Climate change is expected to bring warmer temperatures and increased hydrologic extremes including more frequent droughts and longer inter-storm periods (e.g., Seager et al., 2009; Pachauri et al., 2014). Although warming-induced lengthening of the growing season and increased atmospheric CO₂ concentrations have generally facilitated vegetation growth (e.g., Keeling et al., 1996; Myneni et al., 1997; Keenan et al., 2014), enhanced hydroclimate variability has often led to increased periods of plant water stress (Anderegg et al., 2012), species-specific drought responses (Clark et al., 2011; Brzostek et al., 2014), xylem cavitation (Hoffmann et al., 2011), and subsequent widespread tree mortality (Adams et al., 2009; Klos et al., 2009; McDowell and Allen, 2015). For these reasons, water availability has become more widely recognized as a key driver of ecosystem response to climate change than before, in terms of carbon cycling (van der Molen et al., 2011), vegetation water use (Wullschleger and Hanson, 2006), and species distributions (Stephenson, 1990; Crimmins et al., 2011; VanDerWal et al., 2013) across different scales.

In addition to regional climate variability, topography provides variation in hydroclimate through topoclimate variation, lateral soil water redistribution, and differences in soil depth and storage. The interaction of climate and topography supports a wide range of microclimate and soil moisture conditions for both xeric and mesic tree species (Dobrowski, 2011; McLaughlin et al., 2017), and promotes high productivity and biodiversity especially in mountain forest ecosystems (Davis and Goetz, 1990; Beckage and Clark, 2003; Emanuel et al., 2011; Clark et al., 2014). In mountainous forested catchments, water consumption by vegetation downslope (green water) usually depends on water flow generation from upslope (blue water). Hillslope-riparian-stream connectivity by dominant subsurface hydrologic flow processes plays a key role in runoff generation (Jencso et al., 2009; Detty and McGuire, 2010; McGuire and McDonnell, 2010) and soil moisture organization at the watershed scale (Western et al., 1999; Ali and Roy, 2010). Downslope flows can mitigate the impact of droughts in convergent topographic areas (e.g., Hawthorne and Miniat, 2018). Therefore, topography-mediated soil moisture conditions provide an important control on the patterns of forest water use (Tromp-van Meerveld and McDonnell, 2006; Mackay et al., 2010), periodic water stress (Vicente-Serrano et al., 2008; Ford et al., 2011), tree growth (Clark et al., 2014; Elliott et al., 2015; Martin-Benito et al., 2015), mortality (Berdanier and Clark, 2016; Tai et al., 2017), and species distribution (Day et al., 1988; Crimmins et al., 2011). It is therefore important to understand climate-vegetation-topography-hydrology interactions and feedbacks to predict landscape-scale responses of forest ecosystems to ongoing climate change (e.g., Hoylman et al., 2018).

Forest vegetation often adjusts leaf area amount and duration in response to water and nutrient availability (e.g., Nemani and Running, 1989), which is mediated by lateral hydrologic flows along topographic gradients (Hwang et al., 2009). Hydrologic partitioning between localized water use and drainage often influences emergent vegetation dynamics in space and time (Thompson et al., 2011), which can be used as a simple diagnostic to infer underlying water balance patterns along hydrologic flow paths (Brooks et al., 2011; Voepel et al., 2011; Hwang et al., 2012; Hoylman et al., 2018). Furthermore, close interactions between hydroclimate variability and vegetation dynamics (e.g., large-scale mortality, growing season duration, etc.) have been demonstrated by measurable shifts in seasonal streamflow dynamics and forest water yield at the watershed scale (Adams et al., 2012; Bearup et al., 2014; Hwang et al., 2014, 2018; Creed et al., 2015; Kim et al., 2017, 2018). However, there have been few studies on feedbacks between climate change, lateral soil moisture distribution, and long-term forest canopy patterns at the watershed scale.

Following decades of active fire suppression in the eastern United States, fire-intolerant, mesophytic tree species (e.g., red maple and tulip poplar) have increased in southern Appalachian forests compared to fire-tolerant, xerophytic oak and hickory, often called as forest “mesophication” (Nowacki and Abrams, 2008, 2015). These long-term forest mesophication trends also have a great implication in understanding forest responses to frequent droughts under climate change. Red maple and tulip poplar typically exhibit as isohydric stomatal responses to declining soil water potentials, while oaks are typically anisohydric (Choat et al., 2012; Klein, 2014; Roman et al., 2015; Hwang et al., 2017). Anisohydric trees allow leaf water potential to drop as soil dries, so they maintain greater stomatal conductance to continue gas exchange under moderate droughts (Martínez-Vilalta et al., 2014). In contrast, trees that do not allow leaf water potential to drop as soil dries are known as isohydric, which close stomata to maintain a stable leaf water potential at the expense of CO₂ uptake. For this reason, forest mesophication under fire suppression has been suggested to lead to greater sensitivity to frequent droughts under changing climate and reduced C sink across the eastern deciduous forests (Brzostek et al., 2014; Roman et al., 2015). However, this argument does not sufficiently consider changes in water balance patterns along hillslope gradients although mesophytic trees are usually found more at downslope topographic positions (Day et al., 1988).

Working in humid, mountainous, forest catchments, we hypothesize that climate-vegetation-topography-hydrology interactions will manifest in the following ways over time with warming:

To test these hypotheses, we combined several long-term data sets to examine the role of topography, lateral hydrologic flows, and localized water use and growth at forested headwater catchments in the southern Appalachian Mountains, United States.

In this study, we first show that the evapotranspiration has increased, and streamflow yield given precipitation has decreased in the three reference catchments during the growing season using mass balance-based hydrologic metrics (ET, RR and HI; Figure 3). Significance levels were greater for the two normalized hydrologic metrics, RR and HI, than ET, suggesting that these two metrics may better capture vegetation water use patterns with normalizing precipitation variability. These long-term trends in three hydrologic metrics effectively represent increasing localized forest water use and decreasing runoff generation given precipitation over the period. Second, we show that hydrograph declines are becoming steeper and more linear over time through empirical recession slope analyses (Supplementary Figure S4). The late recession behavior is mostly contributed by slow response units of the catchment (mostly upslope portions with longer flow path) (Woods and Sivapalan, 1997; Li and Sivapalan, 2011), and shallow subsurface flow is a main source of sustained base flow in the study site (Hewlett and Hibbert, 1963). Therefore, steeper declines and more linear responses in hydrograph recession suggest that the upslope regions are potentially less hydrologically connected to streams over time (Harman et al., 2009; McGuire and McDonnell, 2010; Hwang et al., 2012). Third, low streamflow dynamics also showed more frequent and prolonged drought periods during the growing season over time, which closely corresponded with root zone soil moisture patterns in the study watersheds. This suggests that despite relatively high annual precipitation (ca. 1,800 mm at low elevations), the study watersheds are moving toward seasonally-drier conditions with greater canopy water use, lower runoff production given precipitation, and potentially lower downslope subsidy.

More frequent and prolonged dry periods could be also partially explained by the increased seasonal and interannual precipitation variability over time. This site has been characterized by an increasing length of inter-storm periods and total rainfall amounts per storm over the period of warming (Laseter et al., 2012; Burt et al., 2018). However, changing precipitation patterns cannot fully explain the increasing ET signals in that more frequent and less intense rainfall usually provides optimal conditions for vegetation water use (both transpiration and interception), and our site is experiencing the opposite. In addition, stormflow dynamics in the study watersheds are characterized by threshold behavior that is a combined function of antecedent soil moisture and storm precipitation (Scaife and Band, 2017); therefore, greater rainfall amounts per storm could lead to higher streamflow generation by subsurface stormflow. Furthermore, pan evaporation has not increased in the study site with warming (Figure 2C), thus atmospheric forcing cannot explain the increasing ET trends in hydrologic records (Roderick and Farquhar, 2002). Therefore, increasing ET signals may be better explained by vegetation responses to changing climate and ongoing forest mesophication (Creed et al., 2014; Caldwell et al., 2016; Hwang et al., 2018; Kim et al., 2018) rather than directly driven by climate forcing variables.

Leaf area patterns have been homogenized along hydrologic flow paths over time (approaching the unity) in all the study catchments, as shown in the long-term remote sensing (Figure 4) and supported by long-term field data in one catchment (Figure 5). Although the ratios of NDVI values decreased <1% over time, there are two main reasons why these signals are not trivial. First, the NDVI metrics from Landsat imagery were from aggregated greenness signals at a 30-m resolution (Hwang et al., 2011a), which may provide inaccurate representations both in topographic and vegetation classifications between upslope and downslope positions (Figure 1). Second, the percent changes in NDVI values should be interpreted as greater changes in LAI values due to the non-linear relationship between NDVI and LAI values in the study site (see Figure 4 in Hwang et al., 2009). Our field-based basal area increment and leaf area index data also showed that the vegetation upslope has grown more than downslope vegetation at least 20% in the similar ratio metrics in the basal area and LAI datasets (Figure 6C), as well as the standard deviations. The homogenization patterns were also more statistically significant at the drier low elevation catchments (WS02, WS14, and WS18; Figure 4) where we also had more significant seasonal drying signals in streamflow dynamics (Figure 3), compared to the wetter high elevation catchments (WS27 and WS36). WS27 might show slightly different patterns both in the NDVI ratios and standard deviations (Figure 4) possibly due to the damage by an ice storm in 2006 at high elevations (>1200 m), reflected in long-term basal area and litterfall data (not shown here).

The xerophytic trees upslope showed greater and more consistent growth over time, while mesophytic trees downslope showed lower growth with larger inter-annual variation (Figure 6). The canopy conductance of the trees downslope also showed greater declines and sensitivity to relative soil moisture deficit than the trees upslope (Figure 7). In other words, mesophytic trees downslope behaved more isohydrically, while xerophytic trees upslope showed typical anisohydric behavior under the moderate drought condition. Although shallow soils remained consistently wetter downslope than upslope, even in dry years (Supplementary Figure S7), vegetation downslope was more responsive to mild drought stress than upslope. This suggests that trees downslope may be experiencing more frequent drought stress and subsequent lower growth due to combined effect of frequent droughts, more water use by vegetation upslope, and potentially lower downslope subsidy over time.

A recent study also showed the strong dependency of vegetation downslope on upslope water subsidy in the study site. Hawthorne and Miniat (2018) showed that despite greater transpiration, there was greater overnight recharge of soil moisture in the downslope plot, driven by downslope flow or hydraulic redistribution. This may indicate the strong dependency of vegetation downslope on upslope water subsidy through lateral hydrologic redistribution, shown to have been potentially decreasing over time in long-term hydrologic records above. This also suggests that emergent decreases in hydrologic subsidy to downslope areas over time might be driven by both changing precipitation patterns, forest mesophication (more mesophytic trees), and subsequent increased ET mostly by up- and midslope tree communities, which occupy major portions of watershed landscapes (Day et al., 1988). Recently, Caldwell et al. (2016) also attributed declining water yield to a shift toward mesophytic dominance in the study site that uses more water than xerophytic oak species (Figure 7).

The divergent growth responses of trees between up- and downslope may be explained by stomatal responses to mild drought stress across the forest landscape (Meinzer et al., 2016). A recent study in the study site also demonstrated that anisohydric oaks in the upslope plot could maintain relatively high transpiration rates in the spring until presumably hydraulic adjustments (embolism) were incurred following the first major dry period (Hawthorne and Miniat, 2018). Meinzer et al. (2013) also showed that water use by maple and poplar trees was twice as sensitive to soil drying compared to oak species, while oaks were relatively insensitive to drying. This may allow trees in the upslope plots to take advantage of warmer springs with earlier greenup (Hwang et al., 2018; Kim et al., 2018; Oishi et al., 2018) and increasing atmospheric CO₂ concentrations. Although CO₂ fertilization can decrease transpiration by improving water use efficiency especially during dry periods (Warren et al., 2011), we did not see any decreasing vegetation water use signals from the long-term hydrologic records (Figure 3). Similar increasing trends of ET were also recently reported at other undisturbed forested watersheds in the southern Appalachians (Hwang et al., 2018). This suggests that the potential CO₂-driven reduction in transpiration might be outweighed by other factors, such as lengthened growing season (Hwang et al., 2011b, 2014) and greater vegetation growth by CO₂ fertilization (Figure 5; Frank et al., 2015; Ward et al., 2018).

However, with increasing ET upslope and frequent droughts, water subsidies to downslope vegetation would decline over time as indicated in the long-term hydrologic records (Figure 3 and Supplementary Figures S4, S5). Mesophytic, isohydric trees downslope appeared to respond more to increased hydroclimate variability due to their dependence on downslope flows and subsidy (Figure 8). A recent tree-ring study in the study site showed that downslope tree species were more sensitive to hydroclimate variability than trees upslope with larger inter-annual variation (Elliott et al., 2015). They demonstrated that radial growth of oaks was greater than maple, birch and tulip poplar on upslope sites in dry years, while the latter species had higher basal increments than oaks on downslope sites in wet years. This pattern has also been confirmed across the eastern deciduous forests using Forest Inventory and Analysis data (Brzostek et al., 2014). This suggests that the isohydric/anisohydric continuum along hillslope gradients will play an important role in forest ecosystem responses to climate change, which would be closely mediated by changes in partitioning between localized water use and lateral hydrologic flows.

The divergent responses of vegetation to soil moisture deficit are not likely attributed to differences in vertical root structures between up- and downslope communities. In this study, we found that trees downslope showed a greater response to mild drought than those upslope, even within the same tree species in a normal year (Figure 7). Recent studies in the study site reported rooting depth and vertical distribution from 27 soil pits across different topographic positions and vegetation types (Hales et al., 2009; Hwang et al., 2015; Hales and Miniat, 2017). They found that roots were distributed deeper and more evenly in wet, hollow (areas of convergent topography) locations, compared to drier, nose (divergent topography) landscape positions. However, maximum rooting depth (around 1 m), total root biomass, and root frequency did not vary systematically between dry/nose and wet/hollow topographic positions and among different tree species. They did report observing distinct tap root structures around the depth of saprolite at several nose pits. Although many studies reported that deep tap roots play an important role in vegetation water use via hydraulic lift during dry periods (Nepstad et al., 1994; Canadell et al., 1996; Caldwell et al., 1998; Siqueira et al., 2008), the study by Hawthorne and Miniat (2018) reported greater overnight recharge in soil moisture at downslope rather than at upslope plots. This suggests that stomatal behavior and seasonally high local water use, rather than differences in rooting distributions, may be responsible for greater upslope growth and LAI over time compared to downslope vegetation.

Our study is in contrast with results from drier and colder ecosystems (e.g., Anning et al., 2013). Bunn et al. (2005) showed that tree ring growth patterns of Pinus balfouriana in the Sierra Nevada Mountains showed stronger correlations with temperature at wet and high convergence areas (downslope), while they correlated more with precipitation at dry and low convergence areas (upslope). Adams et al. (2014) also reported that tree ring growth of P. contorta and P. ponderosa at wet downslope areas showed decoupled responses to regional temperature and precipitation patterns, contrary to trees at dry upslope areas. In the western United States, vegetation water use can be usually decoupled from dominant lateral hydrologic flows during the growing season (Brooks et al., 2010) due to a seasonally dry climate (Tague et al., 2008; Tague, 2009). These abiotic factors would lead to less tight coupling between vegetation dynamics and watershed-scale hydrological behavior especially in dry regions (Adams et al., 2012). In the southern Appalachians, hydrologic subsidy by lateral hydrologic flows often leads to a gradient in plant-available water during dry periods (Yeakley et al., 1998), which may indicate strong dependency of downslope vegetation use on upslope water subsidy. This highlights the need to understand landscape-scale ecosystem responses to changing climate as connected systems between upslope and downslope via associated dominant lateral hydrologic flows.

Convergent or downslope topographic areas are often considered to be potential locations of thermal (climatic) microrefugia where local environment conditions may be decoupled from regional climate conditions (e.g., Dobrowski et al., 2009). While this argument has been mostly driven by topographic effects on temperature regimes, such as adiabatic lapse rates with elevation (e.g., Dobrowski et al., 2009; Gollan et al., 2014) and cold air drainage to valley bottoms (Novick et al., 2016), few studies related the changes in topography-mediated water balance patterns with vegetation responses to climate change (but see Crimmins et al., 2011). With an increase in upslope water use and subsequent decreases in hydrologic downslope subsidy, the effect of increased hydroclimate variability with warming should be amplified in downslope topographic positions, where mesophytic trees have already acclimated to greater soil moisture availability through physiological and rooting strategies. Considering ongoing forest “mesophication” under a history of fire suppression across the eastern United States deciduous forests, this study suggests that mesophytic trees downslope may be more vulnerable in terms of growth reduction and enhanced mortality due to the combined effect of frequent droughts and decreased lateral hydrologic flows under changing climate.

In this study, we combined long-term streamflow, remote sensing, soil moisture, tree basal area, litterfall, and sap flux datasets to explain long-term changes in vegetation patterns at the hillslope to watershed scales. We showed increasing evapotranspiration, decreasing streamflow yield given precipitation, and potential decreasing upslope subsidy to downslope topographic positions over the period of warming. This led to emergent homogenization of canopy density (leaf area) patterns along the hydrologic flow paths, supported by both long-term remote sensing and field observations. Xerophyric trees upslope showed greater growth over time compared to mesophytic trees downslope, closely mediated by the isohydric/anisohydric continuum along the hydrologic flow paths. This study also suggests that the changes in hydrologic partitioning between localized water use (green water) and lateral hydrologic flows (blue water) mediates the divergent responses of vegetation between upslope and downslope topographic positions. We speculate that with forest growth and more frequent droughts, they may become more hydrologically disconnected and increase the ratio of ET to Q, and as a result downslope mesophytic trees become more susceptible to frequent drought due to their strong dependency of green water use on blue water generation upslope. Our findings highlight the need to understand the underlying hydrologic balance along hillslope gradients to predict how forested mountain ecosystems may respond to climate change, possibly reinforced by ongoing forest mesophication under active fire suppression.

The datasets generated for this study can be found in the https://coweeta.uga.edu/dbpublic/dataset_details.asp?accession=1046.

TH, LB, and CM designed the overall research plan, analyzed and interpreted the data, and wrote the manuscript. CM, JV, JK, and PB are in charge of long-term climate, hydrology, soil moisture, and vegetation data, and contributed to writing of the manuscript. CS participated in Landsat TM data analysis and contributed to writing of the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.