Reconstructing groundwater and lake level histories in Northern Wisconsin: isolation of groundwater’s influence on tree rings from climatic and environmental drivers

Tree rings can reveal long-term environmental dynamics and drivers of tree growth. However, individual ecological drivers of tree growth need to be disentangled from the effects of other co-occurring environmental and climatic conditions in tree rings to examine the histories of stand- to landscape-level ecological processes. Here, we integrate ecohydrological theory of groundwater–tree interactions with dendrochronological approaches and develop a new framework to isolate water-level effects on tree rings from climate induced variability in tree ring growth. Our results indicate that changing depth to groundwater within 1–2.3 m of the land surface exerts a substantial influence on red pine growth and this influence can be quantified and used to reconstruct long-term groundwater and lake level histories from tree ring patterns in Northern Wisconsin. This research suggests a substantial influence of groundwater on tree growth with implications for improving the mechanistic understanding of climate-induced tree mortality and reduce uncertainty in forest productivity models. Further, this is a transferable approach to isolate and reconstruct strong environmental drivers of tree growth that co-occur with other environmental signals.


Motivation
Long-term ecological and environmental studiesand the data that support them-are critical for characterizing and quantifying ecological responses to change. Long-term data enable us to understand the structure and functioning of ecosystems (Cody andSmallwood 1996, Clutton-Brock andSheldon 2010), link biological patterns to environmental variability (McGowan 1990, Likens 2012, and inform the management of human influences on ecosystems and the services they provide (Ducklow et al 2009, Nelson et al 2011. Long-term ecological data sets are needed to understand complex ecosystem phenomena that occur over a prolonged period, to reveal unique and/or extreme events, to quantify the effectiveness of conservation management activities like ecosystem restoration, to provide core data for developing, parameterizing, and validating simulation models, and to act as platforms for collaborative studies, thus promoting multidisciplinary research (Lindenmeyer et al 2012). Long-term data additionally provide an enduring resource to support evidence-based policy, decision making and ecosystem management (Lindenmeyer et al 2012). To holistically understand ecological phenomena, it is also necessary to observe system behavior in many locations, which provides insight to predict landscapescale processes, to understand complex community and population dynamics, to categorize regions vulnerable to environmental disturbances, to reveal ecological 'hotspots' and critical ecosystem interfaces, and to determine the scaling of ecosystem processes (Mirtl et al 2018). However, a recent, broad review of the spatial and temporal domains in ecology clearly identifies spatially distributed, long-term data (e.g. field, automated in-situ sensing, remote sensing, and paleo-reconstruction) as one of its most underrepresented, but critically important data needs (Estes et al 2018).
Trees and tree-ring chronologies have been used as readily available long-term and widespread environmental sensors. Tree growth responds to myriad environmental conditions including water, nutrient, and light availability, forest management, insects, fires, competition, and climate among others (Bowman et al 2013). There are approximately three trillion trees around the world (Crowther et al 2015); trees exist in almost every ecosystem type and can record a long, ecologically meaningful period at annual resolution. Dendrochronologists have exploited trees as climate sensors to reconstruct century-to-millennial records of paleoclimate because climate is a primary driver of tree growth (Martinelli 2004). However, to realize the full potential of the multiple signals encapsulated in tree rings, ecological drivers of interest must be isolated from masking, primary drivers, like climate. Isolating the influence of ecologically important tree growth drivers from climate requires tailored sampling designs and associated analyses than those that are currently available and has remained a challenge in dendroecology (i.e. the study of tree rings through time for ecological applications that disentangle relevant ecological signals; Amoroso et al 2017). Thus, new, transferable methodologies that provide a way to achieve a clearer understanding of the mechanisms underpinning tree-ring patterns while controlling for co-occurring environmental variables, like climate-that potentially bias results if unaccounted for-are critically needed.
One specific knowledge gap is how historic groundwater levels influence tree growth and how this interaction could be isolated from climate in tree ring signals. In arid environments, changing groundwater levels strongly influence tree growth (Zhou et al 2019), and recent research has highlighted depth to groundwater (DTG) as an underappreciated driver of tree growth in humid regions (Ciruzzi and Loheide 2021). However, groundwater level histories rarely span more than a few decades in most parts of the world and are notoriously challenging to reconstruct (Jackson et al 2016). Filling groundwater level data gaps would provide the required context to understand historic variability and trends in groundwater levels and the response of groundwater levels to extreme weather events (e.g. prolonged and/or severe droughts/deluges), which are needed to characterize the impacts of recent climate changes for sustainable planning and management of water resources for the natural and built environments (Chen et al 2004, Hanson et al 2006, Holman et al 2011, Jackson et al 2015, Booth et al 2016. Tree growth chronologies often span much longer time scales than most groundwater level records; thus, combining ecohydrology theories related to groundwater-tree interactions and dendrochronological approaches to extend the history of these interactions offers a way forward for studying spatially distributed, long-term groundwater-tree interactions and filling groundwater level data gaps by leveraging trees as hydrological sensors.

Conceptual model and overall approach
A recent ecohydrological conceptual model of groundwater-tree interactions in sandy temperate forests was developed to investigate the influence of DTG on groundwater use and tree growth in sandy temperate forests (figure 1(a), Ciruzzi and Loheide 2021). Sandy soils in temperate forests drain water relatively quickly leading to 'soil droughts' (Li-Ping et al 2006) and more frequent water stress in trees than is often assumed for regions where annual precipitation exceeds potential evapotranspiration. Climatic variability leads to variability in tree growth (Martinelli 2004), but trees that use shallow groundwater are buffered from the impacts of water stress (e.g. reductions in tree growth) and the strength of this interaction is, in part, controlled by DTG. The buffering effect of groundwater on tree growth weakens as DTG deepens and, conversely, strengthens as DTG becomes shallower (Ciruzzi and Loheide 2021). However, DTG also varies with climate characteristics (Watras et al 2014); during dry years DTG becomes deeper and becomes shallower in wet years. While this conceptual model provides the ecohydrological basis to potentially leverage and constrain the history of groundwater's influence in individual trees, disentangling groundwater's effect on tree growth from other drivers of productivity, including climate, has remained unexplored.
A strategic approach is needed to isolate the influence of groundwater in tree-ring patterns from all other drivers of tree growth (figure 1(b); sections 2.1-2.4). First, we sampled trees free of groundwater influence and assumed these trees represent baseline tree growth conditions, primarily driven by climate. We then sampled trees that varied by DTG and assume these trees are influenced by baseline conditions plus DTG. We then form two end member cohorts of tree ring chronologies: a cohort free of groundwater influence representing baseline conditions (the climate signal) and a cohort representing optimal growth under shallow groundwater conditions with the baseline conditions (the optimal GW depth signal). Assuming all other site conditions affecting tree growth rates are equal (section 2.1), variability in tree growth between these two end members represents variability in groundwater influence and therefore variability in groundwater (a) Shallow groundwater enhances tree growth, but the strength of this interaction depends on depth to groundwater. (b) Our approach involves measuring DTG near trees across a depth to groundwater gradient, extracting tree cores and measuring tree ring widths, and applying tree growth transformations (section 2.3). From these data, we define two cohorts of trees ring chronologies representing end member time series: (1) optimal growth conditions under shallow groundwater (blue) and (2) trees free of groundwater influence (red). Position between these two end members (black) represents percent influence of groundwater (PING) on tree growth (section 2.4) and is related to groundwater depth. (c) Reconstructing groundwater levels from tree cores (section 2.5) can fill in temporal data gaps in observed groundwater and lake level records and be applied in the region where shallow groundwater is sufficiently influencing tree growth. levels (section 2.5). DTG in highly conductive sediments in this region varies synchronously with lake levels (Watras et al 2014), thus capturing the historical influence of groundwater on tree growth near lakes can provide insight into lake histories as well. Using this conceptual model and understanding of groundwater-tree interactions in sandy temperate forests, we designed an ecohydrological observatory to explore the following two research objectives: (1) Quantify the percent influence of groundwater (PING) on tree growth along a DTG gradient (1-9 m).
(2) Reconstruct DTG and lake level histories from tree cores.  Lowry et al 2007, Ciruzzi and Loheide 2021. At ten sites in the Trout Lake Watershed in Northern Wisconsin with long-term groundwater level data (⩾15 years of data) covering a DTG range 1-9 m (Magnuson et al 2022) (figure S1), we sampled tree cores from Pinus resinosa Ait. (common name: red pine) trees to identify the two end member cohorts ( figure 1(b)). We sampled P. resinosa trees because this species is distributed throughout the forest, is easy to core, has large, readily-identifiable annual growth rings, and is responsive to climatic and environmental drivers. We collected 1-2 radial tree cores at DBH height from P. resinosa trees within 15 m of groundwater wells with at least 15 years of DTG data. We used a 15 m radius to ensure little to no DTG variation among individual trees near the groundwater well, and topography was generally flat near these wells. Trees that were visibly higher or lower in topography near the well were not sampled (typically 1-2 trees/site). Since the forest composition is mixed deciduous, most sites only had a few P. resinosa trees (3-7 trees/site, 10 total sites, 48 total trees) within 15 m of the groundwater wells at each site. We also targeted trees near the shores of three lakes, within approximately 5-10 m of the lake shoreline, to extend DTG reconstructions to lake level reconstructions. We selected trees growing at the same relative elevations 1-2 m above each lake level for this analysis (3-4 trees/lake, 10 total trees). Two of the lakes were within the study watershed, and one was outside of the watershed approximately 20 miles west of the watershed boundary (figure S1). In our study, we controlled for rooting depth by sampling the same species of tree, P. resinosa with root extinction depth between 0.1 and 3 m (Stiell 1978, Farrar 1995 and for soil textures by only sampling trees in locations with the same sandy soil texture.

DTG data and processing
DTG at each of these wells are manually collected 4-9 times a year and represent DTG for specific days, with higher sampling frequency during the summer months (Magnuson et al 2022).
For each groundwater well, we quantified the average summer (May-September) DTG with co-located tree growth measurements. The resulting annual average DTG time series were used to characterize sites into two cohorts: (1) trees in shallow groundwater environments (DTG < 2.5 m; 5 wells; 22 trees; 38 tree cores); and (2) trees in deep groundwater environments (DTG > 4 m; 5 wells; 26 trees; 50 tree cores) as used in section 2.4. This threshold was determined from previous ecohydrological research that demonstrated P. resinosa growth shows maximum enhancement when DTG is in the range of ∼1-2.5 m, and that tree growth is unaffected by DTG when/where DTG > 4 m (Ciruzzi and Loheide 2021).

Tree growth data and processing
The tree cores in this study were prepared and processed in the same manner as in Ciruzzi and Loheide (2021), which produced tree growth chronologies indicative of environmental variability after controlling for tree size and age. To summarize, we calculated the basal area increment (BAI; i.e. the annual increase in cross-section area of a tree) from measured ring widths, which minimizes the relationship between tree size and growth, but may not account for the entire biological growth trend (e.g. age-growth relationships; Peters et al 2015). Next, we aligned all BAI time series based on age and modeled the mean of all the series with a curve-fitted spline representing the regional growth curve (Briffa and Melvin 2011).
Dividing an individual BAI time series by the regional growth curve yields a ratio-based basal area growth index (BAGI; dimensionless units), which minimizes impacts from both age-growth and size-growth trends and is a tree growth time series indicative of environmental variation. The BAGI time series are then realigned to calendar years. A BAGI value of 1 is the average growth response across all trees of similar age. Values above 1 and below 1 represent above average and below average growth years, respectively. Previous research used this approach and found that mean annual tree growth in shallow groundwater environments (DTG < 2.5 m) was significantly higher than tree growth in areas with deeper groundwater (DTG > 4 m) (Ciruzzi and Loheide 2021). We hypothesize that variability in annual growth rings between shallow and deep end members is indicative of groundwater depth within the range of groundwater depth that separates these end members.

Quantifying the PING on tree growth
For our analysis we separated tree growth time series into two cohorts: (1) the 50th percentile of tree growth from trees in historically deep groundwater environments (DTG > 4 m), representing baseline conditions; and (2) The 95th percentile of tree growth from trees in historically shallow groundwater environments (DTG < 2.5 m), representing maximum tree growth as influenced by groundwater plus baseline conditions. Variation in tree growth between these two cohorts is assumed to be entirely due to variations in DTG. We calculate the PING as: where, BAGI o t is the BAGI observed at a site in a given year (t), BAGI d t is the 50th percentile of BAGI in a given year (t) from trees in historically deep groundwater environments (DTG > 4 m), and BAGI s t is the 95th percentile of BAGI in a given year (t) from trees in historically shallow groundwater environments (DTG < 2.5 m).

Reconstructing groundwater and lake level histories
A linear regression model related all years with paired groundwater depth and PING between 0 and 1. In total, 62 data pairs were used to create a relationship between PING and DTG. This model was used to reconstruct historic DTG time series at sites with trees that are influenced by shallow groundwater (i.e. sites with PING between 0 and 1). The same model was used to reconstruct historic DTG time series for trees near lake shores. We assume DTG near lakes in highly conductive soils is synchronous to lake levels on annual time scales (Watras et al 2014) and present observed and modeled lake levels as water level anomalies, calculated as deviations from the mean when both indicators were available. Three groundwaterfed lakes were chosen relatively near the groundwater sites (two lakes are in the Trout Lake watershed, and one lake is approximately 20 miles west of the watershed) from the NTL-LTER database (Magnuson et al 2022). Lake level sampling measurement frequency varied among the lakes, but generally were manually measured 1-10 times a year as water level elevation for the length of observations (n = 22−32 years). Therefore, the resulting summer lake level averages do not represent a true annual mean, but an average of the observations for which there are available data. For each lake, the average DTG determined from its tree cohort was used to create an offset elevation relative to the lake level before converting to modeled lake level anomaly.
Since groundwater level data were used to create the PING-DTG relationship, the groundwater reconstruction can be viewed as model calibration; however, the lake level data are independent with reconstruction serving as model validation. Individual years and sites when and where PING > 1 and PING < 0 represent years and sites that are outside the sensitivity range of this model and are removed from the reconstructed DTG time series. The standard deviations (SDs) of the water level observations and modeled observations, root mean squared error (RMSE), and Nash-Sutcliffe model efficiency coefficients (NSE; Nash and Sutcliffe 1970) were calculated at sites with reconstructed groundwater level histories. NSE, a metric commonly used to evaluate the performance of hydrological models is calculated as: where DTG o is the mean of observed DTG, DTG m is modeled DTG, and DTG t o is observed DTG at time t. Both RMSE and NSE are indicative of model performance. NSE is additionally useful to discern whether the model performs better, similar, or worse than using the observed mean as the model itself. NSE values range from −∞ to 1. An NSE = 1 indicates a model that perfectly matches observed data, an NSE = 0 indicates the model performs just as well as assuming a model of the observation mean, and NSE < 0 indicates the model performs worse than using the mean.

The PING on tree growth
Tree growth chronologies in shallow and deep groundwater environments overlapped from 1930 through 2015 (figure 2(a)), providing an opportunity to quantify the PING on tree growth for 85 years. Tree growth under optimal shallow groundwater conditions remained higher than tree growth in deep groundwater conditions for all overlapping years ( figure 2(a)). At six sites, historic DTG means were known to be deeper than 2.5 m and these sites showed near-0 or negative PING values throughout their lifetimes suggesting no groundwater influence because the groundwater is too deep. Three sites with historical DTG means shallower than 2.5 m showed positive PING values between 0% and 100%. We observed groundwater influence for P. resinosa tree growth within a relatively narrow range of DTG between 1.19 and 2.25 m.

Reconstructed groundwater and lake level histories
Groundwater and lake level variations were reconstructed from tree ring patterns (figures 3(a) and (b)). While the accuracy (RMSE) of the model performs well for these groundwater sites and lakes, the aggregated model performance (NSE) appears to depend on sufficient variation in water level and groundwater influence. Generally, NSE values below 0.5 are considered poor, 0.5-0.65 are satisfactory, 0.65-0.75 are good, and >0.75 are very good (Moriasi et al 2007). While RMSE were similar among all reconstructions (RMSE range 0.15-0.25 m), the SD and NSE differed. NSE is sensitive to the range of variability captured between the model and observations. Since wells like K6 and K75 have more variation that is captured by the model than well K56, the NSE for K6 (NSE = 0.78) and K75 (NSE = 0.53) is much higher than K56 (NSE = −0.22), even though the latter has the smallest RMSE.
Conditions when and where PING > 100% or PING < 0% are indicative of years with near-optimal (DTG = ∼1.3 m, green shading in figure 3) or deeper (DTG > 2.3 m, purple shading in figure 3) groundwater levels, respectively. While these time periods are not included in the measures of fit described above because the method does not provide a single water level estimate, the method does provide valuable, categorical information on groundwater or lake level by indicating it is in a shallow or deep phase relative to the Relating the percent influence of groundwater on tree growth and depth to groundwater. (a) Quantifying the percent influence of depth to groundwater on tree growth. Tree growth cohorts are separated by depth to groundwater classes. The blue line with blue dots represents a time series of optimal tree growth under shallow groundwater conditions and tree growth that is 100% influenced by groundwater. The red line with red dots represents a time series of tree growth unaffected by groundwater and tree growth that is 0% influenced by groundwater. The percent influence of groundwater on tree growth (PING) is calculated as the percent between these two end-member time series. For example, an individual site is shown with the black line with gray dots. In 2015, the PING for this site is 70%. The percentages in each year are related to the average summer depth to groundwater in the same year as show in (b), where the data coordinate highlighted in (a). The PING-DTG relationship was created from three sites that showed groundwater influence (n = 62 data pairs of PING-DTG data). sensitivity range. The accurate categorical prediction during these extremes is not reflected in the goodness of fit metrics, including RMSE and NSE, but is a potentially useful prediction on its own.

Implications for filling data gaps in groundwater level observations using tree cores
The methodology used here to reconstruct groundwater levels using tree cores adds to the relatively few studies that reconstruct groundwater level histories, including studies using statistical and process-independent (neural networks) methods (e.g. Conrads and Roehl 2007), process-based models (e.g. Najib et al 2008), environmental proxies (e.g. Ferguson and St. George 2003), and other tree core studies (e.g. Perez-Valdivia and Sauchyn 2011, Zhou et al 2019). This study is unique among tree core studies because we minimized co-occurring climate conditions and quantified the PING on tree growth as the dependent variable before reconstructing groundwater levels from tree ring indices (Perez-Valdivia and Sauchyn 2011, Zhou et al 2019).
Inundated wetlands, rivers, and lakes fed by persistent groundwater discharge (water table rising above land surface, DTG ⩽ 0) cover ∼15% of the global land surface, ∼2% additional land is covered by less frequently inundated wetlands (0 < DTG ⩽ 0.25 m), and ∼5% to ∼15% of global land includes DTG or its capillary fringe within the rooting depth of upland plants (within 0-3 m of the land surface), adding to ∼22% to ∼32% of global land area influenced by shallow groundwater (Fan et al 2013). This represents a substantial portion of global land area influenced by groundwater and the widespread potential to use the approach developed here to leverage trees as hydrological sensors to fill in long-term groundwater level data gaps where groundwater-tree interactions are strong. While we have focused on the temporal aspect of reconstructing water level time-series here, an important application of the method is in mapping the spatial extent of both current and historical groundwater levels and their influence on ecosystems. The ubiquity of trees across the landscape provides an unprecedented ability to tap into spatially distributed and historic, longterm groundwater records that quantify groundwater depth in the depth range of growth sensitivity or classify a location as having groundwater levels too shallow or too deep to influence growth.

Environmental processes influencing the strength and predictive capacity of groundwater-tree growth interactions
Groundwater is an under-appreciated but substantial driver of tree growth and if widespread at local, regional, and global scales a better understanding of groundwater-tree growth interactions is necessary to reduce uncertainty in forest productivity estimates related to climate. Increases in the frequency, duration, and/or severity of drought are predicted to induce widespread tree mortality (Allen et al 2010) and lower groundwater levels (Peters 2003), which in areas where groundwater-tree growth interactions are strong could result in amplified reductions in productivity. Mechanistic understanding of climateinduced tree mortality requires improved knowledge of belowground processes, including drivers of soil moisture conditions and water availability to tree roots (e.g. Brunner et al 2009, Allen et al 2010. Local, regional, and global forest productivity and mortality models often include detailed algorithms describing aboveground physiological processes but treat belowground processes as a 'black box' and rarely are groundwater processes included (Phillips et al 2016). Thus, better understanding of groundwater-tree growth interactions may help constrain uncertainty in climate-induced tree mortality and improve the accuracy of forest productivity models. Rooting depth and soil texture are important drivers of groundwater-tree interactions in temperate, humid forests that that could confound results and interpretations, but if properly accounted for may increase the predictive capacity of groundwater-tree growth interactions, including reconstructing water levels with tree rings. We did not observe groundwater influence in P. resinosa trees deeper than ∼2.3 m, which is consistent with the maximum rooting depth of this species between 0.1-3 m (Stiell 1978, Farrar 1995, and suggests that rooting depth is a control on PING-DTG relationships. We recommend future studies explore multiple tree species with different rooting depths and profiles to increase the predictive range of using tree cores to reconstruct DTG histories. For example, maximum rooting depths for common temperate forests (including mixed species) is 4.0 ± 0.4 m (Canadell et al 1996, Stone andKalisz 1991). Following a similar approach in this study, investigating groundwater-tree growth interactions in temperate trees with deeper roots can potentially increase the range of reconstructed groundwater levels as observed in this study and fill in data gaps outside the sensitivity of P. resinosa trees, (e.g. filling in the deep groundwater level data gaps from 2008 to 2012 at site K75; purple regions in figure 3). Further, we did not observe evidence of groundwater flooding or oxygen stress that would reduce tree growth (Sojka 1992). We suggest that the recent observations of substantially shallow groundwater levels and record high water levels in much of Northern Wisconsin (Watras et al 2022) could be used to investigate the influence of extremely shallow groundwater that may reduce tree growth and would enable a greater range of sensitivity for reconstructing groundwater and lake levels.
Soil texture influences the strength, sensitivity, and optimal DTG for groundwater-vegetation interactions under variable climate (Zipper et al 2015) and it is expected to influence groundwater-tree growth relationships in temperate forests. When water derived from precipitation events is depleted, shallow groundwater subsidizes transpiration and tree growth in forests with sandy soils leading to enhanced growth compared to trees without access to groundwater (Ciruzzi and Loheide 2021). As groundwater deepens, the ability of root zone soils to retain water that falls as precipitation plays a critical role in determining transpiration and productivity, and therefore plants in finer soil textures (i.e. more water retentive) tend to be less dependent on groundwater than those in coarser-grained soils (Zipper et al 2015), which likely leads to weaker and less frequent groundwater-vegetation interactions. The DTG range of groundwater influence may also extend to a deeper depth in finer textured soils than in coarser soils. In finer textured soils the upward movement of water from the water table due to capillary action (i.e. the capillary fringe) is higher and extends up to a few meters above the water table . The capillary fringe in very coarse soils is small to negligible . Therefore, assuming rooting depths are similar in coarse-and fine-grained soils (Foxx et al 1984), groundwater-tree interactions in finer-grained soils may be influenced by a deeper DTG on the order of meters thereby increasing the potential range of reconstructed DTG. Future work should explore groundwater-tree growth interactions and groundwater level reconstruction using tree cores for species with different rooting depths and for different soil textures to determine how these considerations may expand the range of groundwater-tree growth sensitivity for reconstructing historic groundwater and lake level histories.

Isolating ecological signals in tree ring patterns that co-occur with environmental and climatic conditions
Though the new methodology developed here was applied to investigate groundwater-tree interactions, conceptually this framework can be extended and applied to investigate any singular mechanism influencing tree growth and reconstruct its history. Transferring this methodology to isolate other ecological drivers of tree growth from co-occurring environmental and climatic signals will require similar sampling strategies, including controlling for other site characteristics except for the signal of interest. Importantly, the ecological signal of interest will need to exert a strong control on tree growth at the annual scale and will need to vary over either the spatial scale of interest or the lifetime of the trees. We suggest this methodology is powerfully useful for revealing the historic influence of ecological conditions that have outsized effects on tree growth compared to baseline conditions, whether that is enhanced growth, as observed in this study, and/or growth limitations. Though not an exhaustive list, investigating other tree growth drivers related to nutrient variability (Lévesque et al 2016), insect infestation (Anderegg et al 2015), fires (Guiterman et al 2015), and disease (Das et al 2016) may be potentially strong signals that can be extracted using this approach.

Conclusions
This study demonstrates the feasibility and potential of dendroecology to isolate and reconstruct specific ecological signals that may be masked from other co-occurring signals, including climate. Assuming all other site conditions were equal, we sampled and processed two end member tree growth time series, one end member representing the deepgroundwater (baseline), climate-mediated signal, and the other end member representing optimal growth under shallow groundwater conditions. Varying tree growth between these two cohorts represents varying groundwater influence that, in tandem with modern observations of groundwater levels, were leveraged to reconstruct groundwater and lake level histories. Thus, we suggest that trees can be used as hydrologic sensors to fill in historic groundwater and lake level data gaps in forests where and when groundwatertree interactions are strong. This research also highlights the substantial influence of groundwater on tree growth that if widespread at local, regional, and global scales could help improve mechanistic understanding of climate-induced tree mortality and reduce uncertainty in forest productivity models. Overall, this innovative dendroecological methodology can be extended to investigate other singular drivers of tree growth, quantify their influence, reconstruct their histories, and extract widespread, long-term information to inform stand-to landscape-level ecology.

Data availability statement
The data that support the findings of this study are openly available at the following URL/DOI: www.hydroshare.org/resource/ee4eec731ee5414d8e dab62fe6105f66.