Integrated Conduit Systems Discharging at Springs Dominate the Flow-Field in Karst Aquifers
River reversals into karst springs: a model for cave enlargement in eogenetic karst aquifers
Jason Gulley, Jonathan B. Martin, Elizabeth J. Screaton, Paul J. Moore*
Department of Geological Sciences
P.O. Box 112120
University of Florida
Gainesville, FL 32611-2120
1
Abstract
Most conceptual models of epigenic conduit development assume that conduits sourcing karst springs form as water that is undersaturated with respect to carbonate minerals flows from recharge to discharge points. This process is not possible in springs fed by distributed recharge that is transmitted through aquifer matrix porosity, such as unconfined aquifers in eogenetic carbonate rocks. Diffusely recharged water has a long residence time within the aquifer, and thus would have equilibrated with the aquifer rocks prior to discharge to the conduits. The Upper Floridan aquifer (UFA) has high matrix permeability (~10-13 m2) and many springs lack discrete inputs of undersaturated allogenic water in their recharge areas. Consequently, another explanation for their development is necessary. During flooding of the SuwanneeRiver in north-central Florida, water highly undersaturated with respect to carbonate minerals commonly recharges the UFA through spring vents, and solution scallops oriented away from the vents suggests most dissolution along conduit walls occurs during these flow reversals. During a single flow reversal at the Peacock Spring cave system, flood water was capable of dissolving up to 3.4 mm of the conduit wall rock. Dissolution occurs as flow reversals follow pre-existing features that include joints and paleo-water table caves. Lack of speleothems in conduits in the UFA has been used as evidence that the caves formed in the phreatic zone; however, flooding would dissolve any speleothems that may have formed during previous subaerial exposure. Conduit enlargement during flow reversals suggests that dissolution can progress in the normal upstream directions and this process may be an important driver of dissolution in any karst aquifer with outflows to surface water that are subject to flooding. Flow reversals would also introduce dissolved organic carbon and oxygen into the groundwater and provide important energy sources for cave ecosystems as well as altering redox chemistry of the aquifer water.
Introduction
Conduits commonly form in diagenetically mature carbonate aquifers with low matrix porosity and permeability(termed telogenetic karst by Vacher and Mylroie, 2002) when undersaturated allogenic runoff flows into discrete recharge points such as sinkholes or swallets. This recharge dissolves the rock along joints and bedding planes, thereby expanding these preferential flow paths into conduits (Palmer, 1991). Fully mature conduits thus often link recharge and discharge points in these systems.
In contrast, processes forming conduits remain poorly understood in aquifers with high matrix porosity and permeability(termed eogenetic karst by Vacher and Mylroie, 2002). These aquifers typically occur in tropical marine settings and have not undergone burial diagenesis that would occlude the primary depositional porosity and permeability. Because the high permeability matrix allows rapid infiltration of recharge as diffuse flow through the surface (e.g., Ritorto et al., 2009), point recharge (i.e. allogenic recharge) at sinking streams is less common than in telogenetic karst aquifers and generally only occurs where streams flow off confining layers onto the carbonate aquifer (e.g., Screaton et al., 2004). Where sinking streams do exist, their potential for focused dissolution is greatly diminished because of the large volumes of water stored in the matrix porosity, which is commonly equilibrated with carbonate minerals of the aquifer (Moore et al., 2009).
Little allogenic recharge occurs in the eogenetic karst of the Upper Floridan aquifer (UFA), except for where streams flow off the edge of the confining layer and into the unconfined aquifer. Nonetheless, many of Florida’s springs discharge from laterally extensive phreatic conduit systems (Florea and Vacher, 2007; Martin and Gordon, 2000). Because of their distance from the coast, these conduits could not have formed from mixing of fresh and saline water as has been proposed for caves in the eogenetic limestone of the Yucatan (Smart et al., 2006) and the Bahamas (Mylroie and Carew, 1990). The general lack of allogenic recharge limits input of water undersaturated with respect to carbonate minerals into pre-existing high-permeability zones at the upstream end of the conduits. These conduits are assumed to have formed in the phreatic zone and not been subject to past subaerial exposure because most lack speleothems, unlike conduits in the Yucatan and the Bahamas. Conduit dissolution has been proposed to occur in phreatic zone from “headward sapping” in which high permeability zones act as low resistance drains and cause flow paths to converge and concentrate dissolution, further focusing flow and dissolution (c.f., Rhoades and Sinacori, 1941; White, 2001). However, water from the UFA is generally saturated with respect to calcite (Martin and Gordon, 2000; Moore , 2009), and thus headward sapping is unlikely to form the conduits found there. Consequently, the origin of submerged eogenetic karst conduits remains unresolved.
In this paper, we use legacy data and new observations from the SuwanneeRiver watershed in north-central Florida to suggest that reversals of springs during flood events provide a mechanism to form or enlarge conduits. Spring flow reverses when river stage increases faster than the hydraulic headsin the aquifer. Although backflooding of air-filled caves has been observed in telogenetic karst regions (White and White, 1989) and some Florida springs have been reported to reverse (Opsahl et al., 2007), the importance of chemical processes such as dissolution during spring reversals has not been evaluated. We establish that surface water is undersaturated with respect to carbonate minerals during high discharge events and use observations of solution scallop direction in two water-filled conduit systems to support the concept that dissolution occurs during spring reversals. We collected high-resolution specific conductivity records within two conduit systems and geochemical data during reversal of one spring to document the influx of highly undersaturated water during flooding. These data allow an assessment of the magnitude of dissolution by flood waters.
Study Locations
The Suwannee River watershed in north-central Florida is entirely underlain by the Floridan Aquifer System (FAS), a thick sequence of limestone and dolomite that is subdivided into the UFA, a middle confining unit (where it exists), and the Lower Floridan aquifer (Miller, 1986). The Cody Scarp generally marks the boundary between the confined and unconfined regions of the UFA and separates the NorthernHighlands and Gulf Coastal Lowlands physiographic areas (Fig. 1). In the Northern Highlands, the UFA is overlain by the confining siliciclastic Hawthorn Group and the Surficial Aquifer System. Water sources to the SuwanneeRiver include the Surficial Aquifer System and runoff, which provide tannic-rich water due to organic matter contributions from wetlands. Downstream of the Cody Scarp, the Suwannee River watershed transitions to being sourced by the UFA, including discharge from more than 100 springs (Rosenau et al., 1977; Scott et al., 2004). These springs include 9 of Florida’s 27 first magnitude springs, which are defined as having a discharge of > 2.8 m3/sec, (i.e., > 100 cfs: Meinzer, 1927). At baseflow, these springs discharge water that is saturated with respect to carbonate minerals, reflecting equilibration with the aquifer rocks.
Groundwater of the UFA has higher specific conductivity than surface water because of the high dissolved load of carbonate minerals but it has lower dissolved organic carbon concentrations, and thus does not have the characteristic tannic stain of water from the surficial aquifer system or surface water draining the Northern Highlands. Differences in specific conductivity and staining between groundwater and surface water are particularly strong during floods (e.g., Moore et al., 2009), providing natural tracers that allow separation of flood water flowing off of the Northern Highlands from groundwater of the UFA.
Flooding is common in winter and spring from rainfall associated with cold fronts and in late summer and fall from tropical storms (Grubbs and Crandall, 2007). These floods frequently elevate river water levels that have their headwaters in the Northern Highlands above the hydraulic head of the unconfined UFA(Martin and Dean, 2001; Martin et al., 2006; Ritorto et al., 2009). The changing hydraulic gradients cause springs to reverse, as shown by whirlpools at spring vents and changes in water levels and tannins in wells up to 4.8 km from the river (Crandall et al., 1999).
We investigate flow and chemical composition of water at two springs (Madison Blue and Peacock) that discharge within the SuwanneeRiver watershed in north-central Florida (Fig. 1). Madison Blue Spring is classified as a first magnitude spring that contributes baseflow discharge of 2.0 to 3.9 m3/sec to the Withlacoochee River via a short stream connected to the spring vent known as a spring run (Rosenau et al., 1977; Scott et al., 2004). The WithlacoocheeRiver is a major tributary to the SuwanneeRiver, and Madison Blue Spring is located about 12 km upstream of the convergence of the two rivers (Fig. 1). More than 8 km of passages have been mapped in Madison Blue Spring (Fig. 2A). Additional passages are known but have not yet been surveyed.
Peacock Spring, locatedabout 67 km downstream of Madison Blue Spring and 2.3 km north of the SuwanneeRiver (Fig. 1), lacks a conduit connection or a spring run to the SuwanneeRiver.The ‘spring’ is technically a group of water-filled sinkholes (karst windows) that lead to 7.5 km of mapped conduits (Fig 2B). The SuwanneeRiver periodically floods and inundates the spring along a normally dry channel that connects the river to the entrance of the conduit system. The channel contains a sill that restricts direct infiltration of river water into conduits to times when river water elevation exceeds~8 masl (Rick Owen, Florida Department of Environmental Protection, Personal Communication).
We have also made cave-diving observations of conduit wall morphology at Madison Blue and Peacock Springs, as well as at two other locations, Little River and Cow Springs, that were not sampled for this study. Little River Spring is located about 18 km downstream from Peacock Spring and Cow Spring is located a few hundred meters north of the Suwannee River about 3 km southeast of Peacock (Fig. 1).
Methods
Legacy flow data were collected by the Suwannee River Water Management District (SRWMD) and the United States Geological Survey (USGS) for the WithlacoocheeRiver near Madison Blue Spring at Lee, Florida (USGS station 02319394) and the SuwanneeRiver near Peacock Spring near Luraville, Florida (USGS station 02320000). The Lee station is approximately 10 km downstream from Madison Blue Spring and the Luraville station is approximately 3 km upstream from Peacock Springs (Fig. 1). Discharge data from Madison Blue Spring were provided by the USGS from continuous velocity measurements from a current meter (USGS station 02319302). Chemical composition data was also collected by the SRWMD for water discharging from Madison Blue Spring and from the rivers at the Lee and Luraville stations. Similar flow and chemistry data are unavailable for Peacock Spring. We used the chemistry data to calculate calcite saturation indices (SIcal) using PHREEQC with the LLNL database (Parkhurst and Appelo, 1999). We define SI values here as the log of the ion activity product divided by the equilibrium constant for calcite dissolution reaction.
Two observation periods of spring reversals are reported here: one in fall 2008 and the other in spring 2009. During fall 2008, Tropical Storm Fay passed through the area causing a minor flood. This event was recorded by a Schlumberger Conductivity-Temperature-Depth (CTD)-diver that was installed at the entrance to Peacock Spring to make time-series measurements of specific conductivity (SpC) and temperature (T).
In spring 2009, specific conductivity and T were also monitored at 20-minute intervals during and following major flooding in April 2009 with CTD-divers installed at the entrances to Peacock and Madison Blue springs and at six locations within conduits sourcing these springs (Fig. 2). CTD-divers were installed within the conduits at distances from the main spring vent of 152 m (Martz Sink), 610 m (Courtyard) and 1097 m (Back Section) in the Madison Blue Spring conduits, and at distances of 214 m (between Pothole and Olsen sinks), 884 m (Challenge Sink), and 1067 m (Distance Tunnel) in the Peacock Spring conduits. CTD-divers have accuracies for Tof + 0.1º C and SpC of +1%. The CTD-divers record pressure to a maximum depth of 10 m of water, which was exceeded during most of the flood and thus we have no data for water depths. CTD-divers were also installed at the Lee and Luraville stations during spring 2009 but the flood covered both sensors with sediment and prevented any data collection.
To complement the SpC and T data, water was collected six times (16 and 24 April, 1, 8, and 15 May, and 14 July 2009) during the April 2009 flood and its recession at the Luraville station and from two sinkholes, Challenge and Orange Grove sinks, that intersect Peacock Spring conduits (Fig. 2B, Table 1). Samples could not be collected at Madison Blue Spring during the April 2009 flood because roads to the spring were submerged, making the spring inaccessible. Water was collected by extending a PVC tube from the banks to directly above the center of the sinkholes and to about 5 m from the banks of the river. The tubing was connected to a peristaltic pump, which drew water into an overflow cup. The water was monitored for its SpC, T, dissolved oxygen (DO) concentration, and pH using a calibrated YSI model 566 multi-parameter field meter, and pumping continued until all values stabilized. Following stabilization, samples were collected in PVC bottles for analyses of alkalinity and major element concentrations and kept chilled until measurement. Samples for measurements of cation concentrations were preserved with nitric acid. Alkalinity was titrated within one day of collecting the samples using the Gran method (e.g., Drever, 1997) and the major element concentrations were measured using a Dionex Model 500DX ion chromatograph (IC) in the Department of Geological Science, University of Florida.
Most of the samples collected during the first three weeks of the flood had very low solute concentrations and were at or near the detection limit of the IC. Consequently, their charge balance errors are large, averaging around 17%. Charge balance errors (CBEs) are less for samples collected during the flood recession, averaging around 4%. We used PHREEQC (Parkhurst and Appelo, 1999) to calculate SIcal based on these data. Charge balance was alternately forced on Ca2+ and alkalinity concentrations by increasing or decreasing the concentration of Ca2+ and alkalinity within the calculations until charge balance was achieved to assess the impact of CBEs on calculated calcite saturation index. Forcing charge balances changes the SI less than 1 SI unit for samples with high CBEs and for most samples changes the SI less than 0.l SI unit. While an error margin of 1 SI is large, the sample with largest CBE was still had an upper calcite SI of nearly -4 and reflects that the water is capable of dissolving considerable amounts of calcite.
While some data was lost due to equipment being damaged by flooding or logistical constraints, data that were collected clearly demonstrate large volumes of undersaturated river water flow into springs during springs reversals. Missing data include discharge at Madison Blue spring for the April 2009 flood, which destroyed the USGS discharge gauging station at Madison Blue Spring. Peacock Spring is not gauged because it lacks a spring run. We thus estimated the rate and volume of river water intruding into the springs by dividing the distance between CTD divers by the time it took for flood water to pass them, as estimated from changes in SpC of the water. We assume river water flowed into the conduits during the time of decreasing and sustained low conductivity and that springs began to discharge when the SpC rose following the maximum flood elevation. To estimate the total amount of recharge during a reversal, our calculations assume flow was maximum at the start of the reversal and decreased linearly until the reversal stopped. We convert the flow rate to a volume of water based on an estimated average conduit diameter of 3 m, which is consistent with our observations of the water-filled conduits. There are large variations in conduit diameter and cross-sectional morphology in both systems, 3 m is considered to be a rough average. More accurate assessment of influx volumes would require detailed conduit cross section measurements and either continuous velocity measurements or head data from the conduits and surface water, which were not available.
Results
SICal response to elevated river discharge. Legacy data demonstrate thatthe Withlacoochee and Suwannee rivers have an inverse exponential relationship between discharge and SIcal (Fig. 3). Highest river discharges approach 500 m3/sec in the WithlacoocheeRiver at the Lee station and 1000 m3/sec in the SuwanneeRiver at the Luraville station. Water during high flow events can reach SIcal values of < -4 at both stations across a wide range of discharges and most likely reflects the importance of antecedent aquifer heads as a control on river water chemistry.