen stress 0.20 m (6.06)water stress 0.55 m (45.2)

[24] The relationship between the oxygen stress and vegetation communities shows the best predictive power at 200 cm, 115 cm and 200 cm for dry, moist, and wet communities, respectively (Figure 6b). The sum rank of all three vegetation communities under oxygen stress is at a minimum at 175 cm below land surface (see minimum of solid line in Figure 6b). However, this oxygen depth threshold is highly correlated (correlation coefficient equal to −0.92) to the water stress depth threshold at 55 cm (Figure 7b). As a result, using an oxygen depth threshold of 175 cm will provide little new information for predicting vegetation communities. There is a local minimum in the sum rank of log β at a oxygen depth threshold at 20 cm that has a correlation coefficient of −0.67 with respect to the 55 cm water stress depth threshold (Figure 7b). This new local minimum depth threshold for oxygen stress will give addition information for predicting vegetation communities, which was not available at the global minimum. As a result the 20 cm oxygen stress depth threshold is used for all three vegetation communities.

4.3. Hydrologic Niches

[25] The global minimum values for the sum rank of log β for water stress and local minimum for oxygen stress parameters were used to select the best predictors, which were sum exceedance oxygen stress with a 20 cm threshold and sum exceedance water stress with a 55 cm threshold (see Figure 6). These predictors were then used to create a hydrologic niche model in HyperNiche for each of the three vegetation communities (Figure 8). The contour map shows the probability of occurrence of a dominant species in a vegetation community (i.e., Pdry, Pmoist, and Pwet) based on current hydrologic conditions from the calibrated groundwater flow model years 2003–2008. These probabilities are based on the results of using the local mean-Gaussian weighting function in the nonparametric multiplicative regression model to fit a predictive model to the observed vegetation data. Regions classified as out of range (see gray area in Figure 8) are regions of the parameter (predictor) space where too few vegetation data are present to reliably create a nonparametric multiplicative regression model. This out of range region does not mean that a given community will not be present within the hydrologic niche, it simply means that there are insufficient data to calibrate the model and is a limitation of the data set. In some cases, such as high water stress and low oxygen stress, it is very likely that this out of range area would be dominated by dry vegetation and may even contain drier vegetation communities than were observed in the data set. In other cases, such as high oxygen and high water stress, novel hydrologic conditions would be experienced and no expectation of vegetation composition exists. The log β values for the best hydrologic predictors are given for each vegetation community in Table 2.

Figure 8.

Probability of dominance for vegetation communities based on (a) dry, (b) moist, and (c) wet hydrologic conditions.

[26] Results from the nonparametric multiplicative regression for the dry vegetation community show a greater dominance under low oxygen stress and high water stress conditions. In locations of low oxygen stress (<15 m day), Pdry increases from 20% to around 60% with an increase in the sum exceedance value for water stress from a minimum value of 0 m day to a maximum of 240 m day (Figure 8a). At low water stress conditions (<80 m day), the Pdry drops from around 20% to 0% as a result of an increase in oxygen stress. Pdry varies gradually and shows the tradeoffs of changes in water stress and/or oxygen stress on the probability of dominance (Figure 8a). Similar relationships have been observed by Silvertown et al. [2001] regarding the tradeoffs between wet and dry conditions on the distribution of vegetation communities.

[27] Unlike the dry community, the maximum Pmoist does not occur at the extremes of water stress or oxygen stress (Figure 8b). The maximum Pmoist occurs within the meadow as a result of a combination of mild water and oxygen stress. This moist vegetation appears to survive under a wider range of hydrologic conditions. As a result, the probability map shows a relatively constant probability of 40%–60% dominance over most of the hydrologic conditions, which declines substantially under high oxygen stress (>20 m day). The maximum Pmoist occurs under intermediate hydrologic conditions when dominance of both dry and wet communities declines.

[28] Pwet has the highest probability of occurrence under large oxygen stress and low water stress (Figure 8c). Pwet reaches almost 100% probability and is likely due to these species' adaptations for tolerating extended periods of saturation during the growing season. To some extent, the hydrologic niche of the wet community is an inverted representation of that of the dry community; however it appears that the wet community is more tolerant of water stressed conditions than the dry community is of oxygen stressed conditions and thus survives under the widest range of hydrologic conditions of the three communities considered (compare Figures 8a to 8c). The relationships developed here are undoubtedly affected by the nature and thickness of fine grained sediment layers. It is likely that if the site had more fine grained sediments the vegetation response would be less affected by water stressed conditions due to greater water retention in the soil column but would be more sensitive to oxygen stress conditions for unadapted vegetation due to reduced oxygen diffusion. The opposite would be true for sites with coarser grained sediments. Spatial variability of soil properties could adversely affect the predictive capability of the hydroecologic model we have developed, and in cases with strong heterogeneity, prediction of vegetation composition could be improved by incorporating spatially variable soil information.

4.4. Water Stress and Oxygen Stress

[29] The observed vegetation response to both oxygen stress and water stress demonstrates an ability to explain the current vegetation communities at discrete locations where vegetation surveys were taken. Application of these relationships involves expanding the predictions from point locations to the whole meadow. Predicting vegetation composition spatially requires not only the hydrologic niches shown in Figure 8, but also maps of the sum exceedance values for water stress and oxygen stress at 0.55 m and 0.20 m depths, respectively. These maps of sum exceedance values for oxygen stress (Figure 9) and water stress (Figure 10) can be generated using the spatially and temporally continuous results from the groundwater flow model. The fine-scale structure of these maps is the result of both the high-resolution lidar data representing the topography and the simulated water table, which are used to calculate depth to the water table. This high-resolution mapping allows for steep gradients in hydrologic conditions to be represented over short distances while also capturing broad trends across the entirety of the meadow. Figure 9 shows higher oxygen stress values are observed in zones near the Tuolumne River, in intermittent channels, and in abandoned channels, particularly in the southwestern portion of the meadow. Figure 10 shows that dry regions primarily exist along the edge of the meadow and in the region of the meadow near Delaney Creek where the land surface is at a slightly higher elevation, creating a greater depth to the water table. In combination, the water stress and oxygen stress metrics indicate that the terrace near Delany Creek has the driest hydrologic regime, the southwestern part of the meadow has the wettest regime, and intermediate conditions occur in the region surrounding and to the north of Unicorn Creek. Note that some combinations of sum exceedance values may be out of the range of the vegetation probability distribution relationships (i.e., gray area in Figure 8).

Figure 9.

Spatial distribution of simulated oxygen stress across Tuolumne Meadows based on the sum exceedance value representing water levels above a depth of 0.20 m.

Figure 10.

Spatial distribution of simulated water stress across Tuolumne Meadows based on the sum exceedance value representing water levels below a depth of 0.55 m.

4.5. Vegetation Maps

[30] Combining results from the habitat model (Figure 8) and maps of oxygen stress (Figure 9) and water stress (Figure 10), we estimate the probability of dominance for a given vegetation community over the full extent of Tuolumne Meadows (Figures 1113). The dry vegetation has a probability of dominance ranging from 0 in the abandoned stream channels to 0.4–0.5 in the higher-elevation regions on the terrace near Delaney Creek (Figure 11). The moist vegetation has a probability of 0.4–0.6 in a majority of the meadow (Figure 12); this result supports the notion of a vegetation community that is tolerant, to some degree, of both dry and wet conditions. The wet community map shows a probability of dominance ranging from 0.2 to 0.4 along the edges of the meadow and in the higher-elevation region in the northwest (Figure 13). Within the abandoned channels and along the Tuolumne River, particularly in the western half of the meadow, the probability of dominance of wet vegetation is as high as 0.8.

Figure 11.

Probability of occurrence as a dominant dry vegetation community.

Figure 12.

Probability of occurrence as a dominant moist vegetation community.

Figure 13.

Probability of occurrence as a dominant wet vegetation community.

5. Conclusions and Implications

[31] A range of environmental factors control the distribution of vegetation communities within mountain meadows. In the work presented here, we have identified two of these factors. The first hydrologic metric quantifies wet conditions that induce oxygen stress with a sum exceedance value above a 0.20 m depth threshold, and the second hydrologic metric quantifies dry conditions that induce water stress with a sum exceedance value below a 0.55 m depth threshold. Many other conditions also affect the distribution of vegetation; however, these wet and dry indicators are shown to be useful predictive metrics that describe much of the observed variation in vegetation composition. It is well known that other variables including soil moisture, soil texture [Dodd et al., 2002], soil pH [Chytrý et al., 2003], redox conditions [Howes et al., 1981] and nutrients [Tilman, 1987] affect vegetation composition (see Callaway [1995] for review). The success of using only two carefully selected hydrologic metrics may largely be due to strong correlations between these hydrologic metrics and other controlling factors. For example, increased soil moisture is positively correlated to shallow water table conditions, fine-grained texture, and reduced oxygen availability. It could be argued that precision in the prediction of vegetation dominance only comes with the incorporation of all variables controlling vegetation. However, the additional uncertainty and complexity with each new variable would likely minimize the marginal improvements in prediction capability that could be gained. Extremely large computational demands for such a model coupling variably saturated groundwater flow, nutrient transport, and plant competition or growth on top of the current models (i.e., hydrologic watershed model, streamflow routing, and groundwater flow model) would limit the transferability of such an approach.

[32] Even considering the simplifying constraint of limiting the predictive variables to a single set of wet and dry metrics to predict vegetation dominance, there are multiple metrics to be evaluated that are associated with varying depth thresholds. This research determined that 0.20 m and 0.55 m were the most appropriate sum exceedance depth thresholds for oxygen stress and water stress, respectively. Even though these oxygen stress and water stress sum exceedance depth thresholds were determined empirically, they likely have physical meaning in terms of controlling vegetation in high-elevation meadows. Choosing a wet and dry indicator facilitates the evaluation of the extreme conditions to which plants are exposed; these seasonal extremes include high water levels caused by snowmelt in the spring and drought conditions at the end of the growing season when the tributary streams within the meadow dry up. The depth thresholds used to calculate the sum exceedance values are within the root zone of typical meadow species [Martin and Chambers, 2002]. Martin and Chambers [2002] observed the greatest root density at depths of 0.25 m or shallower, which corresponds to the threshold determined here for the onset of oxygen stress. When the water table is above this depth threshold, the majority of the roots are affected by saturated conditions that impede oxygen diffusion and limit root respiration. Similarly, at depths below 0.55 m, very few roots are present, especially for species in the wet vegetation community, and plants are no longer capable of accessing water from the water table and associated capillary fringe. The rooting density profile of each community is likely adapted, to some extent, to their specific hydrologic niche [see Guswa, 2008]. For example, water tolerant plants can tolerate shallow water table conditions and tend to concentrate their roots very near the land surface. In the case of dry vegetation communities, these plants outcompete other communities as a result of deep roots that can access water greater than 0.55 m below the land surface.

[33] The nonparametric multiplicative regression performed in this study resulted in hydrologic niche segregation of three meadow vegetation communities. Similar to Silvertown et al. [1999, 2001], these niches revealed a tradeoff between oxygen stress and water stress metrics as a determinant of vegetation type. The dry meadow vegetation community had the greatest likelihood of occurrence in regions with high water stress, but little to no oxygen stress. The opposite was true for the wet vegetation community, which occurs in regions with high oxygen stress and low water stress. In between these two extremes, the moist vegetation community is found, in regions with moderate water stress and moderate oxygen stress. Modeling the hydrologic niche of wet and dry communities resulted in relatively minimal niche overlap. For the moist community, while the hydrologic niche reveals a high-probability region at moderate water stress (170 m day) and moderate oxygen stress (15 m day) in Figure 8b, there is moderate overlap with the hydrologic niches of wet and dry communities where predictions are less certain, and vegetation communities of mixed composition may occur.

[34] Linking a watershed-scale hydrologic model to a fine-scale (meadow) groundwater flow model is critical for simulating both historical and future water table fluctuations. Without an understanding of the temporal variability of these water table fluctuations it is impossible to predict vegetation communities within a riparian zone. Variability in both space and time in the movement of water within the contributing watershed has been shown to have significant effects on groundwater levels within the meadow [Lowry et al., 2010]. In high-elevation meadow environments the flux of water from the hillslope as a result of snowmelt is particularly critical for understanding seasonal water table fluctuations and must be accounted for by transient, and spatially variable, boundary conditions at the edge of the meadow. In addition, the timing of cessation of flow in ephemeral tributaries flowing through the meadow, which is also controlled by watershed dynamics, was a primary control on groundwater recession that begins when these channels become dry. It is possible that in some watersheds these spatial and temporal fluxes have less variability and can be simulated as a steady state condition, but in our study, large-scale, watershed processes exhibited strong control on local, riparian groundwater dynamics.

[35] Results presented here have applications for management of meadows within the Sierra Nevada mountains. Predictive models of future climate change within this region already exist and can be used to determine the impact of nonstationary hydrologic conditions within high-elevation meadows. Using these data, it would be possible for resource managers to identify potential shifts in vegetation communities within Sierra Nevadan meadows on the basis of simulated wet and dry hydrologic metrics described here. This method also facilitates hypothesis testing of potential restoration strategies within these riparian ecosystems in order to maintain and/or restore specific vegetation communities. This predictive restoration modeling approach could save time and financial resources by identifying both meadows that will be most affected by climate change as well as restoration techniques most suitable to maintain or enhance ecological function.

[36] On the basis of his travels through Yosemite National Park, John Muir, an advocate for preservation of wilderness in the Sierra Nevada, wrote “When we try to pick out anything by itself, we find it hitched to everything else in the universe” [Muir, 1911, p. 211]. In this case, we find riparian meadow vegetation to be controlled by processes at the intersection of surface hydrology, groundwater hydrology, ecology, and geomorphology. The observed vegetation in a riparian zone cannot be explained or predicted without a system level understanding of the hydrology at scales ranging from the local riparian zone, to the hillslope, to the watershed. Along the climate-hillslope–riparian zone–stream continuum, cascading hydrologic flows control the timing, magnitude, duration, and spatial distribution of water excess and shortage. The resulting water and oxygen stress conditions control the suitability of riparian areas for specific vegetation communities and define their hydrologic niche. This tight linkage between hydrology and ecology offers not only a means for prediction of vegetation patterns as demonstrated in this research, but could also be applied in the opposite direction as an approach for synthesis. Because riparian vegetation is influenced by hydrologic processes upstream and upslope, its distribution and composition may offer clues into the hydrologic processes of the watershed as well as insight into the vulnerability of the system to future change.

Acknowledgments

[37] This material is based upon work supported by the National Science Foundation under grants CBET-0729838 and CBET-0729830. We would like to thank the National Park Service for allowing us to conduct this research in Yosemite National Park and for providing access to their facilities. We would like to thank Nicoleta Cristea, Jeff Deems, Fred Lott, David Cooper, Evan Wolf, and Jim Roche for assistance with model setup, well installation, and data acquisition. This work would have been impossible without the extensive vegetation survey conducted by Elle Kramer, Joy Fischer, Kate Wilkin, Liz Ballenger, and Lusetta Nelson from the National Park Service. We would also like to thank Eric Booth for the many insightful conversations regarding vegetation modeling. Finally, we would like to thank three anonymous reviewers and the associate editor for their helpful comments which helped to improve the manuscript.

Ancillary