Chemistry, Growth, and Fate of the Unique, Short‐Lived (2019–2020) Water Lake at the Summit of Kīlauea Volcano, Hawaii

Less than a year after the 2018 Kīlauea caldera collapse and eruption, water appeared in newly deepened Halemaʻumaʻu crater. The lake—unprecedented in the written record—grew to a depth of ∼50 m before lava from the December 2020 eruption boiled it away. Surface water heightened concerns of potential phreatic or phreatomagmatic explosions but also offered a new means of possibly identifying eruption precursors. The U.S. Geological Survey Hawaiian Volcano Observatory (HVO) monitored the lake via direct visual observation, webcams, thermal imaging, colorimetry, and laser rangefinders. HVO also employed uncrewed aircraft systems to sample the water and measure near‐lake gas composition. The lake's δD and δ18O indicate a groundwater source with substantial evaporation. The initial sample had a salinity (total dissolved solids concentration) of 71,000 mg/L and was rich in sulfate (∼53,000 mg/L), iron (∼500 mg/L), and magnesium (∼10,000 mg/L). Subsequent samples were slightly more dilute. The water's pH (∼4), δ34S (+4.3‰), and surface temperatures (up to 85°C) suggest, rather than significant scrubbing of magmatic volatiles, leaching of basalt and reactions with sulfate minerals resulted in high concentrations of sulfate and other solutes. Thermodynamic modeling and precipitate mineralogy indicate that water composition was controlled by iron oxidation and sulfate dissolution. Although the lake exhibited no detectable precursors before the next eruption, and phreatic or phreatomagmatic explosions did not materialize, our multi‐parameter approach to monitoring yielded an enhanced understanding of the hydrologic, geologic, and magmatic conditions that led to the formation of the unique and short‐lived lake.


Introduction
Throughout the 19th and 20th centuries, and into the 21st, volcanic activity at Kīlauea Volcano was almost uniformly typified by effusive eruption of lava lakes, lava flows, and lava fountains (Holcomb, 1987;Wright & Klein, 2014).Indeed, in more recent history, the Puʻuʻōʻō eruption on the volcano's East Rift Zone was in a state of near-constant eruption of lava flows for 35 years since its onset in 1983 (Heliker et al., 2003;Orr et al., 2015).A lava lake in Halemaʻumaʻu summit crater added a second persistent eruption site for the decade between 2008 and 2018 (Patrick, Orr, et al., 2021).The lower East Rift Zone eruption and associated summit collapse of 2018 brought about the end of this decades-long effusive era (Neal et al., 2019;Patrick et al., 2020).Less than a year after the end of the 2018 activity, an unprecedented (in the written record; i.e., the last ∼200 years) water lake began to form in the newly deepened Halemaʻumaʻu (Figures 1 and 2; Nadeau et al., 2020;Swanson, 2019).
Caldera collapse alone was enough to cause concern over increased potential for large, explosive eruptions, as Kīlauea is known to cycle through alternating periods of effusive-and explosive-dominated activity (Swanson et al., 2014); the addition of external water further exacerbated concern over a potential increase in explosive activity.Highly explosive periods have followed caldera collapse events (Swanson et al., 2015), including phreatic and phreatomagmatic events that took place in a deepened caldera through a surficial water body (Mastin, 1997;Mastin et al., 2004).Though a crater lake had not previously been observed in Kīlauea's written history, some stories and chants (mele) in Hawaiian oral tradition have been interpreted as possibly alluding to the presence of such water at the surface in association with the same explosive periods (Swanson, 2008;Swanson et al., 2012).Most recently, the explosive activity of 1924 (Jaggar & Finch, 1924) was long thought to have been caused by the incursion of groundwater into a heated volcanic conduit when the crater collapsed beneath the water table (e.g., Dvorak, 1992;Stearns, 1925).Ultimately, modeling showed that such a scenario was not feasible in 2018, nor would it have been likely in 1924; over the course of years, a column of magma heats the surrounding rock enough to prevent the inflow of liquid water for months to years following the initiation of cooling (Hsieh & Ingebritsen, 2019).The beginnings of the water lake at Kīlauea were spotted on the ground (517.5 m a.s.l.) roughly 14.5 months after the lava lake drained out of sight, which confirmed that the summit region had cooled sufficiently to allow liquid water to infiltrate.
Though groundwater flow modeling suggested that water-driven explosions were unlikely immediately following the draining of a long-lived lava lake (Hsieh & Ingebritsen, 2019), the appearance of a water lake brought with it the possibility of phreatic or phreatomagmatic explosions upon the eventual renewed influx of magma into the shallow summit region.Injection of basaltic magma into standing water can lead to fragmentation, owing to the interaction between magma and external water, which results in phreatomagmatic eruption (often Surtseyanstyle; e.g., Belousov & Belousova, 2001;Houghton et al., 2015;Lorenz, 1987;Manville, 2015;Thorarinsson et al., 1964).Additionally, an increased influx of magmatic gas into the lake can result in fast shallowing of the subsurface vapor-liquid boundary, leading to rapid vaporization of liquid water and explosion in volcanic systems with a summit crater lake (e.g., Montanaro et al., 2022;Stix & de Moor, 2018).
With geologic evidence for explosive events at Kīlauea occurring in association with surface water (e.g., Mastin, 1997;Mastin et al., 2004;Swanson & Houghton, 2019), and with established linkages between the summit groundwater level and chemistry and magmatic activity at Kīlauea (e.g., Hurwitz & Anderson, 2019;

Visual and Thermal Observations
The emerging lake was initially spotted by the helicopter crew of a lidar survey on 25 July 2019; HVO staff visually confirmed the presence of water a few days later, on 1 August 2019, during another helicopter overflight.Subsequently, helicopter overflights provided aerial photos and thermal images that were used to produce visible and thermal orthomosaics using Agisoft MetaShape structure-from-motion software.
On the ground, field crews made visual observations several times per week from the western rim of Kīlauea's crater, approximately 1.2 km line-of-sight distance from the lake.Given the long distance, binoculars and cameras with powerful zoom lenses were necessary.Thermal images were collected with a handheld thermal camera during field visits as well, providing higher resolution images of lake surface temperatures compared to the continuous, stationary thermal camera described below.The field visits used a FLIR Systems T1020 thermal camera with an image size of 1,024 × 760 pixels, using a lens with a horizontal field of view of 12°.
Continuous observations of the lake were accomplished with both visible and thermal webcams positioned on the west rim of the crater, approximately 1.2 km from the lake surface.The visible webcam had 4K resolution with a wide field of view (112°horizontal), which covered the entirety of Halemaʻumaʻu, with images captured every 10 min.The thermal camera was a FLIR Systems A655sc, with a resolution of 640 × 480 pixels and a horizontal field of view of 45°, with images captured every 2-6 min.This thermal camera is an upgrade to the previous models used on Kīlauea (Patrick et al., 2014), but used the same power and telemetry system and a similar acquisition routine (Patrick et al., 2022).Three uncrewed aircraft system (UAS, or drone) missions provided opportunities for close-up visible and thermal imagery of the lake surface.During water sampling and gas measurement UAS flights (described below), the aircraft was equipped with a DJI Zenmuse XT2 camera recording both thermal and 4K visible videos.
Lake color was measured quantitatively on a routine basis during field visits to the western caldera rim beginning in June of 2020 using a Konica Minolta CS-160 colorimeter (Figure 3; Patrick et al., 2023).The colorimeter has a measurement angle of 1/3°, allowing the user to target precise portions of the lake surface that displayed different colors of water.A smaller subset of data was obtained in mid-January of 2020 using a Konica Minolta CS-100 (1°a ngular field of view).Raw colorimetric data were plotted on a CIE 1931 color space chromaticity diagram (Smith & Guild, 1931) such that the dominant wavelength (color) of each data point could be extracted.Results were used to track temporal changes possibly related to changes in water chemistry or subaqueous processes (e.g., Ohsawa et al., 2010).

Lake Growth
Lake depth and volume were tracked by integrating measurements from a Safran Vectronix Vector23 laser rangefinder with pre-lake digital elevation models (DEMs).The laser rangefinder site location and elevation were determined by a Trimble Geo 7X kinematic GPS unit with both horizontal and vertical accuracies of 0.2 m (Patrick, Swanson, et al., 2021).Field crews measured vertical distances to various points on the lake shoreline at least once per week (Figure 3); with the points placed on a lidar-based DEM generated following the 2018 caldera collapse (Mosbrucker et al., 2020), the distances allowed for determination of lake depth, lake surface area, lake volume, and lake growth rate (Patrick, Swanson, et al., 2021).
Precipitation data from a calibrated rain gauge run by the National Park Service and located near the Hawaiʻi Volcanoes National Park Kīlauea Visitor Center were obtained from Air Resource Specialists, Inc (https://ardrequest.air-resource.com/).The gauge's location to the northeast of Kaluapele (Kīlauea's caldera; Figure 1) typically receives ∼75% more rainfall than farther west within the caldera, where the lake was located ( Giambelluca et al., 2013), so we consider the rain gauge as providing a maximum constraint on direct rainfall input to the lake.
Lake surface area, lake surface temperature (from FLIR imagery and in situ thermistor), and wind speed above the lake surface (estimated via FLIR observations of steam cloud motion at the lake surface) were used to determine evaporative loss from the lake through time via methods outlined in Hurwitz et al. (2012) and Pasternack and Varekamp (1997).

Water Sampling and Chemical and Isotopic Analyses
Given the relative inaccessibility of the lake and potential hazards associated with flying an on-board crew into the steep-walled crater, we obtained water samples via UAS (Nadeau et al., 2020) from a launch site on the western rim of Halemaʻumaʻu (19.403 N,155.293W; ∼1,140 m a.s.l.) using methods similar to those outlined in Terada et al. (2018).For each campaign, we used a DJI Matrice 600 Pro Government Edition hexacopter that had been outfitted with a customized slingload mechanism (Figure 3; Peek et al., 2023).To this slingload mechanism, we attached 10 m of polypropylene cord, at the bottom of which we affixed a weighted 1 L high-density polyethylene HydraSleeve (with Drone Cone attachment; https://www.hydrasleeve.com/).Small waterproof thermistors (Comark PDQ400) were attached to the cord to record the maximum temperature encountered at each depth.Before each flight, the sampling components were decontaminated with 70% isopropyl alcohol spray.A new water sample sleeve was used for each collection.A plastic tarp was spread over the loose tephra of the processing area to maintain cleanliness.Biological study of the water samples was not pursued as the UAS payload capacity was insufficient for such investigations.
The UAS pilot maintained situational awareness during flight via live video feed from the dual thermal/visible camera (DJI Zenmuse XT2) mounted on the UAS.Multiple visual observers outfitted with binoculars assisted with monitoring each flight.Different colors of flagging tape were tied at known spacing along the cord above the HydraSleeve to aid in assessment of when the sampler had been submerged and how far the aircraft was from the surface of the water.Once the sampler was beneath the water surface at the intended depth (up to 5 m), the pilot initiated a quick increase in aircraft altitude to ensure adequate opening and filling of the HydraSleeve; slower ascent often resulted in only partial filling of the sampler (Peek et al., 2023).See also Appendix A.
We obtained eight water samples in total over three sampling missions on 26 October 2019 (HM19-01), 17 January 2020 (HM20-01A,B,C), and 26 October 2020 (HM20-02A,B,C,D).The 2019 sampling effort consisted of a single dip of the sampler into the lake, which yielded a sample volume of ∼750 mL.Both 2020 campaigns comprised multiple sampling flights; in January, three ∼750 mL samples were obtained, and in October, four samples of up to ∼750 mL were collected.When multiple samples were obtained as part of the same campaign, we attempted to retrieve water from differently colored regions of the lake.
During each sampling campaign, temperature, specific conductance, and pH measurements of the sampled water were made using a calibrated hand-held meter (ExStik II) immediately following the return of the UAS to the launch site.Samples for major element chemistry were collected in two pre-rinsed 60-ml high-density polyethylene bottles and filtered in the field with a 0.45-μm syringe filter.One bottle from each sample was acidified in the field with nitric acid to a pH of ∼2 for cation analysis.For the first sample (HM19-01), the entire sample was filtered in the field, but upon arrival at the laboratory, additional precipitate had formed, which was vacuum filtered (0.45 μm) and analyzed separately.For subsequent samples, an unfiltered fraction was sent to the lab and vacuum-filtered at room temperature.Yellow outlines are at 2-month intervals from 1 January 2020-1 November 2020.Cyan outline is 20 December 2020, the last date at which the lake was present.Note that ledge in foreground obscures near (west) end of lake from view; dashed cyan line approximates obscured lake shoreline.
Details of the chemical and isotopic analyses are provided in Peek et al. (2023), and here we provide a short summary of the methods.Because of high solute concentrations, samples were diluted by 100 to 1,000 times to ensure in-range analysis for all solutes.Concentrations of major cations and Si (reported as SiO 2 ) were measured using inductively coupled plasma optical emission spectroscopy (ICP-OES; Thermo Scientific iCAP 6000) and concentrations of trace elements were measured using inductively coupled plasma mass spectrometry (ICP-MS; NExION 300Q, Perkin-Elmer).Analytical uncertainties are ≤5% for major elements.Anion concentrations were measured with a Dionex ICS-2000 ion chromatograph, with analytical uncertainties of <3%.All analyses were carried out at USGS laboratories in Menlo Park, California.Because the samples were diluted, the analytical uncertainty is higher for some elements with relatively low concentrations, and some elements had concentrations that were below the detection limits.
Water samples were analyzed for the stable isotopes δ 18 O, δD, and δ 34 S (of dissolved sulfate) at the USGS Reston Stable Isotope Laboratory (RSIL) in Reston, Virginia.Aliquots for δ 18 O and δD (reference standard VSMOW) analysis were stored separately in 30 mL glass bottles.Analysis of oxygen isotopes was conducted according to the methods of Epstein and Mayeda (1953) with a precision of ±0.1‰, and the hydrogen isotopes were analyzed according to the method described in Kendall and Coplen (1985) and Coplen et al. (1991) with a precision of ±1.5‰.The δ 34 S was determined for samples from each of the three lake sampling campaigns as well as for groundwater samples from a National Science Foundation (NSF)-funded well (colloquially "Keller Well"; referred to here as "NSF well"), located ∼1 km to the south of Halemaʻumaʻu (Figure 1; Keller et al., 1979) collected in February 2010 and January 2020 (Peek et al., 2023).All δ 34 S analyses (reference standard VCDT) involved precipitation as BaSO 4 followed by conversion to SO 2 with an elemental analyzer, and subsequent analysis with a continuous flow isotope ratio mass spectrometer (Brenna et al., 1997) following methods described in Révész et al. (2012), with estimated analytical uncertainties of ±0.4 ‰ (2σ).
Eight days after collection of the first sample (HM19-01), it was analyzed for SO 3 2 by titration with an iodideiodate titrant in an acid solution with a starch indicator ("Ripper" chemistry, ASTM D 1339-84, Sulfite Ion in Water, Test Method C).Subsequent attempts to titrate samples on-site immediately after collection were not successful.

Mineralogy of Solid Precipitates
Because the sampled lake waters were at near-saturation with respect to some minerals, additional precipitate formed in field-filtered samples as they cooled to lower temperatures en route to processing in the laboratory.Accordingly, such samples were filtered a second time; the solids that accumulated on the second filters were analyzed using several methods with a goal of determining their mineralogical and chemical compositions.
Scanning electron microscope (SEM) imaging with energy dispersive X-ray spectroscopy (EDS) analysis was performed on secondary precipitates (formed after filtration and subsequent cooling) from an unacidified field sample (HM19-01) using a Tescan Vega3 SEM equipped with two Oxford Instruments X-Max N 150 mm 2 EDS detectors.The mineralogy of secondary precipitates (HM19-01, HM20-01A,B,C) was determined by X-ray diffraction (XRD) on an Enraf-Nonius diffractometer equipped with a primary Germanium (111) monochromator and an INEL 120°curved position sensitive detector (PSD) at the Natural History Museum (London, United Kingdom).Data were collected from 7 to 120°2θ using Co K-alpha 1 radiation with operating conditions of 40 kV and 30 mA.Data were collected for 20 hr.Mineral identification was performed using the ICDD PDF-4+ database.Filtered solids (both secondary and original precipitates) were digested by heating overnight in a sealed Teflon vial with 300 μL concentrated HCl, 100 μL concentrated HNO 3 , and 500 μL concentrated HF.Postdigestion, samples were dried, then re-dissolved in 300 μL concentrated HNO 3 and diluted with Milli-Q water.Digest samples were analyzed for elemental content by ICP-OES using a ThermoFisher ICAP 6500 Duo.

Alteration Minerals From Kīlauea Summit
Alteration minerals collected from various sites near Kīlauea summit prior to the 2018 collapse and from blocks ejected from Halemaʻumaʻu during small lava lake explosions in 2008 and March-April 2018 were identified by XRD.Initial analyses were performed on a Bruker D8 Advance diffractometer at the University of Hawaiʻi at Hilo; repeat and additional analyses were performed at the USGS in Menlo Park, California, on a Rigaku Multiflex X-ray diffractometer with Cu K-alpha radiation.Pattern matching for mineral identification used the ICDD PDF-4 database.Complementary investigation of mineral phases was carried out on a Hitachi S3400N SEM equipped with an Oxford Instruments X-Max large area EDS detector at the University of Hawaiʻi at Hilo.
Sulfur isotope measurements of minerals were conducted by combustion in an elemental analyzer coupled to an isotope ratio mass spectrometer following the methods of Révész et al. (2012).The 2σ analytical uncertainty of the δ 34 S analyses is ±0.4‰.

Near-Lake Gas Chemistry
During the October 2019 sampling campaign, gas compositions within Halemaʻumaʻu were measured with a multi-GAS instrument affixed to a second DJI Matrice 600 Pro Government Edition hexacopter.The multi-GAS included sensors to measure water vapor, carbon dioxide (CO 2 ), and sample pressure (PP Systems SBA-5); SO 2 (City Technology, Ltd.T3ST/F); and hydrogen sulfide (H 2 S; City Technology, Ltd.T3H); as well as a GPS unit (Garmin GPS 18x) and a datalogger (Campbell Scientific CR300-WIFI).A small pump integrated in the multi-GAS drew in ambient atmosphere for on-board analysis with a temporal resolution of ∼1 Hz, and the entire sensor package was powered directly off the aircraft.The responses of the CO 2 , SO 2 , and H 2 S sensors were checked at ∼1,160 m a.s.l.elevation after the flights using portable gas standards (CO 2 = 1,000 ppm ± 2%, SO 2 = 10 ppm ± 10%, H 2 S = 10 ppm ± 10%).The CO 2 response was within 1% of the standard gas value and the SO 2 and H 2 S sensors were accurate within 2% and 4%, respectively, when an ideal gas pressure correction was applied to the sensor output.Two missions were flown: the first to ∼100 m above the lake surface and the second toward fumarolic areas on the crater walls above the lake.

Water Mass Balance
Over the duration of its nearly 17-month lifetime, the lake reached a maximum depth just over 50 m (Patrick, Swanson, et al., 2021).Depth increased fastest upon the initial appearance of the lake, reaching 10 m deep in the first 2 months of growth, then progressively slowed over time (Figure 2; Figure 4a).Water flow rate into the lake also increased over the entirety of the lake's existence, though the rate of increase began to slow appreciably about 3 months into lake growth.By the lake's boiling off at the start of the December 2020 eruption (Cahalan et al., 2023), the increase in flow rate had slowed further.Rainfall data, in conjunction with the lake surface area time series, show that the cumulative volume of rainfall directly into the lake was approximately 55,000 m 3 as determined using the rain gauge near the visitor center, or ∼6% of the lake's ultimate volume of ∼915,000 m 3 of water.The enhanced rainfall at the eastern end of the caldera and rain gauge site (Giambelluca et al., 2013), compared to the lake's location, implies that direct rainfall input to the lake was likely ≤5% of the total lake volume.A finer-scale examination of a major rainfall event in January 2020 underscores the fact that rainfall directly onto the lake surface was not a controlling factor in the lake's growth (Figure 4b).
Though the effects of rain were negligible regarding lake volume, evaporative loss from the water's surface was significant (Figure 4a).Examination of steam movement across the lake surface allowed for estimation of nearlake wind speed, which was found to commonly be ∼2 m/s.Thermal imagery and in situ sensors also yielded lake surface temperature, found to range from approximately 65 to 85°C (described in further detail below).For a wind speed of 2 m/s and an average water surface temperature of 70°C, using equations from Hurwitz et al. (2012), we have determined that the cumulative volume of water lost to evaporation over the lake's lifetime exceeded the final lake volume by more than 80%.With more conservative estimates of 1 m/s wind and a water temperature of 65°C, evaporative loss was of a similar magnitude to lake volume through much of the lake's existence, but still exceeded the final lake volume by ∼10%.Use of greater wind speed and lake temperature values in evaporation calculations increases the disparity between lake growth and evaporative loss.Accordingly, evaporative loss was a large part of the lake's mass balance, equivalent to over 50% of the net influx into the lake, and possibly significantly more.

Lake Color, Dynamics, and Thermal Properties
Select photos exhibiting the evolution of the lake's color are shown in Figure 5a and in Appendix A. When it first appeared, the water in the lake was relatively clear and tinged a green to greenish-blue hue.As the lake grew and deepened, its color and turbidity evolved; within the first 2 weeks, the water began to appear more distinctly green and cloudy, particularly in the central portion.The lake edge remained green to greenish-blue ; green dots).The cumulative evaporative (pink lines) and discharge volumes (black lines) were determined by polynomial fitting of lake volume and evaporated volume per day followed by integration of the polynomial curves.Discharge into lake was calculated by adding lake volume to evaporated volume, subtracting the volume of rain that fell directly onto the lake surface.Gray field brackets on lake depth and lake volume span the range from minimum to maximum estimates (Patrick, Swanson, et al., 2021).Vertical gray lines denote water sample dates.(b) Lake volume (m 3 ; black line with green dots) and daily direct rainfall input volume (m 3 ; light blue line) for 3 months bracketing largest rainfall event during the lifetime of the water lake.Raw data from Patrick, Swanson, et al. (2021) and NPS rain gauge.and clear.Yellow hues became more apparent in the center of the lake by the middle of August 2019, and by mid-September 2019, the lake was more uniformly yellow-green.October 2019 saw the emergence of orangebrown tones among the yellows and greens.By early spring of 2020, green colors no longer dominated the lake.After ∼9 months (mid-late April 2020), the lake was largely orange-tan through darker reddish-brown, though clear, green to greenish-blue water remained a transient presence at select portions of the lake margins throughout the evolution of colors.
Lake color was systematically monitored over time using webcam imagery; a profile along the E-W axis of the lake was extracted from daytime images, and profiles from a series of images were stacked to highlight temporal changes (Figure 5b).Proportions of lighter and darker brown varied with time; green hues also appeared frequently at the lake margins (Appendix A).These temporal patterns were compared to precipitation records, seismicity, ground tilt, air temperature, humidity, and barometric pressure, but no correlation was obvious.
Colorimetry data (Patrick et al., 2023) were not obtained regularly until June 2020, after the lake had transitioned to predominantly orange and brown tones with dominant wavelengths most typically in the 575-590 nm range (Figure 6); where measurements of green-appearing influx at the edges of the lake were possible, they exhibited dominant wavelengths ranging from ∼530 to 565 nm.Two test data sets obtained in January 2020, when the lake appearance had a mix of greens, yellows, and oranges, plotted in an intermediate range of ∼565-580 nm, though some of the difference may be due to the use of an older colorimeter.
Groundwater in the summit region of Kīlauea is known to be warm; the bottom of the NSF well near the summit has been measured to be in the range of 80-90°C in recent years (Hurwitz et al., 2019).Water was warm in the lake as well; surface temperatures ranged from 65 to 85°C, based on thermal camera measurements.Thermistor measurements made during water sampling yielded temperatures on the lower end of the range (∼65°C).Thermal imagery confirmed that the greener areas at lake margins were warmer influx zones, with surface temperatures of ∼80-85°C.The same zones remained hot over time, though a green hue was not always obvious.Three main water influx zones existed (east, north, and west lake margins); color boundaries often indicated a confluence of water flow from different influx zones and exhibited temperature steps on the order of a few degrees (Figure 7).In September 2020, additional imagery revealed the existence of small areas of cooler, brown upwelling in a few locations along the north margin of the lake (Appendix link A21).Bubbling may have been present in association with the upwelling, perhaps related to a recently submerged fumarole field along the northwest portion of the lake, but the small size of the upwelling features and distance to the lake surface prevented collection of imagery with resolution sufficient to confirm the presence of bubbles.Patches of cooler, dark brown upwelling were also noted sporadically where rockfalls impacted the lake surface.
Lake color and surface temperature were clearly linked, though the relationship varied in time and space.Within each groundwater entry zone into the lake, the temperatures showed systematic cooling trends with distance from the shoreline that related to gradual changes in color; for example, in the western influx zone, a shift from dark brown to lighter brown represented a decrease in surface temperature.The darker browns were present along the shoreline where new influx was hottest (in contrast to the aforementioned cooler, dark brown upwelling elsewhere), and the water cooled with distance from the shoreline, becoming lighter in color.In the northern influx zone, the hottest water along the shoreline had a gray hue, with the water cooling with distance and transitioning to tan.In the eastern influx zone, the hottest water along the shoreline, at the area of influx, was green and cooled with distance as it became more tan-colored.

Chemical and Isotopic Composition of the Lake Water
The chemical and isotopic compositions of the eight water samples are presented in Table 1 and in Figures 8-10.The salinity, often defined and referred to in this text as total dissolved solids (TDS), ranged from 71,000 mg/L in October 2019 (about twice seawater concentration) to 55,000 mg/L in October 2020, amounting to a 23% dilution since the first sample was collected.Despite the significant dilution (x100-1,000) of samples that was necessary for laboratory analysis, the percentage charge imbalance (([(sum cations-sum anions)/(sum cations + sum anions)/2]•100), where the sum of ions is in milliequivalents per liter) for all samples was less than 4% (Table 1).
The pH of all samples ranged from 3.8 to 4.5 with no trend through time.Despite collection from differently colored patches of the lake surface, there were no significant chemical differences between the three samples collected in January 2020 or between the four samples collected in October 2020.
Sulfate (SO 4 2 ) was the major dissolved anion in all samples, followed by Cl and F , and the major dissolved cations (by mass and molar) were Mg 2+ >Na + >K + >Fe 2+,3+ >Ca 2+ >Mn 2+ >Li + (Table 1).We note that it is possible that some colloidal iron particles may have been able to pass through the 0.45 μm filter such that not all of the iron in the filtered samples was necessarily dissolved (Nordstrom, 2011).The one sample analyzed successfully for sulfite, HM19-01, contained only 20 mg/L sulfite (SO 3 2 ), less than 0.04% of the sulfate concentration.The concentrations of most dissolved species in the October 2020 samples were ∼20-30% more dilute compared with the October 2019 sample.Notable exceptions to that pattern of dilution included As, B, Ca, F, Fe, and SiO 2 ; As decreased by ∼70%, B decreased by just ∼10%, Fe(T) decreased by ∼35%, F and SiO 2 decreased by only ∼15%, and Ca 2+ remained nearly constant (Table 1).The cation composition of the lake water significantly differed from the composition of Kīlauea basalt (Figure 8), particularly in that the water was strongly depleted in Ca 2+ compared with typical basalt.The lake water was relatively enriched in Mg 2+ compared with groundwater from the nearby NSF well, where the water is more enriched in Na + and K + (Figure 8; Hurwitz et al., 2019).Oxygen (δ 18 O) and hydrogen (δD) stable isotope values ranged between 1.18‰ and 0.81‰ and between 9.40‰ and 6.86‰, respectively (Figure 9).The δD values became isotopically lighter with time, but the δ 18 O values from October 2019 and October 2020 were constant within analytical uncertainty.The δ 18 O and δD values plotted to the right of the meteoric water line and were isotopically heavier compared with other waters from the Kīlauea region (Figure 9).
Stable isotope values of sulfur (δ 34 S) in aqueous sulfate ranged from 4.28 to 4.42‰ (Table 1) and were all within analytical uncertainty (Figure 10).δ 34 S values of NSF well water samples from 2010 to 2020 are 6.9‰ (n = 1) and 6.4‰ (n = 2), respectively, likely indicating minimal sulfur isotopic variation of the well waters over 10 years.

Mineralogy and Chemistry of Filtered Solids
SEM imaging of the solid precipitate removed from the filter of sample HM19-01 revealed homogenous material with a maximum grain diameter of ∼500 nm and a mode of ∼200 nm.Elemental composition by EDS analysis (n = 15) included S, P, Fe, Mg, Na, and K.
Alteration samples dominated by sulfate mineralogies had a relatively narrow δ 34 S range of 1.3‰ to +3.4‰ (n = 8).Native sulfur crystals from material ejected during explosions from the Halemaʻumaʻu lava lake in March and April of 2018 were isotopically lighter, with a δ 34 S range of 14.5‰ to 12.4‰ (n = 2).Native sulfur crystals collected in May and June of 2018 from degassing fissures in the lower East Rift Zone (Figures 9 and 10) had a similar δ 34 S range from 14.4‰ to 8.9‰ (n = 7).

Gas Chemistry
The bulk chemistry of the volcanic gases did not vary appreciably between the two UAS flights, both conducted on 26 October 2019.The first UAS gas flight descended ∼510 m from its launch point (∼1,140 m a.s.l.) to ∼630 m a. s.l.near the center of the crater to assess gas emissions from the lake surface.Coherent above-background H 2 O, CO 2 , and SO 2 anomalies were detected, with the highest levels found closest to the crater bottom (maximum ΔH 2 O = 760 ppmv, ΔCO 2 = 40 ppmv, ΔSO 2 = 0.3 ppmv, where "Δ" Figure 8. Ternary molar diagrams of major cations in the 2019-2020 Kīlauea water lake, NSF well waters (Hurwitz et al., 2019), and fresh, unaltered Kīlauea summit whole rock basalts (Wright, 1971).
Figure 9. δD versus δ 18 O for various waters in Kīlauea summit region, including lake water samples (triangles) and high temperature (∼300°C) fumaroles (Hinkley et al., 1995;diamonds).NSF well data from Hurwitz et al. (2019).Rain data from Scholl et al. (1995).Solid black line is the global meteoric water line (GMWL; Craig, 1961) and gray lines are evaporation of water with an initial δD = 30 and relative humidity (R.H.) of 0.7 and 0.8 above the lake-air interface following the formulation in Gonfiantini (1986).
The second flight focused on emissions from sulfur-encrusted fumaroles along the northern crater wall.The 12-min flight descended ∼240 m into the crater (∼900 m altitude) and encountered heterogenous gas plumes with variable amounts of H 2 O, CO 2 , and SO 2 (maximum ΔH 2 O = 1,000 ppmv, ΔCO 2 = 37 ppmv, ΔSO 2 = 0.3 ppmv).Despite the variable compositions encountered, the gases were overall of similar H 2 Orich and SO 2 -poor character to those measured during the first flight, with H 2 O/CO 2 ratios from approximately 26-49 (n > 39, r 2 > 0.60) and CO 2 /SO 2 = 66 (n = 153, r 2 = 0.64).One coherent H 2 O-CO 2 -SO 2 plume had a bulk composition of 97.64% H 2 O, 2.32% CO 2 , and 0.04% SO 2 (H 2 S was not detected); other plumes contained H 2 O and CO 2 only.Concentrations of gases, particularly SO 2 , would likely have been much higher very close to the sulfur-precipitating fumaroles, but UAS battery life and safety of the aircraft prevented such a proximal approach.

Thermodynamic Modeling
To better understand the processes controlling the chemical composition of the Halemaʻumaʻu water lake, we used the USGS geochemical speciation code PHREEQC (Parkhurst & Appelo, 1999) and the WATEQ4F v.2 thermodynamic database (Ball & Nordstrom, 1991) with an expanded form of the Debye-Hückel equation.We determined saturation indices (SI) for a range of minerals, where SI = 0 indicates equilibrium, SI > 0 indicates solution supersaturation with respect to the mineral, and SI < 0 indicates undersaturation.Factors such as reaction kinetics, which can also influence whether a mineral dissolves or precipitates, are not considered in the calculations.
As inputs, we used the chemical composition of the first collected sample from the lake in October 2019 (HM19-01) and of sample HM20-02A from the last sampling campaign in October 2020 (Table 1).We calculated the saturation indices at both 70°C and 80°C and, because we were unable to measure dissolved oxygen concentrations immediately following sampling to constrain Fe-oxidation, we also performed the calculations with either 50% Fe 3+ or 100% Fe 3+ .Given the moderately acidic lake water pH, total inorganic carbon and alkalinity were not included in the calculations.Because most specific minerals identified among filtered solids (Section 3.4) are not well-represented in the thermodynamic databases, we relied on descriptions of alteration minerals from Kīlauea, as in Section 3.5 and in previous studies (Keller et al., 1979;Macdonald, 1944;McCanta et al., 2014;Stone & Fan, 1978), as constraints.

Discussion
The 2019-2020 Kīlauea summit water lake was unique in recent history at Kīlauea and was also chemically unique among volcanic lakes globally.That novelty, as well as the increased potential for phreatic/phreatomagmatic explosion hazards, led to an intense focus on the evolving lake and implementation of many different monitoring methods to track that evolution.Here we synthesize our varied measurements and observations to discuss the chemistry of the lake, including the sources of the water and sulfur; the reactions governing the hydrogeochemistry of the lake system; the relationship of the lake to potential eruption hazards; and possibilities for future crater lake studies.

Lake Dynamics and Chemical Evolution
The 2018 Kīlauea caldera collapse significantly altered the summit groundwater system and the summit degassing regime.Ahead of the collapse, during the 2008-2018 summit lava lake eruption (Patrick, Orr, et al., 2021), groundwater was separated from the magma column by a steam zone; the rock nearest the magmatic conduit and lava lake was heated such that liquid water could not exist in that zone (Hsieh & Ingebritsen, 2019).During the same timeframe, SO 2 and other gases rose within the magma column and degassed freely from the surface of the lava lake, yielding emission rates averaging ∼5,000 t/d over the duration of the decade-long summit eruption (Elias et al., 2018).The 62 collapse events in 2018 resulted in ∼500 m of total subsidence across a collapse area of ∼5 km 2 (Anderson et al., 2019;Neal et al., 2019).The previously open lava lake conduit drained and eventually was filled with collapse debris.Concurrently, SO 2 emission rates dropped from 5,000-10,000 t/d SO 2 in May 2018 to 1,000-4,000 t/d in June 2018, and were near instrument detection limits of <50 t/d by late September 2018 (Kern et al., 2020;Nadeau et al., 2023) as magma withdrew to greater depths and was repressurized to the point at which SO 2 is still highly soluble in basaltic melt (Gerlach, 1986;Lerner et al., 2021).Subsidence within the summit caldera also lowered the groundwater base level and thus increased the hydraulic gradient toward the newly deepened crater.Despite this heightened gradient, the collapsed material was initially still hot enough to continue preventing flow of liquid water into the crater (Hsieh & Ingebritsen, 2019).Only once the former magma conduit rocks had  (Parkhurst & Appelo, 1999).Calculations were carried out with water temperature at 70°C (dark colors) and 80°C (light colors) and at 100% Fe 3+ (circles) and 50% Fe 3+ (diamonds).Minerals with positive SI are expected to precipitate from the water.
sufficiently cooled to allow for the presence of liquid water did the steam zone disappear and the local water table likely become mostly contiguous, if still depressed.As groundwater progressively flowed toward the crater to reduce the hydraulic gradient around the collapse area, the water table eventually intersected the bottom of the deepened crater and created the incipient lake.
It had been previously established that groundwater near Kīlauea summit is recharged locally (Scholl et al., 1996) and is affected by the magmatic system; NSF well water chemistry changed ahead of the 2008 summit eruption as a result of scrubbing of magmatic volatiles (Hurwitz & Anderson, 2019) and water level in the well decreased in association with a magmatic intrusion in 2001 (Hurwitz & Johnston, 2003).However, a crater lake within Kaluapele was unprecedented in written historical times.Accordingly, monitoring the chemistry of the newly formed lake became a priority, both to determine the composition of the unique water body and to assess whether the groundwater or lake water was scrubbing magmatic gases.The low SO 2 emission rates after the 2018 activity were initially attributed to a lack of shallow magmatic degassing, but the presence of the water lake raised the possibility that scrubbing could already be masking higher gas emissions than the continued low SO 2 emission rate measurements would suggest, or that it would later mask increasing emissions upon an eventual renewed magma ascent beneath the crater.Increased SO 2 emissions could also have been masked owing to subsurface cooling and gas-phase or gas-rock reactions, resulting in conversion of magmatic SO 2 to H 2 S and/or the precipitation of sulfides or sulfates (e.g., Hedenquist & Lowenstern, 1994;Heinrich, 2007;Henley & Fischer, 2021).
Water sampling via UAS in October 2019, January 2020, and October 2020 allowed for the following key observations regarding lake chemistry (see also Table 1): (a) the first sample (HM19-01), collected approximately 3 months after the lake was first observed, had a TDS of 71,000 mg/L with high concentrations of sulfate (53,000 mg/L), chloride (630 mg/L), iron (527 mg/L), and magnesium (10,400 mg/L); (b) water pH did not systematically change over time and had a fairly narrow range of values (3.8-4.5);(c) the TDS and the concentrations of most solutes, including SO 4 2 , decreased between October 2019 (first sample) and October 2020 (last collected samples) by 20%-28%; (d) solutes that deviate from this dilution trend are As, B, Ca, F, Fe, and SiO 2 , suggesting that they were reactive; (e) dissolved Ca 2+ concentrations remained nearly constant while the concentrations of Fe and As decreased more than other solutes (∼35% and ∼70% dilution, respectively); (f) despite the overall dilution trend, the total mass of dissolved solutes increased significantly as the lake grew in volume; (g) the values of δ 18 O and δD were shifted from the global meteoric water line (GMWL) due to significant evaporation; and (h) the δ 34 S signature of the lake water was essentially constant across all sampling campaigns (4.3-4.4‰,n = 8).
Additional important observations include: (a) the lake formed within a significant talus pile from the 2018 caldera collapse, which presumably consisted of highly fractured altered and unaltered basaltic lava; (b) the steady rate of water inflow into the lake and the insensitivity of the lake's water level to local rainfall events (Figure 4b) indicate that the lake was dominantly fed by groundwater; (c) sulfur-bearing alteration minerals from surface samples around Halemaʻumaʻu (pre-collapse) were dominantly sulfates and native sulfur, with no observed sulfides; (c) sulfur-precipitating fumaroles up to ∼200°C (as measured via thermal camera) were visible on the walls of the collapse crater, indicating localized degassing; (d) PHREEQC calculations indicate that hematite, jarosite, and goethite, are supersaturated, and amorphous iron oxyhydroxide, amorphous silica, anhydrite and gypsum are at near-saturation ( 1 < SI < 1); (e) SO 2 emission rates during the lake's existence (<50 t/d) remained two orders of magnitude lower than the emission rates from the lava lake present in Halemaʻumaʻu (averaging ∼5,000 t/d) prior to the 2018 collapse (Elias et al., 2018(Elias et al., , 2020;;Kern et al., 2020); and (f) lake water at the time and depth of sampling (up to 5 m below lake surface) was ∼65°C (Table 1) as measured by in situ thermistors, which is slightly lower than indicated by thermal cameras (∼65-85°C).

The Source of Water Into, and Evaporation From, the Lake
The isotopic compositions (δ 18 O and δD) of all lake water samples are within analytical error and plot to the right of the GMWL (Figure 9).In comparison, the δ 18 O and δD composition of local precipitation varies across the caldera, but values fall generally along the GMWL, with δD values ranging from 38‰ on the south side to 22‰ on the east side, and 30‰ on the north and west sides of the caldera (Scholl et al., 1996).If we consider an average δD value of 30‰ for local precipitation and a lake water temperature of 70°C, then following Gonfiantini (1986), the evaporation lines that bracket the lake samples correspond to a relative humidity between 0.7 and 0.8 (Figure 9); this is consistent with relative humidity generally between 0.6 and 0.95 near Kīlauea summit as measured by permanent HVO meteorological instrumentation.These calculated evaporation trends do not consider the effects of high salinity (Sofer & Gat, 1975), but the magnitude of such an effect is within the large uncertainties of the calculated lines (Bowen et al., 2018).There is no indication of magmatic water input, which is consistent with observations from earlier high temperature (∼300°C) and near-boiling fumaroles near the summit crater that emitted steam comprising mostly meteoric water (Hinkley et al., 1995).
Accordingly, the Kīlauea lake's stable isotopic composition is consistent with significant evaporation (Figures 4a  and 9) of locally recharged meteoric water (Scholl et al., 1996).We note that this significant evaporation from the lake is in contrast with inferences made in a previous modeling study where it was asserted that, based on panevaporation rates of approximately 1.5 m/year, evaporation was likely a secondary factor in the crater-lake water mass balance (Flinders et al., 2022).Using calculated evaporation volumes in conjunction with lake surface area (Patrick, Swanson, et al., 2021), we find daily evaporation depth equivalents of ∼0.1 m/day, or ∼36.5 m/year, throughout much of the lake's lifetime.Our observations suggest that evaporation rates from the lake surface were substantially larger than typical pan-evaporation rates (likely owing to high water temperatures), implying that the water inflow rates calculated by Flinders et al. (2022) are significantly underestimated.
Though evaporation of meteoric water likely dictated the δ 18 O and δD signature of the lake, we also explored possible oxygen exchange between SO 4 2 and H 2 O that would deplete the H 2 O in 18 O (Chiba & Sakai, 1985).
Such an oxygen isotope shift requires extremely high concentrations of SO 4 2 that may have only been present in the very early days of the lake's presence, when the lake volume was very small (Figure 4a).Further, the rates of oxygen-isotope exchange between dissolved SO 4 2 and H 2 O are extremely sluggish for the lake's measured temperature and pH (exchange half-life on the order of 100s-∼1,000 years; Chiba & Sakai, 1985;Rye et al., 1992;Seal et al., 2000), such that significant oxygen exchange over the lake's 17-month existence is not likely to have occurred.

Acid-Base and Redox Reactions in the Lake
A global compilation of volcanic lake waters reveals a bimodal pH distribution with numerous acidic (pH 0.5-1.5)and near-neutral (pH 6-6.5) lakes, and very few volcanic lakes within a pH range of 3.5-5 (Marini et al., 2003).Bimodal pH distributions were also found in acid mine drainages (Bigham & Nordstrom, 2000;Cravotta et al., 1999) and in the hot springs of Yellowstone National Park (Nordstrom et al., 2009), though in both instances, the more acidic of the two modes has a somewhat higher pH (2-4 and 1-5, respectively) than the acidic mode of the volcanic lake pH distribution.Reaction path and speciation modeling show that, at a temperature of 65°C, pH buffers exist both in the acidic (HSO 4 /SO 4 2 at pH = 2.5) and neutral (H 2 CO 3 /HCO 3 at pH = 6.3) regions, but that there are no buffers in the intermediate pH region of 3.5-5 (Marini et al., 2003).The pH values of the Kīlauea lake do not fall near either mode of the bimodal distribution of volcanic lakes worldwide, yet SO 4 2 concentrations are comparable to those measured in the hyper-acidic crater lakes of Kawah Ijen, Indonesia, and Poás, Costa Rica (∼77,000 mg/L and up to ∼167,000 mg/L, respectively; Delmelle & Bernard, 2015), and higher than those measured at other acidic lakes, including Ruapehu's (New Zealand) crater lake (15,250-21,400 mg/l; Christenson & Wood, 1993).This implies that unique processes were taking place to produce the mildly acidic, sulfate-rich Kīlauea lake waters.
Based on abundant studies on acid mine drainage (e.g., Bigham & Nordstrom, 2000;Nordstrom, 2011Nordstrom, , 2020)), we suggest that iron oxidation had an important influence on the pH and composition of the Kīlauea lake waters.Iron could have been sourced either from high temperature leaching of basalt (Hurwitz et al., 2003), or from dissolution of Fe-sulfide and -sulfate minerals previously deposited on the margins of the lava lake prior to the 2018 collapse.As the rate of the oxidation would likely have depended on a variety of factors (e.g., temperature, pH, physical state, particle size; e.g., González-Santana et al., 2021), we have limited data to constrain the rate at which iron was oxidized.However, if oxidation reactions were the primary control on water pH, then at least some of the iron that was initially in the lake must have been in the reduced form (Fe 2+ ).Unfortunately, we were unable to measure the ratio of Fe 2+ /Fe 3+ or make electrometric measurements of redox potential following sampling.
Though we lack direct chemical measurements confirming high concentrations of reduced iron, a long-term presence of Fe 2+ in the lake is likely, based on both lake pH and water color.High concentrations of dissolved iron (as opposed to colloidal iron) at pH > 4 suggest that it was likely dominated by reduced iron (Fe 2+ ); high concentrations of Fe 3+ (e.g., >10 mg/L) at pH > 4 are not possible unless the concentration of dissolved organic carbon is high and has strong complexing capacity for Fe 3+ (D.Kirk Nordstrom, U.S. Geological Survey, written communication, 2021).Also, PHREEQC geochemical speciation calculations show that Fe 3+ would primarily exist in the hydrolyzed form (Fe(OH) 2+ ) and would readily precipitate as amorphous Fe(OH) 3 .Onda et al. (2003) and Ohsawa et al. (2010) showed that relevant chemical species can contribute to crater lake color at different wavelengths: colloidal sulfur produces Rayleigh scattering, which causes a blue coloration (∼480 nm dominant wavelength); absorption by dissolved Fe 2+ results in a greenish tinge (∼535 nm); and absorption by dissolved Fe 3+ yields dominant wavelengths in the yellow range (∼580 nm).Kīlauea's crater lake initially appeared bluish-green to green (Figure 5), which attests to the early dominance of Fe 2+ over any Fe 3+ present.Though yellow tones indicative of an increasing ferric iron proportion became apparent only ∼2 weeks into the lake's evolution, clear greenish influx zones with dominant wavelengths in the range of ∼530-560 nm, presumably dominated by reduced iron, at the lake edges persisted as an ephemeral presence throughout the lake's lifetime (Figure 6).Accordingly, we assume that the proximal groundwater feeding the lake was consistently rich in Fe 2+ and continued to supply Fe 2+ into the lake through the duration of its lifetime.
Regarding the potential source(s) of Fe 2+ , alteration minerals collected from locations around the rim of Hale-maʻumaʻu prior to the 2018 eruption (Section 3.5 and McCanta et al., 2014) provide context for some of the minerals that could have been potentially dissolved into the water.The assemblages predominantly comprise sulfate minerals (gypsum, anhydrite, jarosite, natrojarosite, pickeringite, alunite), native sulfur, amorphous silica (opal/opal-CT), and minor titanium dioxides.None of the minerals identified were sulfides, although this could be the result of sampling only at the ground surface where minerals are oxidized.Indeed, halides were also not identified, but have been found previously, generally where reducing conditions exist, such as in fumaroles and lava lake drill holes at Kīlauea (Naughton et al., 1976).
We propose that the Fe 2+ was most likely sourced largely from basalt, and generally not from sulfide or sulfate minerals, for several reasons: (a) the large talus pile from the 2018 caldera collapse consisted of highly fractured basaltic lava; (b) iron in subaerial Kīlauea basalt is mostly reduced (Carmichael & Ghiorso, 1986); (c) there are no observations of sulfide alteration minerals in the summit area; (d) the concentrations of Mg and Mn in the lake water are high, and Mg-or Mn-bearing sulfides are relatively rare and have not been observed as alteration minerals anywhere at Kīlauea; (e) the δ 34 S of low-temperature sulfide minerals (typically 50‰ to 20‰; Rye, 2005;Seal, 2006) is inconsistent with the sulfur isotope composition (Figure 10) and moderate pH conditions of the lake water (discussed below).We note that pickeringite was among the sulfate alteration minerals identified; it is not iron-bearing, but related members of the halotrichite group contain Fe 2+ (Ballirano, 2006).If any such minerals were present in association with the identified pickeringite, they could have contributed minor amounts of Fe 2+ in addition to the larger contribution by the basalt.
If Fe 2+ oxidation controlled Kīlauea's lake chemistry, we can also examine the effects on pH of those controlling processes.The oxidation of Fe 2+ initially increases water pH (e.g., Nordstrom et al., 2015): Fe 3+ then hydrolyzes and water pH decreases: (2) The iron then precipitates from the water as a mineral phase (goethite), causing a net decrease in water pH and leading to a metastable equilibrium: An alternative to Equations 2 and 3 incorporates sulfate to form schwertmannite (Schoepfer & Burton, 2021) Studies of natural systems and acid-mine drainage suggest that during the oxidation of Fe 2+ -bearing solutions and neutralization of acid water through biotic or abiotic pathways, ferric (Fe 3+ ) oxide minerals, such as micro-to nano-crystalline goethite, schwertmannite, and ferrihydrite (Fe(OH) 3 ) precipitate.The precipitate is usually a mixture of phases of uncertain composition and crystallinity (Nordstrom et al., 2015).Laboratory and field studies have shown that ferrihydrite tends to dominate at pH values > 5.5 but is thermodynamically unstable and acts as the predominant precursor for transformation to goethite (Nordstrom, 2011).Jarosite tends to dominate at pH 0.8-2.5 (Nordstrom et al., 2015), and schwertmannite at intermediate pH values; both of these are favored over ferrihydrite or goethite in waters with elevated sulfate concentrations.
The possible initial formation of schwertmannite in the Kīlauea lake between 14 September 2019 and 30 October 2019 (onset of orange-brown coloration; Figure 5) is consistent with observations from the summit crater lake of Mt.Shinmoe-dake, Kirishima Volcano, in Japan.There, a change in color from blue-green to brown was attributed to the presence of schwertmannite.A cultivation test from the Shinmoe-dake lake revealed that the lake hosts iron-oxidizing bacteria, which likely mediated schwertmannite formation (Ohsawa et al., 2014).
The presence of schwertmannite would also provide another means by which to decrease the lake's pH; schwertmannite can be converted to goethite at pH values of 2.5-5.5 (Bigham et al., 1996;Schoepfer & Burton, 2021), which produces additional hydrogen ions: Additionally, hematite precipitation can decrease solution pH, via either redox-(e.g., Reed & Palandri, 2006) or non-redox-sensitive reactions (e.g., Ohmoto, 2003;Zhao et al., 2019).However, it is likely to occur only at temperatures over 100°C (Babcan, 1971), which is above the upper end of the range of water temperatures measured at Kīlauea, such that it is not likely to have been an important process in the lake.
Results from PHREEQC simulations (Figure 11) suggest that hematite, jarosite, and goethite are highly supersaturated and ferric oxyhydroxide is slightly above saturation at 70 and 80°C and with 50% and 100% Fe 3+ .However, thermodynamic data on iron minerals at temperatures of 70-80°C are sparse, and studies have shown that there are significant discrepancies between calculated saturation indices and observations in natural systems for the various iron phases (e.g., Nordstrom, 2011;Nordstrom, 2020).In particular, there are no thermodynamic data in the different mineral databases used by PHREEQC for schwertmannite, which is one of the most common minerals to form via direct precipitation of iron in waters at the pH of the Kīlauea lake waters (Schoepfer & Burton, 2021).Despite this, schwertmannite was not identified by XRD of the material that accumulated on the syringe filters, and imaging of the filter sample by SEM did not reveal particles with its sometimes-distinct 'seaurchin'-like surface morphology.However, schwertmannite can be notoriously difficult to characterize, which may be particularly relevant given the poor crystallinity and very fine size (mode of ∼200 nm) of our filtered solids.As mentioned, the material analyzed was secondary precipitates and may not represent the minerals that formed in situ in the water lake, and any further investigation would have been best performed on primary lake precipitates.
There are other known problems with the calculations of iron-bearing minerals in these speciation codes.For example, the solubility of goethite in thermodynamic databases is of the fully crystalline form and, therefore, it will frequently show significant supersaturation, as in our calculations.The calculated saturation indices are further confounded by the potentially higher iron concentrations in the acidified sample than the actual concentration of dissolved iron, a possible result of some colloidal iron passing through the 0.45 μm filters used (Nordstrom, 2011).Jarosite typically forms at pH values < 2.6 in acid mine waters (Bigham et al., 1996), but it can form at higher pH values when SO 4 concentrations are high and there is sufficient K and/or Na.Jarosite does not readily precipitate from aqueous solutions at the pH of the Kīlauea water lake, but was found in altered basalt around Halemaʻumaʻu (McCanta et al., 2014; this study) and in material precipitated from acid mine waters at Iron Mountain, California (Alpers et al., 1989).Further, whereas the aqueous model implemented in PHREEQC is adequate at low ionic strengths, it may become more inaccurate at ionic strengths in the range of seawater and above (Parkhurst & Appelo, 1999).Accordingly, we do not adhere to a strict interpretation of the results of the PHREEQC modeling but rather use the results in the broader context of our study.
In addition to iron, we also considered the possible role of arsenic in the lake's evolution.Arsenic (As) concentrations substantially decreased between October 2019 and October 2020, implying As precipitation from lake waters if we assume a steady supply.Both As(V) and As(III) sorb strongly to iron oxide; however, arsenic sorption is dependent on its oxidation state and the specific type of iron oxide mineral (Dixit & Hering, 2003).At pH 4, As(V) sorption to amorphous iron oxide and goethite is more favorable than that of As(III).In the absence of H 2 S, As(III) readily oxidizes and the oxidation of As in thermal waters is also catalyzed by microbes (Cherry et al., 1979;Gihring et al., 2001;Inskeep & McDermott, 2005;Langner et al., 2001).Adsorption of As(V) by iron oxides reaches a maximum at pH 3-4, and gradually decreases with increasing pH (Hingston et al., 1971).Therefore, the mildly acidic, Fe-oxide bearing nature of the water lake provided favorable conditions for As adsorption and precipitation, likely leading to the large observed decrease in dissolved As over time.
In summary, groundwater flowing into the lake appears to have been dominated by reduced iron, as indicated by the persistent green-tinged influx zones at the edge of the lake.Other colors (e.g., brown, yellow) observed in the lake over its lifetime were indicative of oxidized iron.Iron was likely derived from leaching of the abundant surrounding basalt in the collapse talus pile.Within the lake, oxidation of the iron decreased pH from near-neutral (assumed based on the groundwater pH of ∼7.8 in the nearby NSF well; Hurwitz et al., 2003) and caused precipitation of iron oxides, potentially including hematite, jarosite, goethite, ferric oxyhydroxide, and schwertmannite, as well as sorption of arsenate.As we did not conduct biological analyses, we cannot rule out a biotic role in the oxidation (e.g., Bonnefoy & Holmes, 2012;Weber et al., 2006).

The Sources of Elevated Sulfur and Other Solutes in the Lake
Sulfate (SO 4 2 ) was the main solute dissolved in the lake water (Table 1).The major observations that are pertinent for determining the source of the sulfur are: (a) SO 4 2 concentrations decreased from 53,000 mg/L in October 2019 to 41,000 mg/L in October 2020 (a 23% decrease) while the total mass of SO 4 2 continued increasing as lake volume grew; (b) PHREEQC calculations indicate that the lake water was at, or near, saturation with respect to the Ca-sulfate minerals gypsum and anhydrite, and slightly undersaturated with respect to epsomite (MgSO 4 •7H 2 O) (Figure 11); (c) within analytical error, the δ 34 S values of the lake water (+4.3‰ to +4.4‰) did not change between October 2019 and October 2020 (Figure 10, Table 1); and (d) altered rock samples from around Halemaʻumaʻu (pre-collapse) contained sulfate minerals with δ 34 S values between 1.3‰ and +3.4‰ (Figure 10), and native sulfur from around Halemaʻumaʻu and from fissures in the lower East Rift Zone have δ 34 S values between 14.5‰ and 8.9‰ (Section 3.5).We use these constraints in conjunction with data on water composition from the nearby NSF well (Hurwitz et al., 2021) to constrain possible sources of sulfur in the lake.We consider these possible scenarios: (a) most of the sulfur in the lake was derived from scrubbing of SO 2 released from underlying magma; (b) sulfur in the lake was derived from dissolution of sulfur-bearing minerals in altered rocks in the talus pile that formed following the 2018 caldera collapse; (c) the same as in scenario b, but including additional dissolved sulfate already in groundwater flowing into the lake to produce the measured SO 4 2 concentrations and the δ 34 S composition of the lake water.
Throughout the existence of the water lake, only relatively low SO 2 emissions of <50 t/d were measured at Kīlauea summit (Nadeau et al., 2023), and sulfur-and sulfate-depositing fumaroles were visible on the walls of the collapsed Halemaʻumaʻu (Figures 2a, 3c, and 3d).Coincident to the first water sampling of the lake (October 2019), SO 2 was also directly measured within the collapsed crater by a UAS-borne multi-GAS instrument at low, but resolvable, concentrations (ΔSO 2 = 0.3 ppmv).These observations indicate that "dry" degassing pathways allowed some SO 2 gas to reach the ground surface without dissolving into groundwater.However, it is unknown whether additional, and possibly significant, SO 2 degassing was also occurring beneath the lake or into groundwater feeding the lake.
If degassing SO 2 was dissolving into the lake or groundwater, the principal gas scrubbing reactions (Iwasaki & Ozawa, 1960;Symonds et al., 2001) would have been: Geochemistry, Geophysics, Geosystems

10.1029/2023GC011154
The equilibria of these disproportionation reactions shift strongly to the right below 400°C, mainly converting SO 2 into dissolved sulfate and sulfuric acid, which would significantly acidify the water.Oxidation of H 2 S would further increase sulfate concentrations and decrease water pH (Symonds et al., 2001): Thus, if substantial amounts of SO 2 (released from shallow magma) or H 2 S (released from the hydrothermal system) were being scrubbed, lake water pH would have decreased over time.The calculated δ 34 S values of SO 2 degassing from Kīlauea magma (+1‰ to +3‰; Lerner et al., 2021;Sakai et al., 1982) are slightly lower than those of the lake water, and an input of even a few 10s of t/d of SO 2 completely dissolving into an initially sulfur-poor water (meteoric water) could reasonably produce the measured lake water δ 34 S values and SO 4 2 concentrations (Figure B1).However, this SO 2 input would have greatly acidified the lake water (Equations 6 and 7).Additionally, if SO 2 dissolved into the lake and underwent disproportionation-hydrolysis to isotopically heavy SO 4 2 and isotopically light H 2 S (Kusakabe et al., 2000) and the reduced sulfur species became physically separated from the water (e.g., by H 2 S outgassing or native sulfur precipitation), the δ 34 S values of dissolved lake water sulfate would have become notably heavier than measured.Since neither the strong isotopic effects of SO 2 disproportionation-hydrolysis nor lake acidification were observed, we rule out substantial SO 2 degassing from shallow magma into the lake.Consequently, we conclude that the low measured SO 2 emission rates during the existence of the water lake instead indicated the absence of very shallow magma in the area (Gerlach, 1986;Lerner et al., 2021) until the onset of the December 2020 eruption (Gerlach, 1986;Lerner et al., 2021).
We also rule out dissolution of native sulfur and sulfide minerals as a source of the sulfur in the lake because of their much lighter δ 34 S values (Figures 10 and B1) and their relative insolubility in relevant waters (neutral-mildly acid and <100°C; e.g., Ellis & Giggenbach, 1971).The dissolution of sufficient masses of such minerals into initially sulfur-poor waters to achieve the high SO 4 2 concentrations in the lake would have caused the lake water's δ 34 S values to be far lighter than measured (Figure B1).
A more plausible source for the sulfur in the lake is the dissolution of sulfate minerals from the underlying talus pile and from rocks surrounding the lake.However, the δ 34 S compositions of the lake water (+4.3 to +4.4‰) were slightly heavier than the compositions of sulfate alteration minerals ( 1.3‰ to +3.4‰; Figure 10).The δ 34 S fractionation between Ca-sulfate minerals and SO 4 2 is negligible at temperatures >110°C (Kusakabe & Robinson, 1977;Sakai, 1968), although some studies have found that gypsum can be +1.0 to +1.7‰ heavier than associated SO 4 2 at temperatures <80°C (Raab & Spiro, 1991;Van Driessche et al., 2016), which would result in lower δ 34 S-SO 4 2 than measured in the lake.While it is possible that all the sulfur in the lake was derived from dissolution of sulfate minerals, the slightly lighter δ 34 S values of sulfate minerals measured in our study (Section 3.5) might imply an additional sulfur source with a heavier isotopic composition, including, perhaps, sulfate generated by very minor SO 2 disproportionation.
One possible source of high δ 34 S sulfur could be rainwater; precipitation around Kīlauea's summit is likely dominated by the δ 34 S value seawater of about +21‰ (Paytan et al., 1998;Rees et al., 1978).Mixing of ∼93 wt.% sulfur derived from sulfate minerals (δ 34 S = +3‰) and ∼7 wt.% sulfur from rainwater is consistent with the observed lake water δ 34 S value of +4.4‰.However, the SO 4 2 concentration in rainfall around Kīlauea's summit is only ∼5 ppm (Scholl & Ingebritsen, 1995) and at the rate of groundwater flow into the lake (Figure 4) there would not be nearly enough sulfur to yield the high SO 4 2 concentrations in the lake.Therefore, unless unrealistically high rates of evaporation of the precipitation occurred prior to recharge, groundwater derived from precipitation would not have had a significant effect on the δ 34 S composition of lake water.
An alternative source of high δ 34 S sulfur could be groundwater within Kīlauea's caldera.The highest concentration of SO 4 2 measured in the NSF well within the caldera (Figure 1) was 7,040 mg/L in February 2010 (Hurwitz et al., 2019) with a corresponding δ 34 S value of +6.9‰.This elevated δ 34 S value in the groundwater could result from SO 2 (δ 34 S = +1‰ to +3‰) dissociation in water, which produces +6‰ to +9‰ heavier SO 4 2 and isotopically light H 2 S gas (Kusakabe et al., 2000;Taran & Kalacheva, 2020).A binary mixture sourcing ∼36 wt.% of sulfur from sulfate-rich NSF well type water with a δ 34 S value of +6.9‰ and ∼64 wt.% of sulfur from Ca-sulfate alteration minerals (anhydrite or gypsum) with a δ 34 S value of +3‰ (Figure 10) could closely match both the δ 34 S value and the SO 4 2 concentration of the lake water measured in October 2019, without shifting the lake pH (Figure B1).These calculations are described in further detail in Appendix B.
In conclusion, the range of observations suggests that sulfur in the lake was mostly derived by dissolution of sulfate minerals from the collapsed talus pile and the probable addition of isotopically heavier dissolved sulfate in groundwater feeding the lake, and possibly very minor scrubbing of SO 2 .Sulfate concentrations throughout the duration of the lake were controlled by equilibrium with a Ca-sulfate mineral, most likely gypsum or anhydrite.
Although the concentration of sulfate and most other solutes in the lake water decreased as the lake volume grew (Table 1), the overall mass of dissolved solutes continued to increase, indicating that groundwater flow into the lake continued adding solutes over time.

Relationship to Activity at Kīlauea Summit
The period following the eruption and caldera collapse in 2018, and before the 2020 eruption, represented the first prolonged non-eruptive period at Kīlauea since prior to the onset of the long-lived Puʻuʻōʻō eruption in 1983 (Heliker et al., 2003).While Kīlauea's summit exhibited continuous inflation and above-baseline (relative to pre-2018 levels) earthquake counts between late 2018 and late 2020 (U.S. Geological Survey, 2020), summit SO 2 emission rates remained near detection limits (<50 t/d; Nadeau et al., 2023).As with the unchanging gas measurements, there were ultimately no precursory visible or thermal changes in lake character that might have alerted HVO to the coming eruption on 20 December 2020 (Appendix link A23).Lake color remained the same at least until nightfall on the evening of the eruption, with visible webcam imagery looking unremarkable compared to prior images, and colorimeter measurements made in the early afternoon of the same day were also unchanged compared to recent data.Thermal imagery showed the same minor variations in water lake surface temperature as it had for the lifetime of the lake.Lake level continued to gradually increase, with the final lake level measurement from the afternoon of 20 December showing a lake rise since the previous measurement consistent with the established lake level trend (Figures 2 and 4a).
Whether chemical changes occurred within the lake in advance of the eruption is unknown; the final water samples were obtained on 26 October 2020, nearly 2 months prior to the eruption, and were chemically very similar to the previous samples obtained 9 months and 1 year earlier.However, diffusion modeling in olivine phenocrysts erupted in the initial phase of the 2020 eruption suggests that the magma that eventually erupted was likely injected to shallower levels (∼1-2 km) approximately 60 days prior to the eventual eruption, perhaps in conjunction with two summit-area seismic swarms that took place 22-24 October and 29 October 2020 (Lynn, 2021).With the 26 October 2020 water sampling mission only a few days after the first swarm, and before the second, we cannot rule out the possibility that heat or gases released from a shallow intrusion(s) might have affected lake chemistry, either directly or indirectly (via affecting the groundwater feeding the lake), in the ensuing 2 months.However, added heat or changes in chemistry would perhaps have been reflected in lake color or surface temperature, which did not change.Similarly, magma shallow enough to affect the lake water or summit groundwater chemically might be expected to enhance outgassing of SO 2 from the summit area; SO 2 emission rates measured in early December remained at the low levels (<50 t/d) that had persisted since fall of 2018.Such a lack of SO 2 outgassing increase is consistent with an intrusion depth of 1-4 km (as suggested by the occurrence of deflation-inflation events and the depth of enhanced seismicity preeruption; U.S. Geological Survey, 2020), as Kīlauea basaltic magmas only release significant proportions of their dissolved sulfur upon reaching within a few hundred meters of the surface (e.g., Dixon et al., 1995;Gerlach, 1986;Lerner et al., 2021).Accordingly, we infer that the lake's water chemistry was not affected by the October intrusion.
Perhaps the most clear-cut evidence of the water lake failing to provide precursory signals in advance of the eruption comes from thermal webcam imagery of the December 2020 eruption onset (Figure 12; Appendix link A23).The eruption's initial vent appeared on a block of collapsed caldera floor (obscuring ledge in Figure 2) perched higher than the water lake at the base of the crater.Despite magma ascending, and eventually erupting, in proximity (<200 m horizontal distance) to the lake, the water surface appeared entirely unperturbed both immediately before the eruption and even as lava had begun erupting on the higher caldera block.It was only when lava flowed from the elevated block down into the water at the base of the crater that any effect on the lake was noted (Cahalan et al., 2023).

Implications for Eruption Hazards
There are many examples of crater lakes displaying color, temperature, or chemical changes in relation to volcanic activity.At Mount Spurr (Alaska, USA), the crater lake, previously clear blue-green, turned gray over a short (<weeks) timeframe 3-4 weeks prior to the 27 June 1992 eruption (Keith et al., 1995).This color change was accompanied by enhanced bubbling and geysering, and the lake completely evaporated or drained away as of the afternoon before the eruption.Laguna Caliente at Poás Volcano (Costa Rica) also changed colors from green to gray in conjunction with increases in water temperature and dissolved SO 4 /Cl between 2006 and 2013, a period during which several phreatic eruptions occurred (Fischer et al., 2015).Such gray coloration in volcanic lakes has been attributed to increased amounts of suspended sediments during convective upwelling (Christenson & Wood, 1993).At Aso Volcano (Japan), a shift from blue to green of the Yudamari crater lake was attributed to increased subaqueous fumarole activity and an increasing proportion of in-lake degassing of SO 2 compared to H 2 S (Ohsawa et al., 2010).At a number of volcanic lakes, including at Taal (Philippines) and Kelut (Indonesia), water temperature increases have been detected prior to eruptive activity (Badrudin, 1994;Moore et al., 1966;Trunk & Bernard, 2008).In contrast, Kīlauea's lake exhibited no discernible change in advance of the December 2020 eruption.We suspect this is likely a result of the inherent differences in magma chemistry and behavior at Kīlauea compared to other volcanoes that host crater lakes, most of which are arc volcanoes.
Magmas at arc volcanoes are typically more oxidized and evolved than at Kīlauea, and have higher H 2 O contents (e.g., Plank et al., 2013;Urann et al., 2022;Wallace, 2005), all of which promotes increased degassing of sulfur from substantially greater depths than at dry, tholeiitic systems like Kīlauea (e.g., Scaillet et al., 1998;Wallace & Edmonds, 2011;Webster & Botcharnikov, 2011).Thus, as magma ascends beneath arc volcanoes, there is a greater depth-and likely time-window over which gases may rise to the surface to interact with crater lakes compared to at Kīlauea.
Low-viscosity basalts, like those at Kīlauea, also lend themselves to fast preeruptive ascent rates and narrow conduits or dikes (Wilson & Head, 1981).While basaltic magmas may indeed still erupt through water lakes and induce phreatomagmatic explosions, as with the 2005 eruption of Karthala Volcano (Comoros; Bachèlery et al., 2016), in the case of the 2020 eruption at Kīlauea, narrow fissures were fed by dikes that were small enough to erupt near to (i.e., on the crater walls), but not through, the water lake.A wider conduit, as at more silicic volcanoes, would be more likely to intersect or interact with a water lake, potentially leading to precursory signals.Indeed, depending on magma overpressure and crustal stress regime, it is possible for even basaltic magma to ascend slowly enough that there is sufficient time for gases or heat to influence the surface, including, potentially, a water lake.Though such a precursory signal appears to be uncommon at Kīlauea, the onset of the 2008-2018 summit eruption at Kīlauea was preceded by a months-long increase in SO 2 emission rate (Elias & Sutton, 2012;Wilson et al., 2008).
At volcanoes with prominent rift zones (e.g., Kīlauea, Karthala), it can sometimes be that the final stages of magma migration ahead of eruption are often not only fast, but predominantly lateral rather than upward, thus limiting the opportunity for escaping heat and gases to interact with surface water above the eventual eruption site.For example, the dike feeding the prolonged 2018 lower East Rift Zone eruption at Kīlauea propagated ∼20 km over three days to the site of the first eruptive fissures (Neal et al., 2019).While the 2018 eruption did not involve a phreatic component, previous eruptions in the lower East Rift Zone were indeed phreatic or phreatomagmatic (Moore, 1992).With similar dike propagation times as the 2018 eruption, it may be likely that any precursory hydrogeochemical signals would become apparent only very shortly (minutes-days) ahead of such phreatic/ phreatomagmatic activity.
The difficulty in forecasting eruptive behavior around volcanic lakes is exacerbated by the fact that, as at Kīlauea in December 2020, only a few hundred meters difference in vent location may mean the difference between a potentially explosive eruption through a water lake and lava passively flowing into a lake from above.At Karthala Volcano in recent years, the difference between summit lava lake eruptions (May-June 2006, January 2007) and phreatomagmatic eruptions (April 2005, November-December 2005) has meant the difference between lava contained within the summit crater, with no threat to the local population, and widespread ashfall affecting nearly 250,000 residents, with subsequent years of lahars (Bachèlery et al., 2016).

Future Studies on Incipient Crater Lakes
Kīlauea's 2019-2020 crater lake was unprecedented in modern times, but not geologically (Swanson et al., 2014), and future caldera collapse could contribute to the appearance of another lake.Further, though crater lakes are more common at more silicic systems along volcanic arcs (e.g., Delmelle & Bernard, 2015;Varekamp, 2015), other basaltic shield volcanoes similar to Kīlauea have had documented crater lakes (e.g., Karthala; Bachèlery et al., 2016) and recent phreatic activity (e.g., Karthala and Piton de la Fournaise; Bachèlery, 1981;Bachèlery et al., 1995;Michon et al., 2013;Ort et al., 2016;Savin et al., 2005;Thivet et al., 2020Thivet et al., , 2022)).At these volcanoes, as at Kīlauea, and at any other volcanoes that develop incipient crater lakes, future studies could consider several improvements to our methods that would help with better understanding of the controlling processes, which, in turn, may improve the assessment of potential hazards.Despite the significant payload limitations associated with water sampling using small UAS, they currently provide the safest way to access remote crater lakes.Thus, our recommendations for future studies are based on the use of such UAS systems.
We propose sampling lake waters as soon as possible after their appearance, which may mean that logistical considerations need to be addressed in advance.Our study has demonstrated that important processes that controlled the lake's composition (i.e., the rapid leaching of freshly fractured talus rocks and alteration minerals) were in play before we sampled for the first time (3 months after the lake appeared), as indicated by changes in lake color.Addition of analyses of iron redox state (Fe 2+ and Fe 3+ as compared to total Fe) would help reduce uncertainty in geochemical modeling.We also propose inclusion of biological sampling for cultivation tests as part of the study protocol.Such tests can indicate the presence of iron-oxidizing (or other) bacteria and help distinguish between biotic and abiotic processes that control lake chemistry.In situ dissolved oxygen, conductivity, and pH measurements with high-temperature sensors could help eliminate some of the assumptions on processes (e.g., cooling) that took place during transport of the samples from the lake to the site where they were processed at the top of the crater.Additional chemical analyses (e.g., δ 13 C of dissolved carbonate, etc.) could also prove useful for certain lake systems with higher pH.Utilizing larger, higher-endurance UAS platforms could potentially allow for sampling water from greater depths within the lake, making temperature-depth-redox (via platinum electrode) profiles, recording video from within the water column, and collecting sediments that would enable a better characterization of mineral precipitates.
We also propose improvements to existing continuous, real-time monitoring instruments when a lake is present.An automated colorimeter with high spectral sensitivity could allow for a rapid determination of changes in lake chemistry composition (during daylight hours).Modeling of acquired images can provide information on the attenuation by absorption and molecular scattering as well as on the properties causing surface reflection of the water (e.g., Nugent et al., 2015).A station housing precipitation, relative humidity, and wind speed sensors located closer to the lake would be beneficial for constraining both rainfall input and estimates of evaporation from the lake surface; continuous in situ monitoring of lake temperature would also improve evaporation estimates and could help calibrate thermal imagery.A continuous laser rangefinder tracking the elevation of the water surface could improve calculation and monitoring of the lake's physical dynamics, independent of visibility conditions.All of these improvements to existing continuous instrumentation could enhance the automated alarming strategies routinely employed in real-time volcano monitoring.

Conclusions
The main goals of this study were to infer the hydrologic, geologic, and magmatic conditions that led to the formation of a lake in Kīlauea's summit crater in July 2019, as well as to use a multi-parameter approach to characterize and model the chemical processes that occurred within the lake during its 17-month existence.Based on a wide range of instrumental measurements, chemical and isotopic analyses of lake water collected with UAS during three sampling campaigns, mineralogical and isotopic analyses of alteration minerals, and thermodynamic calculations, we conclude: 1.The stable isotope composition (δ 18 O and δD) of the Kīlauea lake is consistent with significant evaporation of locally recharged meteoric water (Figure 9).The steady rate of water inflow into the lake and the insensitivity of the water level to local rainfall events indicates that the lake was mainly fed by groundwater (Figure 4b).2.Even the most conservative estimate indicates that ∼50% of groundwater inflow to the lake was lost to evaporation.The actual percentage may be significantly higher (Figure 4a).3. The pH of the Kīlauea water lake (3.8-4.5) did not align well with the bimodal distribution of volcanic lakes worldwide (0.5-1.5 and 6-6.5;Marini et al., 2003), yet SO 4 2 concentrations were comparable to those measured in the hyper-acidic crater lakes of Kawah Ijen (Indonesia), Poás (Costa Rica), and Ruapehu (New Zealand).A similar bimodal pH distribution was also found in acid mine drainages (Bigham & Nordstrom, 2000;Cravotta et al., 1999).4. The relatively high concentrations of Fe, Mg, Mn, and other solutes in the lake are best explained by leaching from the fractured basalt in the caldera collapse talus pile.This process was previously hypothesized to explain the chemical composition of the nearby NSF well waters (Hurwitz et al., 2003).However, the lake waters were far more concentrated than those in the NSF well, presumably due to an increased availability of leachable and soluble materials within the collapse talus pile as compared to the area near the well.The frequent presence of green-colored water at lake edges over the lifetime of the lake (Figures 5 and 6) likely indicated the continued input of reduced iron (Fe 2+ ). 5. Minor sulfur degassing into the lake may have occurred but was likely negligible with respect to generating the high dissolved SO 4 2 concentrations in the lake, given the lake's moderate pH.Consequently, the low SO 2 emission rates measured during the water lake era were unlikely to be a result of gas scrubbing and rather indicated the continued absence of very shallow magma.If significant gas scrubbing had occurred within the lake, the lake water should have become markedly more acidic.6.The major element chemical composition, δ 34 S values, and thermodynamic modeling suggest that leaching of basaltic rock talus and dissolution of sulfate alteration minerals into the lake followed by iron oxidation reactions resulted in the unique pH composition of the Kīlauea lake waters (Figure 13).
Figure 13.Cartoon summary of the crater setting and the main processes occurring at key benchmark stages for the water lake at Kīlauea in 2019-2020.See Table 1 for the lake chemistry at the stages depicted in panels (b, c).
Continuous dissolution of sulfate minerals into inflowing groundwater, augmented by SO 4 2 already present in groundwater, was the main process that resulted in high SO 4 2 concentrations.The lake itself was saturated in Ca-sulfates (gypsum or anhydrite) throughout the entire timeline of chemical sampling.With continued lake growth, incoming groundwater diluted the lake despite the high evaporation rates, although the total solute load in the lake continued to substantially increase over time as the lake grew in volume.The potential role of microbial activity in affecting the lake's chemical composition was unexplored.7.There were no observed changes to the lake or in SO 2 emission rates in advance of Kīlauea's December 2020 summit eruption.The physicochemical nature of Kīlauea's high-temperature, low-H 2 O tholeiitic basalt, including low viscosity, rapid ascent, narrow conduits, and shallow late-stage degassing of SO 2 , is likely responsible for the lack of precursory signals in the lake.While phreatic and phreatomagmatic activity is inherently difficult to predict or forecast in any setting (Montanaro et al., 2022), it may be particularly challenging at basaltic systems such as Kīlauea, where a lateral conduit shift of only a few hundred meters may result in very different eruptive outcomes (Figure 13d).

Appendix A: Select Water Lake Multimedia Links
Here we provide a selection of links to videos, animations, and infographics that were produced over the course of the lake's evolution.Based on δ 34 S measurements in Kīlauea basaltic lavas (Sakai et al., 1982) and melt inclusions and matrix glasses (Lerner et al., 2021), magmatic SO 2 is likely to be 1-2‰ isotopically heavier than its source melt, resulting in degassing SO 2 having a δ 34 S range of +1‰ to +3‰, depending on how sulfur-degassed the source melt is (degassed magma is isotopically lighter; Lerner et al., 2021).We assume that the magma underlying the Kīlauea water lake would have been relatively sulfur-undegassed and the δ 34 S signature of ascending SO 2 would therefore be roughly +2‰ to +3‰ (Figure 10).
At the time of the first geochemical sampling campaign (26 October 2019), the Kīlauea water lake contained ∼1 × 10 5 m 3 of water (∼1.03 × 10 8 kg of water, assuming a density of 1,025 kg/m 3 for the high measured TDS concentrations at >60°C).However, considering the extensive evaporation that occurred from the lake, a cumulative total of 2.5 × 10 5 m 3 of water (2.48 × 10 8 kg assuming a density of 993 kg/m 3 without evaporative concentration) had likely entered the lake basin at that point.Assuming that this incoming water had a sulfur concentration and δ 34 S signature matching that of the waters from NSF well within Kīlauea's caldera (7,040 mg/L SO 4 2 ; +6.9‰ δ 34 S), 2.0 × 10 6 kg SO 2 gas input would be required to shift the δ 34 S signature of the lake water to the observed value of 3.4‰ (Figure B1).Averaged over the 86 days of the lake's existence up to that time, this mass of SO 2 input would be attained with steady input rate of 23 t/d SO 2 input, which could be reasonable given the ranges of SO 2 emissions measured at the summit during eruptive quiescence.This amount of SO 2 dissolving into the lake would enrich the lake's SO 4 2 concentration from 7,040 mg/L to 50,000 mg/L (after evaporative concentration), which is close to the measured lake water concentration of 53,000 mg/L.The 7,040 mg/L is the highest SO 4 2 concentration measured in the well; more recent values have been lower (Hurwitz et al., 2019) and would have required additional SO 2 input to the lake.However, the dissolution of SO 2 would cause significant acidification of the water (see Equations 6-8); even the relatively modest amount of SO 2 input into the lake to match the δ 34 S signature would add 6.2 × 10 7 mols of H + into the water lake, which would decrease the lake's pH to <1, unless there was strong pH buffering through other water or rock reactions.Considering that the measured lake water was only mildly acidic (pH 3.8-4.5),we conclude that the input of SO 2 was unlikely to be the main source of sulfur enrichment in the lake, though small amounts of SO 2 scrubbing may have occurred.Consequently, the role of SO 2 scrubbing in the lake was likely relatively minor, and the low SO 2 emission rates measured during that time indicated that very shallow magma was not present.
Other sulfur sources that could have contributed to the elevated SO 4 2 levels in the Kīlauea water lake are sulfur-bearing alteration minerals, such as native sulfur, sulfides, and sulfates.Native sulfur and sulfate minerals were present in fumaroles around Halemaʻumaʻu prior to the 2018 caldera collapse, and many of these alteration minerals would have been present in the talus pile in which the water lake eventually grew.Interestingly, no sulfides have been identified among the alteration minerals sampled during this work, although it is possible that sulfides were present in more reduced subsurface environments.These sulfurbearing alteration minerals were formed over previous decades of steady degassing across caldera floor and by more intense alteration associated with the 2008-2018 lava lake degassing, as evidenced by sulfur-bearing alteration minerals present among material ejected during small explosions induced by rockfalls into the Halemaʻumaʻu lava lake.
Native sulfur sampled at Kīlauea's summit and from 2018 lower East Rift Zone fissures have isotopically light δ 34 S signatures of 14‰ to 9‰.The isotopically light δ 34 S signature of these native sulfur samples likely indicates their formation following disproportionation-hydrolysis reactions of SO 2 in low-temperature magmatic-hydrothermal systems (Kusakabe et al., 2000).Assuming these isotopically light signatures are representative for most native sulfur present at Kīlauea, the dissolution of native sulfur into the water lake would thereby exert significant isotopic leverage on the lake's δ 34 S values.In a binary mixture between NSF well groundwater and native sulfur, 7.8 × 10 4 to 1.1 × 10 5 kg of native sulfur would be required to match the measured lake water δ 34 S measured on 26 October 2019 (Figure B1).However, this amount of native sulfur input would only have enriched lake water SO 4 2 concentrations to ∼40% of the measured values (∼21,000 mg/L vs. 53,000 mg/L measured), while greatly decreasing the lake's pH to ∼1.Additionally, the solubility of native sulfur is very low at the pH and temperature conditions relevant to the lake and groundwater (<100°C; e.g., Ellis & Giggenbach, 1971).Consequently, it is unlikely that native sulfur dissolution was a major source of sulfur into the lake.If sulfides were present among the boiling point temperature fumaroles that were common around Halemaʻumaʻu, they would have very isotopically light δ 34 S signatures due to the extreme fractionation from SO 2 source gas at boiling point temperatures.Combining δ 34 S fractionation factors for SO 2(g) -H 2 S (g) from Richet et al. (1977) and H 2 S (g) -FeS (s) from Li and Liu (2006), assuming that these higher-temperature formulations reasonably extend to 100°C, sulfides precipitating from SO 2 -dominated magmatic gases would be ∼30‰ lighter than the corresponding SO 2 gas, with the isotopic fractionation further increasing at lower temperatures.Consequently, for magmatic gas with a likely δ 34 S value of +3‰, corresponding sulfides would have δ 34 S values of 27‰ at near-boiling point 2 ) concentration and δ 34 S signatures measured on 26 October 2019.These mixing models assume that the composition of waters sourcing the lake water matched that of (a) sulfate-rich and (b) lower-sulfate groundwater sampled in the NSF well in February 2010 (Hurwitz et al., 2019) and January 2020, respectively.Additional possible sulfur sources within Kīlauea's caldera include SO 2 magmatic gas, sulfates (assumed as CaSO 4 ), native sulfur, and sulfides (assumed as FeS 2 ), which are stoichiometrically added to groundwater with δ 34 S ranges assumed for gas δ 34 S (Lerner et al., 2021) or measured in alteration minerals (see Results section).Mixing proportions are determined by iterating to match the observed Kīlauea water lake δ 34 S signature and water volume.Water pH during gas and mineral dissolution is calculated by stoichiometric mass balance, without considering acid neutralization reactions.Groundwater pH is assumed to match the average of NSF well waters measured during 1998-2002 (Hurwitz et al., 2003).A mixture of sulfate-rich groundwater and dissolved sulfate minerals, plus a small input of acid-forming component (SO 2 gas, native sulfur, or sulfide minerals) reasonably reproduces the measured lake water δ 34 S, SO 4 concentration, and pH.Other water and sulfur mixtures fail in one or more parameters (red text).Modeled lake water concentrations include a solute enrichment factor of 2.5x to account for the evaporation of 150% of the observed lake volume water (see text).δ 34 S analytical uncertainty of waters and minerals is ±0.4‰ (2σ).
temperatures.If the hypothetical sulfides are pyrite, the fractionation between SO 2 and sulfide would be somewhat less, due to the higher oxidation state of sulfur in pyrite (FeS 2 ; S 1-) compared to the hypothetical troilite (FeS; S 2-) used in the fractionation calculations above.Regardless, precipitating sulfides would be very isotopically light, and sulfur added into the Kīlauea water lake from dissolved sulfides would exert an even greater isotopic leverage during mixing than native sulfur.Assuming a sulfide δ 34 S value of 27‰, 8.7 × 10 4 kg of dissolved FeS 2 would need to mix with NSF well groundwater to match the lake water δ 34 S measured on 26 October 2019 (Figure B1).This amount of sulfide input would only enrich the lake water SO 4 2 and Fe concentrations to ∼38% and ∼79% of the measured values, respectively (∼20,000 mg/L vs. 53,000 mg/L SO 4 2 measured; ∼420 mg/L vs. 527 mg/L Fe measured), while decreasing the lake's pH to ≤2.
Consequently, neither native sulfur nor Fe-sulfides are reasonable mixing sources to account for the observed δ 34 S, SO 4 2 , Fe, and pH of the Kīlauea water lake.
In contrast, the dissolution of Ca-sulfate minerals from within the collapsed talus pile could reasonably produce the observed δ 34 S and SO 4 2 of the water lake without greatly affecting the lake's pH (Figure B1).We therefore view the dissolution of sulfate minerals as the most likely dominant source of the elevated SO 4 2 within the lake, as discussed in the main text.However, the NSF well groundwater had a near-neutral pH of ∼7.8 (Hurwitz et al., 2003) whereas the Kīlauea water lake had a slightly acidic pH of ∼4.2.As the dissolution of Ca-sulfates into the lake water would not directly affect pH, some other acidifying agent was presumably present in the lake.Section 5.1.2discusses the acidifying effects of iron oxidation on the lake's pH; the dissolution of quite small amounts of SO 2 gas (or native sulfur and sulfide minerals) would also serve to acidify the lake water, while having minimal effects on the rest of the lake composition.Consequently, the most likely set of sulfur inputs into the lake water might be a ternary mixture of NSF well groundwater, ∼4.3 × 10 6 kg of Ca-sulfate (anhydrite) and ∼200 kg of SO 2 gas (or 50 kg of native sulfur, or 190 kg of sulfides [FeS 2 ]), which would closely reproduce the measured lake water δ 34 S, SO 4 2 , and pH measured on 26 October 2019 (Figure B1).

Figure 1 .
Figure 1.Map of Kīlauea summit region.Specific sites referenced in the text are annotated.

Figure 2 .
Figure 2. Halemaʻumaʻu and crater lake, viewed from a permanent web camera to the west of crater.View is to the east.Note abundance of sulfur/ sulfate-bearing fumarolic vents along northeastern sector of the crater wall.(a) Main crater on 1 November 2019.Maximum left-right (approximately north-south) lake dimension in photo is ∼65 m.(b) Closer view with lake outline on various dates.Solid black line is the base photo on 1 November 2019.Dashed black outlines within the lake are approximated from photos taken at different vantage points on 7 August 2019 and 1 September 2019.Yellow outlines are at 2-month intervals from 1 January 2020-1 November 2020.Cyan outline is 20 December 2020, the last date at which the lake was present.Note that ledge in foreground obscures near (west) end of lake from view; dashed cyan line approximates obscured lake shoreline.

Figure 3 .
Figure 3. Methods used to sample and monitor the water lake at Kīlauea summit.(a) The UAS sampling system taking flight, showing depth indicator flags on suspended cord, Drone Cone sampler with HydraSleeve, and temperature sensor.Photograph by S. Warren, U.S. Geological Survey, 17 January 2020.(b) UAS and sampler in flight, with crater wall backdrop.Photograph by P. Nadeau, U.S. Geological Survey, 26 October 2019.(c) Colorimeter on lake rim.Photograph by M. Patrick, U.S. Geological Survey, 21 July 2020.(d) Geologist making laser rangefinder measurements of lakeshore elevation.Photograph by M. Patrick, U.S. Geological Survey, 6 April 2020.

Figure 4 .
Figure 4. (a) Measured lake depth (m; dark blue dots) and cumulative water volumes (m 3 ; green dots).The cumulative evaporative (pink lines) and discharge volumes (black lines) were determined by polynomial fitting of lake volume and evaporated volume per day followed by integration of the polynomial curves.Discharge into lake was calculated by adding lake volume to evaporated volume, subtracting the volume of rain that fell directly onto the lake surface.Gray field brackets on lake depth and lake volume span the range from minimum to maximum estimates(Patrick, Swanson, et al., 2021).Vertical gray lines denote water sample dates.(b) Lake volume (m 3 ; black line with green dots) and daily direct rainfall input volume (m 3 ; light blue line) for 3 months bracketing largest rainfall event during the lifetime of the water lake.Raw data fromPatrick, Swanson, et al. (2021)  and NPS rain gauge.

Figure 5 .
Figure 5. (a) Evolution of lake surface color, from initial transparent blue-green to orange-brown and dark brown.Approximate maximum North-South (left-right) lake dimension is indicated in each photo.Photographs, in chronological order, by S. Conway, J. Babb, P. Nadeau, M. Patrick, D. Swanson, K. Mulliken, D. Swanson, M. Patrick, and M. Patrick, U.S. Geological Survey.(b) Sample timeseries stack of profiles across webcam images to monitor color changes in August/ September of 2020, along with rainfall plot from UWE rain gauge on the north caldera rim.White 'G's highlight instances of enhanced green coloration on west end of lake.

Figure 6 .
Figure 6.(a) Dominant wavelengths of lake surface colors with time.Approximate corresponding colors are displayed in the background.Note break in x-axis.(b) Example of the striking range of colors that could exist simultaneously within the lake.Of note is the persistence of green to bluish-green influx zones at the lake edges into late 2020.Approximate maximum North-South (left-right) lake dimension is 172 m.Photograph by M. Patrick, U.S. Geological Survey, 17 November 2020.

Figure 7 .
Figure 7. Visible (left) and thermal imagery (right) from 21 May 2020.The thermal image represents the maximum temperature from a stack of multiple images as a means to remove the effects of cooler, variable steam above the lake's surface.Note coincidence of sharp visible color boundary with sharp thermal boundary in center of lake, whereas other thermal features are not as strongly linked to water color.Approximate maximum North-South (left-right) lake dimension is 137 m.Photographs by M. Patrick, U.S. Geological Survey.

Figure 11 .
Figure11.Calculated saturation indices (SI) for various minerals based on water chemistry of samples HM19-01 and HM20-02A as determined with PHREEQC(Parkhurst & Appelo, 1999).Calculations were carried out with water temperature at 70°C (dark colors) and 80°C (light colors) and at 100% Fe 3+ (circles) and 50% Fe 3+ (diamonds).Minerals with positive SI are expected to precipitate from the water.

Figure 12 .
Figure12.Imagery from HVO's F1 thermal webcam on 20 December 2020(Patrick et al., 2022).(a) Kīlauea crater and water lake at 21:28:32 local time, approximately 1-2 min prior to eruption.No abnormalities or changes are evident.(b) Image from 21:30:33 showing the onset of lava effusion from a collapsed caldera block perched higher than the water.The lake appears identical to the earlier frame, unperturbed by the nearby lava.Note also the mostly transparent gas plume issuing from the erupting fissure, likely a typical eruptive plume consisting of predominantly water vapor, SO 2 and CO 2 .(c) The next camera frame, captured at 21:32:22.Lava has cascaded from its perched caldera block into the water lake below, causing sudden boiling of the lake water and resultant opaque steam plume.Color/ temperature scale identical for all three frames.

Figure B1 .
Figure B1.Simplified mixing models of possible water and sulfur sources to produce the Kīlauea crater lake water sulfate (SO 42 ) concentration and δ 34 S signatures measured on 26 October 2019.These mixing models assume that the composition of waters sourcing the lake water matched that of (a) sulfate-rich and (b) lower-sulfate groundwater sampled in the NSF well in February 2010(Hurwitz et al., 2019)  and January 2020, respectively.Additional possible sulfur sources within Kīlauea's caldera include SO 2 magmatic gas, sulfates (assumed as CaSO 4 ), native sulfur, and sulfides (assumed as FeS 2 ), which are stoichiometrically added to groundwater with δ 34 S ranges assumed for gas δ 34 S(Lerner et al., 2021) or measured in alteration minerals (see Results section).Mixing proportions are determined by iterating to match the observed Kīlauea water lake δ 34 S signature and water volume.Water pH during gas and mineral dissolution is calculated by stoichiometric mass balance, without considering acid neutralization reactions.Groundwater pH is assumed to match the average of NSF well waters measured during 1998-2002(Hurwitz et al., 2003).A mixture of sulfate-rich groundwater and dissolved sulfate minerals, plus a small input of acid-forming component (SO 2 gas, native sulfur, or sulfide minerals) reasonably reproduces the measured lake water δ 34 S, SO 4 concentration, and pH.Other water and sulfur mixtures fail in one or more parameters (red text).Modeled lake water concentrations include a solute enrichment factor of 2.5x to account for the evaporation of 150% of the observed lake volume water (see text).δ 34 S analytical uncertainty of waters and minerals is ±0.4‰ (2σ).

Table 1
Chemical Composition of Halemaʻumaʻu Water Lake Samples Peek et al. (2023)s are reported in milligrams per liter (mg/L).Concentrations in parentheses are from ICP-MS measurements, whereas all other concentrations were determined by ion chromatography (Cl, F, SO 4 ) and ICP-OES except for SO 3 , which was determined by titration.Sample temperatures are those recorded by in situ thermistors rather than those derived from thermal imagery.All of the data and description of the methods are available inPeek et al. (2023).a Percent difference between final (average of analyses from 10/26/2020 samples) and initial (analysis of 10/26/2019 sample) measurements.