The Peculiar Hydrology of West-Central Florida’s Sandhill Wetlands, Ponds, and Lakes—Part 1: Physical and Chemical Evidence of Connectivity to a Regional Water-Supply Aquifer

The sandhill wetlands, ponds, and lakes of west-central Florida, USA, are an understudied, poorly understood variant of geographically isolated features. Their karst origin and xeric setting impart a characteristic ecohydrology, which has been attributed to their apparent connectivity to a regional water-supply aquifer. This study uses physical and geochemical hydrologic data to provide evidence of this connectivity. The findings presented here debut these rare features, which advances their fundamental understanding at a time when increasing anthropogenic pressures risk further loss and degradation. From these findings, the hydrologic nature of sandhill wetlands, ponds and lakes is characterized, which may be useful in distinguishing them from others with different hydrologic controls and for identifying features of similar connectivity, karst or otherwise, wherever they may be found. Water levels and/or geochemistry were compared for 12 wetlands, five ponds, two lakes, and 12 monitor wells (10 constructed in limestone, two in surficial sand) in west-central Florida. Hydrograph and regression analyses indicate similar widely ranging water levels (2–5 m) for sandhill features and wells that are both analogous in elevation and highly correlated with each other (0.84 ≤ R2 ≤ 0.99). Sandhill feature geochemistry also reflects that of the monitor wells, varying relative to the depth of the rainwater-limestone water mixing zone. Findings here suggest sandhill wetland/water features are surface water expressions of the underlying unconfined regional aquifer hydrology, distinguishing them from isolated features elsewhere and establishing them as a groundwater endmember along the hydrologic continuum.


Background and Introduction
Geographically isolated wetlands (GIWs) and waters (i.e. ponds as lakes) are defined as aquatic islands in a terrestrial landscape (Edwards and Sharitz 2000), those surrounded by upland (Tiner 2003), and other definitions depending upon scale and perspective (e.g. ecological or hydrologic isolation, etc.) (Liebowitz and Nadeau 2003). While not navigable or of commerce value (and thus not federally protected), they contribute valuable functions on the landscape including: flood storage, water table regulation, nutrient/sediment retention, and wildlife and aquatic habitat (Novitzki et al. 1996). Recent studies suggest they do this in aggregate as a complex or portfolio with other nearby wetlands and waters, collectively offering landscape support and benefitting downstream waters along a spatial and temporal continuum (Uden et al. 2014;Cohen et al. 2016;Nowicki 2019). Tiner (2003) described GIWs as being among "America's most valuable and threatened natural resources," having suffered numerous losses and degradation from habitat destruction, altered hydrology, and water pollution, of which they continue to be at risk.
Understanding the hydrology of GIWs/waters is fundamental to their protection. For most, local factors-meteorological, geologic or both-control their hydrology. This is true for Carolina Bays along the Atlantic Coastal Plain (Lide 1 3 et al. 1995); moraine, ice-scour, and kettle ponds in Alaska (Rains 2011); playa wetlands across the Southern High Plains (Tsai et al. 2007); prairie potholes in the Upper Midwest (Hayashi et al. 2016); vernal pools in the Northeast and California (Brooks 2004;Rains et al. 2006Rains et al. , 2008O'Driscoll & Parizek 2008); and others. It also is true of many sandhill wetlands and waters in Nebraska (Ginsberg 1985) and Florida (Jones Edmunds 2006;CH2M Hill 2005). But for a unique variant of GIWs/waters in west-central Florida, apparent hydrologic control is regional, via connectivity to the Upper Floridan aquifer (hereafter U Fldn)-part of the expansive Floridan aquifer system and one of the most productive aquifers in the world (Miller 1990).
The hydrology of sandhill wetlands and waters of westcentral Florida is not well studied. Short of work by Henderson (1986), which documents the close relationship between Hunter's Lake (one of this study's sites) and the regional groundwater system, little is documented of the hydrology of scores of other sandhill wetlands and waters dotting the landscape. This makes them particularly vulnerable to increasing anthropogenic pressure as demands for water and land increase, and a changing climate promises added uncertainty.
This study is one of a pair of studies intended to improve understanding of Florida sandhill wetland/water ecohydrology. Here, physical and geochemistry data from wetland/ water features are compared with that of the U Fldn to provide evidence of their hydraulic connectivity. We posit that if the features are connected to the U Fldn, then their water level elevations, fluctuations, and geochemistry would be the same, reflecting its control over their hydrology; inversely, features not connected to the U Fldn would have different characteristics, reflecting that of a local control(s) (e.g. a perched water table). A hydrologic characterization of sandhill wetland/water features derived from these comparisons also is presented and may help distinguish sandhill features from those that may appear similar, but have different hydrologic controls. In the companion study, geophysical applications and lithologic data are used to identify the specific mechanisms by which sandhill wetland/water features connect with the U Fldn. From this, a conceptual model is presented along with a model of fundamental sandhill wetland/ water ecohydrology (Nowicki et al. in review).
The hydraulic connection of sandhill wetlands, ponds, and lakes to a regional aquifer, as confirmed here, distinguishes them from other GIWs/waters and establishes them as an endmember along the hydrologic source continuum. This finding and the resulting characterization of their hydrologic behavior have important local and regional implications for wetland and groundwater management and regulation. They also have broader scale importance to those studying karst wetland/water features or others under similar hydrologic control, wherever they may be found.

Study Area
The study area includes portions of two west-central Florida counties, Hernando and Pasco, and two physiographic provinces, the Gulf Coastal Lowlands and Brooksville Ridge (White 1970) (Fig. 1). Land surface elevations generally range from 1 to 30 m above sea level (NAVD 1988 datum) across the Gulf Coastal Lowlands (with small relict dune features as high as 37 m) and from 12 to 80 m (NAVD 1988 datum) across the Brooksville Ridge.

Climate
The climate in west-central Florida is humid subtropical, with a 30-year (1980-2010) normal annual rainfall of 1341 mm (Brooksville Hernando Co Airport, Florida, USW00012818, 1981-2010 (Arguez et al. 2010). Most of the rain (57%) falls in the wet season (June-September) as convective storms; the rest falls in the dry season (October-May) as less intense frontal systems. Annual rainfall extremes ranging from 860 mm (Richloam Tower gage WY 1980, SWFWMD 2018 to 2120 mm (Chassahowitzka gage WY 2003, SWFWMD 2018Richloam Tower gage WY 2003, SWFWMD 2018 have been recorded at local gages, generally in association with drought and La Nina events or with tropical storm and El Nino events, respectively. Strong local rainfall variation exists whereby annual differences of 300 mm or more have been reported for gages within 1 km of each other (Hernando County Utilities Department [HCUD], unpublished data). Annual evapotranspiration averages 1000 mm for the region (Bidlake et al. 1993), and annual average lake evaporation can exceed the long-term annual average rainfall (Sacks et al. 1994;Lee et al. 1997;Swancar et al. 2000).

Hydrogeologic Setting
The hydrogeology of the study area varies by physiographic province. In the Gulf Coastal Lowlands, the U Fldn occurs as a sequence of near-surface, highly transmissive karstic limestone that occasionally outcrops. It is overlain by a relatively thin overburden of unconsolidated sand with minor amounts of silt and clay and is considered regionally unconfined (Arthur et al. 2008;Basso 2004). The absence of lowpermeability sediments allows groundwater to move freely between the sand and limestone. The high permeability of the overburden allows little surface runoff, and recharge is relatively high.
In the Brooksville Ridge, low permeability clay sediments discontinuously lie between the limestone and mostly sand overburden. These clay sediments are remnants of the Hawthorn Group and are thickest along the center of the province where land elevation is highest (Arthur et al. 2008). Although not expansive or effective enough to confine the U Fldn (i.e. a surficial aquifer does not occur in the study area), the clay sediments can produce locally perched water table conditions above it (Basso 2004). The clay sediments thin away from the center (westward and eastward), becoming discontinuous or altogether absent.
The regional (U Fldn) hydraulic gradient (for the study area as a whole) is relatively gradual at approximately 0.3 m/ km (May 2015 and varying over time, FDEP 2018) (Fig. 1). The direction of the gradient is generally northwest towards the Gulf of Mexico.

Methods
In this study, physical and geochemical data from 19 wetland/water features are compared with data from 12 U Fldn monitor wells to provide evidence of their hydraulic connectivity. Features whose water level elevations and fluctuations are analogous to those of the U Fldn and whose geochemistry is overlapping would suggest hydraulic connectivity (given the unconfined nature of the U Fldn throughout the study area), while physical and geochemical data to the contrary would suggest hydrologic control by a perched water table. Details of the 31 study sites, the type of data collected at each, and the methods of collection and analyses follow

Study Sites
Surface water elevation and geochemistry data were evaluated from 12 wetlands, five ponds, and two lakes, each ranging in area from 1 to 178 ha and in depth from 1 to  Table 1). (Ponds here are distinguished from wetlands in their permanent inundation and from lakes in their lack of wave action, Nowicki et al. in review). Shallow groundwater elevation and/or geochemistry data were evaluated from monitor wells constructed within the surficial sands of the interior of 10 of the 12 wetlands. Wells were constructed of 5.1 cm inside-diameter PVC, screened over the lower interval, and 1.0-3.0 m in total depth.
Deeper groundwater elevation and/or geochemistry data were evaluated from 12 U Fldn monitor wells located in uplands-10 constructed within limestone and two constructed within the overlying surficial sands. Two of the wells constructed in limestone are located adjacent to surface water features; the other wells are located 1-10 km from the nearest feature. Upland wells in limestone were constructed of 2.4-3.2 cm inside-diameter stainless steel or PVC, with solid casing set to 20-78 m below ground surface (just below the top of the limestone), and with open boreholes in limestone to 37-162 m total depth. Upland wells in surficial sands were constructed of 2.5-5.1 cm insidediameter PVC, screened over the lower interval, and 24 m in total depth. Five of the 12 monitor wells were evaluated only for their water level elevations, five only for their water geochemistry, and two for their water level elevations and geochemistry ( Fig. 1, Table 1).
All of the surface water features and seven of the 12 upland monitor wells were selected from a regulatory environmental monitoring program associated with local groundwater production (Water Use Permit #20005789.009 SWFWMD 2015) and are not evenly distributed among physiographic provinces (Fig. 1). The other five wells (all constructed in limestone) were included in this evaluation to compare their water levels with those of the wetlands, ponds, and lakes. Unlike the other monitor wells, these wells are not adjacent to groundwater production wells and have lengthier, more consistent water level elevation records.

Water Level Data Collection
Water level data obtained for most of the 19 wetlands, ponds, and lakes include two period-of-records (PORs) beginning  (HCUD unpublished data). Data generally include twice monthly staff gage readings when surface water was present, or shallow monitor well readings when surface water was absent. Where both staff gage and shallow monitor well readings were recorded, water levels from the monitor wells were selected for analysis (largely to rule out temporary ponding from recent rainfall during low water periods).
The PORs for groundwater level data for the five U Fldn monitor wells begin between 1967and end in 2018SWFWMD 2018;USGS 2018). Data consist of hourly recordings from pressure transducers aggregated into daily averages. All data were provided in NAVD88 units of feet and converted to metric for this study.
Rainfall data (daily and 1980-2010 monthly normals) were obtained for the Brooksville Hernando County Airport gage located in the center of the study area and are included in site-specific hydrographs ( Fig. 1) (Arguez et al. 2010).

Water Geochemistry Data Collection
Water geochemistry data were collected once each at the end of the 2015 dry and wet seasons (May/June and October, respectively). Parameters sampled include those common to hydrologic sourcing and surface water/groundwater mixing studies: field pH, specific conductance, and temperature; major ions (Na + , K + , Mg 2+ , Ca 2+ , Cl − , and SO 4 2− ); and stable isotopes ( 2 H and 18 O) (Winter et al 1998;Lee et al 2009;Rains 2011;Barbieri 2019).
Sampling at each site was performed along a vertical gradient to examine whether geochemistry varied with depth ( Fig. 2). At the wetlands, ponds, and lakes, one or two surface water samples were collected each season-where water depth was 2 m or less (all but one wetland), one shallow sample was collected at approximately 0.5 m below the top of the water column using the grab method. At the ponds and lakes (and one wetland) where water depth was greater than 2 m, samples were collected at two depths-a shallow sample was collected as described above, and a deep sample was collected using a horizontal Van Dorn sampler at approximately 4.5 m or 0.5 m above the bottom elevation, whichever was less. Grab samples were collected in a 500 mL high-density plastic (hdp) container with a screw on cap; Van Dorn samples were transferred to hdp containers upon retrieval.
At the monitor wells, one to three groundwater samples were collected based on the total depth of the well (Fig. 2). At wells in surficial sands (screened only along the bottom intervals), one sample was collected approximately 0.3 m above the bottom of the well. At wells in limestone (open boreholes throughout all but the uppermost limestone), up to three samples were collected at approximately 23 m, 46 m, and 91 or 137 m-referenced herein as shallow, deep or very deep samples (respectively)-to identify any variation along the vertical gradient. Samples were collected using a peristaltic pump or bailer. Pumped samples were collected directly into 500 mL HDPE containers; bailed samples were transferred to 500 mL hdp containers upon retrieval.
Field parameters-pH, temperature, and specific conductance-for both surface water and groundwater samples were measured using a YSI 556 MPS or the equivalent (YSI Inc., Yellow Springs, Ohio, USA). Samples were then sealed and transferred to the mobile staging area. Samples to be analyzed for dissolved ions were filtered with a 0.45 micron filter and then transferred to separate 30-50 mL hdp containers and maintained at ± 4 °C until analysis. Samples analyzed for cations were treated with nitric acid within 1 week of collection. Samples to be analyzed for stable isotopes were transferred to 30-50 mL hdp containers, completely filled and capped with both laboratory film and airtight caps to ensure isolation from the atmosphere. All sampling equipment was thoroughly rinsed with deionized water between samples.
Cation (Na + , K + , Mg 2+ , Ca 2+ ) and anion (Cl − , SO 4 2− ) composition were determined at the University of South Florida's (USF) Center for Geochemical Analysis or at Advanced Environmental Laboratories, Inc. using a Perkin Elmer Optima 2000 DV ICP-OES or the equivalent. Alkalinity as bicarbonate (HCO 3 − ) was then calculated as the missing ion using the modeling function in AquaChem (Waterloo Hydrogeologic, Inc. 2020). Stable isotope composition ( 2 H, 18 O) relative to the Vienna Standard Mean Ocean Water (VSMOW) was determined at USF's Stable Isotope Laboratory using a Picarro Cavity Ringdown Spectrometer, model L2130-i.

Water Level and Geochemistry Data Analyses
Hydrographs and linear regressions were used to examine the similarity between water levels of wetland/water features and those of the U Fldn monitor wells (Addinsoft 2019). Water levels of features hydraulically connected to the U Fldn would be analogous to those of the U Fldn monitor wells, in both elevation and fluctuation, while water levels in features with no hydraulic connection would not. For example, where a perched water table exists (i.e. one separated from the U Fldn water table by a vadose zone), feature water level elevations would be higher than those of the underlying U Fldn, and fluctuations would be more sensitive, responding more sharply to rainfall and lack thereof than the water table of a large regional aquifer. Disparities such as these would be evident in the hydrographs and poorly correlated in the regression.
Similarity between wetland/water feature geochemistry and that of the U Fldn monitor wells also was examined under the premise that the geochemistry of features hydraulically connected to the U Fldn would overlap that of the U Fldn monitor wells. Overlap of this kind would not be expected for features with no hydraulic connection to the U Fldn. For example, where a perched water table exists, the geochemistry of the wetland/water features would be expected to reflect that of rainfall (e.g. acidic pH, relatively low ion concentrations, and potentially enriched in heavy isotopes), while the geochemistry of features connected to the U Fldn would likely reflect rainwater mixed with water residing in limestone (e.g. circumneutral or basic pH, relatively high ion concentration(s), and potentially depleted in heavy isotopes).
To identify potential overlap between wetland/water and U Fldn geochemistry, various qualitative and statistical comparisons were undertaken. Field pH relative to specific conductance and Ca 2+ were examined using scatterplots (Addinsoft 2019). Major ion composition was compared using Piper diagrams (Waterloo 2020). Heavy isotope enrichment and depletion were evaluated using global and local meteoric water line (MWL) plots. These plots relate isotopic data relative to a global rainfall standard according to the equation δ 2 H = 8 δ 18 O + 10 (Craig 1961); 2 H and 18 O values that plot along this line reflect global precipitation that has not evaporated (e.g. groundwater), while values that plot below the line generally reflect water enriched in heavy isotopes left behind after lighter isotopes evaporate (e.g. surface water).
Similarity in geochemistry was further examined using Agglomerative Hierarchical Clustering (AHC) (Addinsoft 2019). AHC iteratively pairs individual samples into clusters based on their least dissimilarity; it pairs clusters until the desired (or automatically generated) number is reached (Reddy 2018). For this study: sample similarity was based on (collective) specific conductance, Ca 2+ , Mg 2+ , K + , Na + , Cl − , SO 2− 4 , 2 H, and 18 O and measured as the Euclidian distance between objects; Ward's method was selected as the agglomeration method (Ward 1963); and the final number of clusters was generated automatically. Significant differences in individual parameters also were examined by water type (e.g. surface water in wetlands, surface water in ponds/lakes, groundwater from wells in surficial sand, and groundwater from wells in limestone) using the Kruskall-Wallace and Steel-Dwass-Critchlow-Fligner multiple pairwise comparison tests (Addinsoft 2019).

Water Level Analyses
Hydrographs comparing water levels of sandhill wetland/ water features and those of nearby U Fldn monitor wells reveal marked similarity. Elevations are analogous, and fluctuations are synchronous over both the short and long term (Brooksville Ridge features excluded) (Figs. 3a, b). Deviations from the U Fldn occur as two basic types-an elevation offset and behavioral responses. An elevation offset is the median deviation between the feature and well water levels (Fig. 4a). It varies little over a feature's POR and may range from negligible (e.g. Croom Road Marsh, Fig. 3a) to more than 3 m (e.g. Norman Marsh, Fig. 3b), depending mostly on relative feature-well positions along the regional hydraulic gradient (Fig. 1). Elevation offsets are generally small for features and wells in close proximity and during periods of a Fig. 3 a Hydrographs of wetland/water features and U Fldn monitor wells of the Gulf Coastal Lowlands physiographic province (northern, central and southern parts of the study area). b Hydrographs of wetland/water features and U Fldn monitor wells of the Brooks-ville Ridge physiographic province (eastern flank and atop the Ridge feature itself). For clarity, only the most recent 10 of the 6-52-year PORs are shown high or low recharge (i.e. when the gradient flattens between sites). In contrast, behavioral responses are numerous and vary in their magnitude, rate, and timing in response to rainfall or lack thereof (Fig. 4b). When an elevation offset is adjusted (by vertically shifting the axis of the monitor well so its water levels vary at the elevation of the wetland water level), the behavioral responses are more apparent. In general, behavioral responses at the features (relative to the U Fldn wells) show a lower magnitude, but similar rate of incline and a similar onset, but slower rate of decline (except during hydrologic highs, when onset lags, but rate is similar). The less the deviation between feature-well behavioral responses, the greater is their synchronicity.
Synchronicity between feature and U Fldn water levels was statistically examined using linear regression analyses. At all but the Brooksville Ridge features, correlation coefficients are high (R 2 = 0.84-0.99) ( Table 2), suggesting most to all of the variation in the wetland, pond, and lake water levels are explained by variation in the U Fldn water levels. The highest correlations are generally associated with features: distributed across both physiographic provinces; proximal to the monitor wells (generally within 3 km); smaller in area (generally 5 ha or less); and represented by all three types (wetland, pond, and lake), although ponds as a group are more highly correlated (0.94 ≤ R 2 ≤ 0.99) than wetlands or lakes (0.84 ≤ R 2 ≤ 0.98) ( Table 2) suggesting some factor other than the U Fldn (e.g. a perched water table) explains most of their water level variation. To better understand the residual variation among wetland/water features, deviations were examined more closely for three exemplar features (i.e. those expressing characteristically high correlation and low deviation) and three exceptional features (i.e. those with low correlation and high or unusual deviation). At two of the exemplar features, water levels exhibit near-perfect correlation (R 2 = 0.99), despite their stark hydrogeomorphic (and physiographic) differences. Croom Road Marsh is a very shallow, intermittently inundated wetland located along the eastern flank of the Brooksville Ridge, while Chapel Pond is a deep, permanently inundated pond located in the Gulf Coastal Lowlands ( Fig. 1; Table 1). Water levels at Croom Road Marsh are nearly coincident with those of the nearby U Fldn well, owing to a negligible elevation offset and minimal differences in behavioral response (Fig. 5a). Water levels at Chapel Pond deviate some from those of the nearby U Fldn well, both in elevation offset (which is small, 0.1 m, compared to most other sites) and in behavioral responses (which also are small and a b Fig. 4 a Elevation offset example between a sandhill wetland (Capuchin Pond) and a nearby monitor well (ROMP 97 U Fldn). Note the offset is fairly consistent over time. b Behavioral response deviations example between a sandhill wetland (Capuchin Pond) and a nearby monitor well (ROMP 97 U Fldn). The elevation offset has been adjusted (note different vertical axes) to highlight these deviations. Note that while numerous deviations may occur for a given site, the general patterns are consistent across the POR 1 3 similar to those described previously) (Fig. 5b). At the third exemplar feature, Ref 4-a moderately deep, seasonally inundated wetland in the Gulf Coastal Lowlands ( Fig. 1; Table 1)-water levels are highly correlated with the nearby U Fldn well (R 2 = 0.92), but with a greater elevation offset and greater behavioral response deviations (Fig. 5c). The deviations are particularly noteworthy when wetland water levels shift between surface and shallow groundwater phases.
In the three exceptional features, water level deviations from those of the nearby U Fldn monitor wells are uncharacteristic relative to those of the exemplar features. At Weeki Wachee Prairie, a large deep multi-pool wetland in the Gulf Coastal Lowlands ( Fig. 1; Table 1), surface water and shallow groundwater levels are highly correlated to the U Fldn well (R 2 = 0.88), and behavioral responses are consistent with those of the exemplar wetlands (Fig. 5d), but the elevation offset is unexpectedly high (up to 0.5 m or more), given the U Fldn well is in close proximity (at the wetland shoreline). At Banshee Pond, a deep semipermanently inundated wetland along the eastern flank of the Brooksville Ridge (Fig. 1, Table 1), surface water levels are dichotomous relative to U Fldn well groundwater levels. A higher elevation offset (2.0 m) and low correlation (R 2 = 0.24) occur during low water periods, and a lower elevation offset (1.2 m) and high correlation (R 2 = 0.92) occur during high water periods (Fig. 5e). At the third exceptional feature, Sand Point Pond-a shallow, seasonally inundated wetland located along the center of the Brooksville Ridge province ( Fig. 1; Table 1)-the elevation offset is very high (12.5 m); and behavioral Table 2 Results of linear regression of paired water level data-wetland/water feature as dependent variable and U Fldn as explanatory variable Fig. 5 a-c Hydrographs of exemplar wetland & water features (i.e. those with high correlation/low deviation relative to U Fldn). Note U Fldn monitor well water levels (right axis) are adjusted relative to wetland water levels (left axis) to remove elevation offsets (where present) and highlight behavioral responses. a Shallow, intermittently inundated sandhill wetland (Croom Road Marsh) and nearby U Fldn well (WR-6) with negligible elevation offset and minimal behavioral response deviations. b Deep, permanently inundated sandhill pond (Chapel Pond) and nearby U Fldn well (WW FLDN) with small elevation offset and small behavioral response deviations typical for sandhill features. c Seasonally inundated sandhill wetland (Ref 4) and nearby U Fldn monitor well (ROMP TR20-3) with somewhat higher elevation offsets (0.2-0.3 m, surface water and shallow groundwater, respectively) and typical behavioral responses. Highlighted is periodic zig-zag pattern potentially representing the Lisse Effect (Heliotis and DeWitt 1987;Weeks 2002). d-f Hydrographs of exceptional wetland & water features (i.e. unusual or with low correlation and high deviation relative to U Fldn). Note monitor well water levels (right axis) are adjusted relative to wetland water levels (left axis) to remove elevation offsets (where present) and highlight behavioral responses. d Large, multi-pool sandhill wetland (Weeki Wachee Prairie) with U Fldn monitor well (WWP) at shoreline and more distant U Fldn monitor well (WW Fldn). Here, the elevation offset (0.5 m) is unexpectedly high, but behavioral responses are typical for sandhill features. (Note, water levels from the nearby (WW Fldn) well were used as a proxy for the WWP well in the early POR. e Deep, semi-permanently inundated sandhill wetland (Banshee Pond) and nearby U Fldn monitor well (WR-6b). Note the dichotomous elevation offset and behavioral response deviations between low and high water level periods. f Shallow, seasonally inundated wetland (Sand Point Pond) and nearby U Fldn monitor well (ROMP 107). Note the extremely high elevation offset, poor tracking, poor correlation (R 2 = 0.43) and behavioral responses not typical of sandhill features response deviations are both numerous and inconsistent with those of the exemplar features, resulting in poor correlation with U Fldn water levels at the nearby monitor well (R 2 = 0.43) (Fig. 5f).

Water Geochemistry Analyses
Similarity and overlap between the water geochemistry of wetland/water features and that of the U Fldn are apparent in scatterplots relating field pH to specific conductance (pH-Cond) and to Ca 2+ (pH-Ca 2+ ). Two prominent pH-Cond or pH-Ca 2+ domains are apparent for both the dry and wet seasons-one defined by water residing in limestone of the U Fldn whose pH (5.6-8.8) and specific conductance (91-571 μS/cm) or Ca 2+ (10.5-47.7 mg/L) are highest as a group, and one indicative of rainfall whose pH (2.2-7.3) and specific conductance (15-33 μS/cm) or Ca 2+ (0.5-3.1 mg/L) are lowest as a group (Figs. 6a, b). Between these domains is an area where specific conductance or Ca 2+ values are intermediate, suggesting mixing between rainfall and limestone endmembers. While water samples from several wetlands and one or two ponds occur within the rainfall domains for pH-Cond and pH-Ca 2+ and show no clear input from water residing in limestone, many other wetlands and most of the ponds and lakes occur either in the limestone domain or in ▸ the area of endmember mixing. This is true for both seasons, indicating a clear overlap in feature/U Fldn geochemistry. It is important to note, however, that the U Fldn is defined not just by its limestone matrix, but also by the overlying surficial sands, through which its water table may move freely (given the lack of a confining unit between them). Thus if the pH-Cond and pH-Ca 2+ limestone domains were expanded to reflect the broader U Fldn domain (i.e. water within the limestone and surficial sands), then many more wetlands, ponds, and lakes would occur directly within it. Piper diagrams depicting major ion composition by sample type along a vertical gradient are presented in Fig. 7. Diagrams for both the dry and wet seasons indicate a calcium-carbonate water type for most surface water and groundwater samples at all depths, with a wet season increase in Cl − composition at surface waters due likely to a Fig. 6 a Scatterplot of field pH & specific conductance for dry and wet season water samples. Note the overlap of numerous surface water samples within the limestone endmember domains and in the area designated as endmember mixing. (Note, limestone water samples collected from wells adjacent to surface water features at WWP and HL were not included in the domain delineation because these wells were not designed for water chemistry sampling and may have been contaminated with drilling muds or bentonite). b Scatterplot of field pH & calcium ion (Ca 2+ ) for dry and wet season water samples. Note the overlap in surface water samples occurring within the limestone endmember domains and in the area designated as endmember mixing the direct input of increased rainfall. These diagrams show a clear overlap in major ion composition between samples collected from wetlands, ponds, and lakes (i.e. shallow and deep surface water and wetland surficial sand) and samples collected from the U Fldn (i.e. upland surficial sand and shallow, deep, and very deep limestone).
Results from the dry and wet season isotope analyses are presented numerically in Tables 3 and 4 (respectively) and relative to global and local MWLs in Fig. 8. Values for 2 H and 18 O are reported as the relative difference of heavy and light isotope ratios (δ 0/00 ). For the dry season, a clear demarcation is apparent between surface water samples, all of which are enriched in heavy isotopes (δ 2 H = + 2.6 b Fig. 6 (continued) to + 34.6 and δ 18 O = + 0.4 to + 7.9), and limestone water samples, all of which are depleted (δ 2 H = − 19.2 to− 11.9 and δ 18 O = − 4.0 to− 2.9). Surficial sand water samples were variable (δ 2 H = − 16.4 to + 13.4, δ 18 O = − 3.6 to + 2.5), showing isotope enrichment when collected from wetlands inundated at the time of sampling (R4, R8, WS) and isotope depletion when collected from wetlands that were not inundated (CrRi, CrRo, POM, SOP) and from uplands (M5, SH). For the wet season, results were more variable. All groundwater samples-both surficial sand and limestonewere depleted in heavy isotopes (δ 2 H = − 18.6 to − 12.1, δ 18 O = − 3.9 to − 2.8) except for one limestone anomaly (WWP), which was enriched (δ 2 H = + 5.1, δ 18 O = + 0.93) and may reflect a leak in the well's grouting. Surface water samples varied (δ 2 H = − 19.7 to + 9.1, δ 18 O = − 4.0 to + 2.0)-those depleted in heavy isotopes were collected from (Brooksville Ridge) features, which were inundated for only a short period prior to sampling, while those enriched were collected from (Gulf Coastal Lowlands) features, which were inundated much longer, allowing more time for lighter isotopes to evaporate.
While isotopic similarity between surface water features and groundwater from U Fldn monitor wells was not expected, variation in the degree of heavy isotope enrichment among surface water features was explored to see whether some features showed less enrichment due to greater U Fldn groundwater input (evaporation effects and rainfall presumed equal across features). One pond (TP) in particular, showed relatively low enrichment, even at the end of the dry season when maximal enrichment would be expected from greater evaporation and lesser rainfall (Table 3). Interestingly, Ca 2+ concentrations at this pond were the highest of all the surface water features for the dry and wet seasons and higher than water residing in limestone for the wet season (Fig. 6b). If rainfall dilution was the source of lower enrichment at this pond, then the Ca 2+ concentrations also Fig. 7 Piper diagrams for dry and wet season water samples along vertical gradient. Note most samples reflect a calcium-bicarbonate water type in both seasons would be lower, but they are not. This may suggest greater pond connectivity to water residing in limestone due to sitespecific hydrogeologic conditions. Because the site occurs in an area of high secondary porosity, large voids within the limestone matrix may allow groundwater to flow more quickly into and out of the pond, replacing enriched surface water with depleted groundwater, thereby reducing its overall enrichment.
AHC results depicting geochemical similarity between sample types along a vertical gradient are presented in Fig. 9. Water samples were clustered into three classes for the dry season and four classes for the wet season. For both seasons, Class 1 consists of samples at the high end of the vertical gradient comprised mostly of wetland surface water and shallow groundwater (surficial sand), while Class 3 and 4 consist of samples at the low end of the gradient comprised entirely of water in limestone. For both seasons, Class 2 consists of samples across the vertical gradient comprised of all water types and depths. Profile plots show specific conductance and Ca 2+ as the most influential parameters in class agglomeration. The classes and profile plots correspond to the pH-Cond and pH-Ca 2+ domains identified in the scatterplot analyses for rainwater (i.e. Class 1), limestone (i.e. Classes 3 and 4), and mixed rainwater-limestone endmembers (Class 2). The clustering of wetland, pond, and lake samples in Class 2 with those from U Fldn monitor wells further indicates an overlap in their geochemistry, which is apparent in both seasons.
Significant differences in geochemical parameters (temperature, pH, specific conductance, 2 H, 18 O, Na + , K + , Mg 2+ , Ca 2+ , Cl − , SO 4 2− , HCO 3 − ) between water types (wetland, pond/lake, surficial sand, and limestone) were assessed using multiple pairwise comparison tests. In these tests, wetland surface water was evaluated separately from that of ponds and lakes to investigate differences apparent in prior analyses. Results confirm that water from wetlands has significantly lower specific conductance and Ca 2+ than water from ponds/lakes and from limestone (Table 5). Specific conductance and/or Ca 2+ in pond/lake water, however, is not significantly different than groundwater from limestone, suggesting pond/lake features may have better connectivity to water residing in limestone than do wetlands. Wetland connectivity to the U Fldn is not ruled out, however, as wetland geochemistry is not significantly different than that of groundwater residing in the U Fldn surficial sands.

Discussion
With only a single published study describing the hydrology of a sandhill lake in the study area (Henderson 1986), little is known about the scores of other sandhill wetlands, ponds, and lakes that occur, including their connectivity to a large regional aquifer, the U Fldn. Hydrologic controls have been described for sandhill and similar features elsewhere in the state, but these studies point to local hydrologic controls such as perched water tables (Shannon and Brezonik 1972) or a surficial aquifer (Jones Edmunds 2006;CH2M Hill 2005;Lee et al 2009). In this study, we provide physical and geochemical evidence that sandhill wetland/water features in the study area are hydraulically connected to the U Fldn. Findings here debut sandhill wetland/water features for this rare quality, advancing their fundamental understanding at a time when natural resource managers must weigh resource protection against increasing groundwater demands, expanding residential development, and a changing climate. This study also identifies the defining hydrologic characteristics of sandhill wetland/water features that may help resource managers distinguish and evaluate features of similar connectivity, here or wherever they may occur.
In this study, physical and geochemical data from 19 wetland, pond, and lake features were compared to that of 12 U Fldn monitor wells to examine similarity in water level elevations, fluctuations, and geochemistry as indicators of feature-U Fldn hydraulic connectivity. Findings here expand on a study by Henderson (1986), which shows similar water level behavior (and geochemistry) between a lake in this study (Hunter's Lake) and nearby U Fldn monitor wells and between that lake and a wetland in this study (Weeki Wachee Prairie). Similarly widely ranging water levels and "sympathetic fluctuations" between potential sandhill waters elsewhere in the state were reported by others (Deevey 1988), but these features are not hydraulically connected to the U Fldn, nor does it control their hydrology (Sacks et al. 1998;and Swancar et al. 2003).

Physical Evidence of Hydraulic Connectivity
Results show that water levels for all but two wetland/water features were synchronous with those of the U Fldn-their elevations analogous and fluctuations highly correlated (0.84 ≤ R 2 ≤ 0.99), especially between ponds and the U Fldn (0.94 ≤ R 2 ≤ 0.99). Elevation offsets of up to 3 m were noted between features and wells and reflect their different positions along the regional hydraulic gradient (Fig. 1). Differences in residual variation were examined qualitatively and may be attributed to inherent factors (e.g. feature size, shape, and depth) and situation (e.g. antecedent conditions, landscape setting, adjacent land use/land cover, and rainfall intensity/duration). The effects of these factors may be greatest at: (1) the larger lakes and wetlands (e.g. Tooke Lake, Willow Sink) whose greater surface areas may contribute greater losses to evaporation and whose longer shorelines may contribute more opportunities for surface water-groundwater exchange (Wetzel 2001;Lee et al. 1997); and (2) the seasonally inundated wetlands (e.g. Ref 4, Ref 8, String of Pearls Marsh), where water more frequently shifts between surface water and shallow groundwater phases and where accumulation of organic material may slow or speed losses to leakage or evapotranspiration (respectively) or contribute to a rarely noted phenomenon called the Lisse Effect. The Lisse Effect occurs in response to intense rainfall, which inundates the wetland so rapidly it traps air beneath the soil wetting front. The trapped air builds up pressure, which artificially raises the head in the shallow monitor well. When the pressure is released, the water level in the monitor well equilibrates with the water table, reflecting the actual recharge that occurred from the rainfall event (Heliotis et al. 1987;Weeks 2002). On a hydrograph, this would produce a zig-zag response in shallow groundwater levels as shown (for Ref 4) in Fig. 10.
The relative simplicity of the ponds (i.e. smaller size, simpler shape, and single surface water phase) may explain why their residual variation is least among feature types. Similarly, the smaller size, simpler shape, and largely groundwater phase at Croom Road Marsh may explain its near perfect water level correlation with that of the U Fldn monitor well. For this wetland, proximity to the well (< 1 km) explains not only its coincident water level elevations, but is likely a proxy for strong hydrogeologic similarity. This may be the most important factor contributing to the low residual variation between feature and well water levels.
A compilation of the physical attributes of the sandhill wetlands, ponds, and lakes investigated here suggest the following defining hydrologic characteristics: widely ranging water levels (2-5 m or more) that are synchronous with those of the U Fldn (Figs. 3a, b), deviating in consistent and predictable patterns (Figs. 4a, b), which result in very high correlations (84 ≤ R 2 ≤ 99%) ( Table 2). Within these Fig. 9 ACH analysis with profile plots for dry and wet season water samples along a vertical gradient. The ACH algorithm grouped samples into three dry season and four wet season classes: Class 1 consists mostly of wetland surface and shallow groundwater (surficial sand) samples indicating rainwater geochemistry; Class 3 and 4 consist only of limestone samples indicating limestone water geochemistry; and Class 2 includes a mix of all sample types, suggesting a geochemistry of rainwater-limestone water mixing. The profile plots identify specific conductance (Cond) and calcium (Ca 2+ ) as the most influential parameters in generating these groupings, followed by 2 H and SO 2− attributes, these features may vary markedly in their hydrologic expression-from small, shallow elliptical wetlands that remain dry for years (e.g. Croom Road Marsh) to deep circular ponds (e.g. Chapel Pond) and large amorphous lakes (e.g. Hunter's Lake) that maintain permanent inundation ( Figs. 1 and 5a). Each of these features, though markedly different in expression, are exemplars of the sandhill type.
Four of the features investigated were considered exceptional for showing uncharacteristic or unexpected water level behavior relative to the exemplars. At Sand Point Pond, the elevation offset from the nearby well is markedly high (12.5 m) and the correlation markedly low (R 2 = 0.43) (Figs. 3b and 5f). At Perry Oldenburg Marsh, the elevation offset appears much less (0.12 m), but is artificially low considering the 3 m drop in head that occurs between the well (WR-6b) and wetland as a result of the regional hydraulic gradient (see contour elevations, Fig. 1) (i.e. a more proximal well would show a greater offset). More importantly, the offset is highly variable over the POR, and the correlation is very low (R 2 = 0.48) (Fig. 3b). Water levels at both wetlands reflect control by a perched water table(s) due to remnant clay deposits in the otherwise unconfined U Fldn. For their lack of connectivity to or hydrologic control by the U Fldn, neither wetland would be considered the sandhill type.
Two other wetlands, though exceptional, would still be considered the sandhill type. At Banshee Pond, dichotomous Key findings indicate specific conductance and Ca 2+ from wetland surface water samples are significantly different than groundwater sampled from surficial sand and limestone, while surface water sampled from ponds/lakes are not Effect. This occurs when intense rainfall seals the surface and builds up pressure in the monitor well. The pressure creates an artificial rise in head in the well, which equilibrates shortly after to reflect actual recharge (Heliotis et al. 1987;Weeks 2002) water levels-synchronous during periods of high water, but not during low water (Fig. 5e)-are attributed to historical excavation. The excavation deepened the wetland bottom into the limestone residuum, altering the way water drains through it (and increasing the hydroperiod from intermittent to semi-permanent). The residuum, which has a clayey texture, is believed to perch water at a certain threshold elevation (at or around 13.5 m), disconnecting it from the U Fldn water table as it drains. As the water table rises above the threshold, it converges with the perched surface water, and they reconnect. This is evident in the hydrograph where both the elevation offset and correlation coefficient shift markedly between the low and high water periods. This wetland would still be considered a sandhill wetland, but with a modified hydrology due to excavation. Sandhill wetlands/waters of Florida have historically been excavated to capture stormwater and to increase hydroperiods for aesthetic or agricultural purposes, increasing the possibility of scenarios like this at other locations of similar geology.
At Weeki Wachee Prairie, water levels are synchronous and highly correlated (R 2 = 0.88) to the nearby U Fldn monitor well (WWP), but the elevation offset was unexpectedly high (0.5 m) considering the well is located at the wetland shoreline (Fig. 5d). Such is the case at Hunter's Lake whose U Fldn monitor well (HUNT) also sits at the shoreline, but where the elevation offset is comparatively small (0.2 m). The study by Henderson (1986), which describes the close relationship between Hunter's Lake and the U Fldn, may help explain this. Henderson describes Hunter's Lake and Weeki Wachee Prairie as flow-through systems, where groundwater enters from one side and leaves from another. At Hunter's Lake, the U Fldn well is located on the up-gradient shoreline where U Fldn water table contours are gradual and translate to a smaller offset; on the downgradient shoreline, the contours are steeper and translate to a larger offset. At Weeki Wachee Prairie, the U Fldn well is (presumably) located on the down-gradient shoreline, where contours may be similarly steep. Detailed contours are not available to confirm this, but like Hunter's Lake, Weeki Wachee Prairie is a large feature situated at the base of a parabolic dune train (Upchurch et al 2018) and may be subject to similar local gradients. Only at Hunter's Lake and Weeki Wachee Prairie are the U Fldn wells located at the shorelines. Monitor wells evaluated for all other features are located 1-10 km away because adjacent wells do not exist. Water levels at these more distant wells may better reflect the regional groundwater flow system because they are not subject to the local complexities and vertical gradients that are created by the surface water features themselves. Historical excavation or site-specific hydrogeologic conditions at this very large wetland also may affect this wetland's relationship with the U Fldn.

Geochemical Evidence of Hydraulic Connectivity
Results show that the water geochemistry of features in the study area are similar to or overlap that of the U Fldn monitor wells, though not entirely as expected. In many shallow features (e.g. wetlands), specific conductance, Ca 2+ and/or pH are low and reflective of rainwater. In many deep features (e.g. lakes and ponds), these attributes are elevated, reflective of mineralized water in contact with limestone (Figs. 6a,  b). In the remaining features (wetlands, ponds, and lakes included), the attributes are intermediate, suggesting a mix of the rainwater and limestone endmembers. The low specific conductance, Ca 2+ , and/or pH at the other wetlands and ponds do not, however, negate their connectivity to the U Fldn. These features simply do not intersect that part of the U Fldn that contains water influenced by limestone.
Water in the U Fldn is chemically stratified into an upper rainwater lens residing in the surficial sand, a lower body of Ca 2+ -rich water residing in the limestone, and a transitional, or mixing, zone in between (Fig. 11). The position of the mixing zone, which is controlled by the expansion and shrinking of the rainwater lens, and the depth of the features determine what type of water chemistry the features will have. Water in features that are not deep enough to intercept the mixing zone will maintain the chemistry of the rainwater lens. Water in features that are deep enough to intercept the mixing zone (or the limestone water itself) will reflect a mixed rainwater-limestone (or limestone water) chemistry.
The importance of assessing the physical characteristics of feature hydrology to determine connectivity cannot be overstated, as geochemistry alone may be confounded by other factors. This is particularly true of features whose chemistry reflects that of rainwater, which may be attributed to: their occurrence within the rainfall lens of a regional aquifer (as described above), direct rainfall input where a water table is perched, or direct rainfall and groundwater input from a surficial aquifer. Anthropogenic factors also may confound the determination (e.g. input from groundwater influenced by sewage or septic tank drainage, Shannon and Brezonik 1972). For example, Ca 2+ and specific conductance from surface water samples at Sand Point Pond (SPP) and shallow groundwater at Perry Oldenburg Marsh (POM) occur within the area of rainwater-limestone endmember mixing, despite perched water tables at both sites (Figs. 6a, b). The noteworthy vadose zone (12 m) at Sand Point Pond precludes any possibility of a hydraulic connection with the U Fldn (Fig. 3b). Its higher Ca 2+ and specific conductance are likely the result of runoff or leakage into the wetland from an adjacent tree nursery irrigating with groundwater pumped from the U Fldn. At Perry Oldenburg Marsh, the vadose zone is only 3 m thick, so water from the rainwater-limestone mixing zone may have risen into the wetland's shallow groundwater, influencing its geochemistry (although not controlling its water levels) ( Fig. 3b; Table 2).

Other Findings of Regional Hydraulic Connectivity
Few other studies offer evidence of wetland/water hydraulic connectivity to a regional aquifer. Wolfe (1996) and Blood et al. (1997) both examined connectivity for karst wetlands outside of Florida. Wolfe compared water levels of two types of wetlands in the eastern Highland Rim of central Tennessee USA to water levels of wells in the regional Manchester/Fort Payne aquifer. Compound sinks (large, steep sided depressions) were found to connect to the regional aquifer by way of multiple internal sinkholes, while karst pans (shallow features with no internal drainage) showed perched water tables with no connectivity. In Wolfe's study, he was able to use wetland geomorphometry as a tool to identify wetlands likely to be hydraulically connected to the regional aquifer (e.g. compound sinks), whereas in this study, geomorphometry offers no indication, given the great variation that occurs in features (e.g. shallow elliptical wetlands, deep Sandhill wetlands as regional groundwater endmembers along a GIW hydrologic continuum of 25 circular ponds, and large amorphous lakes) that are all similarly connected to the U Fldn. Blood et al (1997) compared water levels of isolated wetlands in the karstic Dougherty Plain of southwest Georgia USA with those of wells in the regional U Fldn. They, too, found wetlands with perched water tables not connected to the U Fldn, as well as wetlands whose hydrodynamics were dependent upon "interconnected groundwater systems". These groundwater systems are believed to reflect different zones within a single unconfined aquifer, the U Fldn, as characterized by Miller (1990) (and if so, hint at a more complex type of wetland-U Fldn connectivity than presented here). The authors do mention the existence of "aquiferemergence wetlands" in the Dougherty Plain, but suggest not all are of this kind because not all reflect "direct aquifer discharge" in their surface water chemistry. Specific geochemistry and water level data are not provided to compare between wetlands and monitor wells, but it is possible the lack of mineralized surface water chemistry in some of the Dougherty Plains wetlands is related to their position above a rainwater lens, as described above for sandhill wetlands in Florida.

Conclusion and Implications of Findings
From raised bogs (Large et al. 2007) and vernal pools (Schlising and Saunders 1982;Rains et al. 2006) to fens (Wilcox et al. 1986) and Carolina bays (Lide et al. 1995), studies of GIWs have long characterized hydrologic control along a continuum of local forces, from precipitation to groundwater of a surficial aquifer. Few studies, however, document hydrologic control by groundwater of a regional aquifer (e.g. Henderson 1986). This study provides physical and chemical evidence that GIW/water features in the sandhill of west-central Florida are hydraulically connected to a large, regional water supply aquifer, the U Fldn. This connection distinguishes them from most other GIW/water features and places them at the far end of the hydrologic continuum (Fig. 12).
Key findings show that water levels of sandhill features range widely (2-5 m), are analogous in elevation, and are very highly correlated to the regional water table. Findings also show that connectivity to the regional aquifer does not guarantee a mineralized chemistry as might be expected. While the chemistry of deeper features (including many lakes and ponds) is mineralized, the chemistry of most shallow features (wetlands) reflects that of rainwater. This is not because shallow features are not connected to the regional aquifer, but because they sit within its upper sandy part, which contains the rainwater lens-above the depth where rainwater mixes with water residing in limestone.
Future studies of interest may further investigate: (1) factors affecting residual variation; (2) features whose water tables (by topographic coincidence) reflect both perching and connectivity to the regional aquifer (HCUD unpublished data); and (3) whether features are flow-through systems like Hunter's Lake (Henderson 1986) or recharge/discharge systems like karstic-ridge wetlands in central Florida (Knowles Jr. et al. 2005). Also, while isolated wetlands and waters in Florida are protected at the state-level, their peculiar ecohydrology (i.e. hydrologic cycles lasting 4-10 years or longer, with ecological expressions varying widely in response) has long spurred speculation and misunderstanding of what constitutes health and impact. This has limited both the appositeness of commonly used regulatory assessment methods and boundary determinations. Findings here advance understanding of their unique character and are intended to support the development of novel assessment methods. They also may assist others studying GIWs in karst terrain or wherever the potential for regional hydraulic connectivity may occur.