Geochemistry, Geophysics, Geosystems

Masking of interannual climate proxy signals by residual tropical cyclone rainwater: Evidence and challenges for low-latitude speleothem paleoclimatology

Authors


  • The copyright line for this article was changed on 6 May 2015 after original online publication.

Abstract

[1] The anomalously low oxygen isotope ratio (δ18O values) of tropical cyclone rainfall can transfer proxy information about past tropical cyclone activity to stalagmite oxygen isotope records. Isotopically distinct stormwater reaches the growing crystal surface as a coherent slug, or after attenuation by mixing with isotopically normal vadose groundwaters. A high-resolution micromilled stalagmite stable isotope record from Belize shows that residual tropical cyclone water from Hurricane Mitch masked the oxygen isotope record of a major El Niño event. On decadal time scales, measured δ18O values are affected by changes in local tropical cyclone frequency. Despite the tropical cyclone masking effect, the structure of the “missing” El Niño event is preserved in the ATM-7 carbon isotope ratios (δ13C values). In tropical cyclone-prone regions, the fidelity of stalagmite oxygen isotope proxy data to recording background climate signals is modulated by temporal variations in local tropical cyclone rainfall, and the sensitivity of individual stalagmites to tropical cyclone masking varies with hydrology. Speleothem δ13C values, unaffected by tropical cyclones, can preserve the underlying structure of climatic variability. For low-latitude speleothems with C–O isotope covariance, intervals in which the δ18O values are significantly lower than δ13C values predict may indicate periods when local tropical cyclone masking of isotope-derived precipitation records is enhanced by greater infiltration of tropical cyclone rain. The temporal structure in stalagmite C–O isotope covariance has paleoenvironmental meaning that may be revealed by exploring factors associated with independent behavior in each isotope ratio, respectively. Tropical cyclone masking presents new challenges to paleoclimatology and a source of hypotheses for paleotempestology.

1. Introduction

[2] Stable oxygen isotope time series from stalagmites is a key source of qualitative and quantitative paleoprecipitation proxy records in the low latitudes [cf. McDermott et al., 2006; Lachniet, 2009; Dayem et al., 2010]. In general, lower oxygen isotope ratios are associated with higher rainfall amounts. It has been recently established that information about tropical cyclone activity can be extracted from stalagmite stable isotope records at semiannual to subseasonal resolution [Breitenbach et al., 2011; Frappier et al., 2007b; Nott et al., 2007]. This stable isotope paleotempest proxy exploits the characteristically low stable isotope ratios of tropical cyclone rain water compared to typical summer season rain in the low to midlatitudes [Lawrence and Gedzelman, 1996]. The sensitivity of individual stalagmites to recording measurable isotopic signatures of tropical cyclone rainwater is dependent upon not only the storm water's isotopic contrast, but also the hydrological characteristics of the conduit(s) that feed a stalagmite's drip source [Frappier, 2008]. The karstic plumbing system feeding each stalagmite thus controls both the amount of isotopically distinct stormwater that reaches the growing crystal surface, and the extent to which the stormwater arrives either as a coherent slug, or after attenuation by mixing with other vadose groundwaters with more typical oxygen isotope ratios (δ18O values). It is thought that fast-drip fracture-fed stalagmites will be more likely to record significant infiltration from individual tropical cyclone rain events that deliver a coherent slug of stormwater, compared to slow-drip stalagmites with more diffuse source water pathways that better reflect long-term average infiltration, including effects such as evaporation and seasonal differences in stormwater infiltration [cf. Ayalon et al., 1998; Baldini et al., 2006; Frappier, 2008; Tooth and Fairchild, 2003]. For sensitive stalagmites, infiltration of tropical cyclone water from storms separated by as little time as 2 weeks can be recorded as separate transient low excursions in δ18O values that persist for one to several weeks of calcite deposition [Frappier et al., 2007b]. Historical tropical cyclone rain events are detectable with a sampling resolution of 2 to 50 samples per year [Nott et al., 2007; Frappier et al., 2007b]. In one such well-characterized stalagmite (ATM-7), the stable isotope ratio time series showed substantial covariation between δ18O and δ13C values; one hallmark of historical tropical cyclone excursions is that the δ18O values undergo a substantial, transient decrease that is independent of trends in δ13C values [Frappier et al., 2007b]. However, all stormwater does not reach the cave simultaneously, and thus the low δ18O value excursions described by Frappier et al. [2007b] do not provide a complete understanding of the effects of tropical cyclones on speleothem paleoclimate proxy data.

[3] In general, diffusion and dispersion during infiltration of tropical cyclone rainwater require that even the most sensitive stormwater slug-transmitting conduits will mix some fraction of isotopically anomalous tropical cyclone water with preexisting vadose groundwaters, and store it in the epikarst for some time following the storm event [Frappier, 2008]. In this paper, I consider the fate and effects of this isotopically anomalous stored cyclogenic water, referred to herein as residual stormwater. Residual stormwater, added to the karst system over the course of a few hours or days, is characterized by the low δ18O values typical of tropical cyclones, which may be several per mille lower than other local meteoric water [Lawrence, 1998].

[4] Recent studies have revealed the importance of tropical cyclone rainfall to hydroclimatic variability on regional and global scales. For example, tropical cyclones contribute 2–3% of global annual rainfall (Williams and Marks, 2009), 8–9% of total hurricane season rainfall across the entire Atlantic Basin [Jiang and Zipser, 2010], 10–20% of total rainfall along the US Gulf and Mexican coasts, and as much as 60% of rainfall in arid areas of northwestern Mexico [Englehart and Douglas, 2001]. About 10% of climatological rainfall in Yucatán can be attributed to tropical cyclones, which shift the region from semiarid to a milder dry tropical climate. In the northern Yucatán, these episodic recharge events often contribute an extremely large quantity of water, sometimes equivalent to 50% of the average annual rainfall [Escolero et al., 2007]. Below, I present evidence that the infiltration of residual stormwater in karst regions has significant and persistent effects on stalagmite stable oxygen isotope proxy records that result, including the masking of interannual to decadal climate signals of interest. I also present an approach to detecting past intervals of high local tropical cyclone activity that exploits the insensitivity of speleothem δ13C values to residual stormwater. Lastly, I explore the implications of this tropical cyclone masking effect for paleotempestololgy and low-latitude paleoclimatology.

2. Observations

[5] The tropical cyclone masking isotope effect is demonstrated through an analysis of the effects of historical tropical cyclone events on a subseasonally resolved stalagmite stable isotope record, with evidence of longer-term trends associated with annual to decadal variations in tropical cyclone rainfall near the cave site.

2.1. Evidence of a Tropical Cyclone Masking Effect in a Published Stalagmite Stable Isotope Record

[6] Evidence from a modern Belize stalagmite, ATM-7, shows that residual tropical cyclone water from a single major hurricane rainfall event can depress the δ18O value of stalagmite calcite for months to years (Figure 1). This stalagmite was micromilled at 20 μm intervals [Frappier et al., 2002, 2007b] and although the growth rate varies between and within hydrological years, the continuous micromilling sampling process ensures that no aliasing effects bias the record. The ATM-7 stable isotope proxy record of El Niño events from 1978 to 2001 [Frappier et al., 2002] is characterized in general by high δ18O values that covary with even larger amplitude interannual peaks in δ13C values. The entire ATM-7 record lags the Southern Oscillation Index (SOI) record of El Niño events by ∼1.5 years due to lags in the teleconnection and water infiltration [Frappier et al., 2002; Frappier, 2006]. However, unlike earlier El Niño events in the record when both δ18O and δ13C values show high values, the 1997–1998 event is different (Figure1): δ13C values are typical for El Niño events, averaging −6.5‰, but δ18O values are anomalously low at −4.5 ‰. In the ATM-7 record of the 1997–1998 El Niño event, the expected increase in stalagmite δ18O values is not evident, yet the δ13C values respond as in previous events (Figure 2). This particular El Niño event was different in that it affected the cave site during an interval of high local tropical cyclone rainfall, in particular, when Belize was affected over the course of a few weeks by the catastrophic Hurricane Mitch, and later by Tropical Storm Katrina.

Figure 1.

Comparison between local annual rainfall, a high-resolution stalagmite stable isotope record, and El Niño events. Annual rainfall for 1975–2000 at Central Farm meteorological station in Belize (blue line), located within 30 km of the cave site, shows neither decadal differences nor any significant lagged or instantaneous correlation to the Southern Oscillation Index [Frappier et al., 2002]. ATM-7 stable isotope record, after Frappier et al. [2007b] showing overall covariation between (top, black) δ18O values and (bottom, green) δ13C values. Major El Niño events recorded by both δ18O and δ13C values are highlighted with yellow boxes, in contrast to the final El Niño event (pink box), which is prominent only in δ13C values. The extended period in the mid-1990s of relatively mild El Niño events (yellow double arrow) is coincident with low δ18O values and tropical cyclones. Measured δ18O value excursions related to local tropical cyclones are highlighted in red. The decadal average δ18O value for the 1990s (gray bar), with eight tropical cyclones is 1.87‰ lower than for the 1980s (gray bar), with only one tropical cyclone. 1979, an extremely wet year in central Belize when local precipitation exceeded three standard deviations above the climatological mean, contains the expected paleorainfall amount proxy signal of low δ18O values in the corresponding annual layer of ATM-7.

Figure 2.

Detail of 1998–1999 in ATM-7 stable isotope record. δ18O values (black) include red highlighted low excursions corresponding to local tropical cyclone events in 1998 (Mitch) and 1999 (Katrina). Both δ13C and δ18O values began increasing together at the onset of the El Niño event, until the first low δ18O value excursion. Three brief low δ18O value excursions correspond to an extended period of high δ13C values (green) related to El Niño.

[7] Hurricane Mitch produced rain in Belize while approaching Guatemala and Honduras from the Caribbean Sea, and again after circling around to the south and entering the Bay of Campeche [National Hurricane Center, 2011]. The storm generated tens of centimeters of rain at local meteorological stations near the cave site [Frappier et al., 2007b], while torrential flooding and landslides killed thousands of people in Guatemala and Honduras [National Climatic Data Center, 1998]. To my knowledge, no rainwater samples from Hurricane Mitch were analyzed for their stable isotope composition, but the stormwater δ18O values were probably extremely low, despite changes in storm intensity along the track [Lawrence and Gedzelman, 1996; Fudeyasu et al., 2008]. The ATM-7 stable isotope record for 1998 contains two low δ18O value excursions, which [Frappier et al., 2007b] interpreted as two infiltrating stormwater slugs from the meandering passage of Hurricane Mitch (1998) through the area. Figure 2 shows that trends of rapidly increasing δ18O and δ13C values in ATM-7 were established in early 1998 as the El Niño event began to affect the area; but, following the two low δ18O value excursions, δ18O values remained low for the duration of the El Niño event. In contrast, δ13C values in ATM-7 continued to increase, and remained high into late 1999 when the local effects of El Niño finally waned. The stalagmite again shows a breakdown in covariation between δ18O and δ13C values in 1999, when Belize was struck by Tropical Storm Katrina (1999). Figure 2 shows that the low δ18O value anomaly in 1999 occurred when δ18O values in ATM-7 drop but never recover, yet at the same time the stalagmite recorded δ13C values that remained high for months, reflective of the persistent El Niño conditions. By the spring of 2000, oxygen and carbon isotope ratios in ATM-7 returned to their typical pattern of covariation.

[8] This pattern of a discrete breakdown in stalagmite δ18O-δ13C value covariation following tropical cyclone events is a clear example of a tropical cyclone masking effect, as evidenced by the suppression of the expected high δ18O values in ATM-7 calcite during this El Niño episode by stormwater from Hurricane Mitch, which depressed δ18O values for months following the storm event. A similar pattern can be observed during the mid-1990s, when several tropical cyclone events occurred during a period of mild El Niño events (Figure 1). During the mid-1990s, ATM-7 recorded weak El Niño events as moderately elevated δ13C values; yet, δ18O values remain relatively low without the high δ18O value signal typical of other El Niño events earlier in the record. The tropical cyclone masking effect described above is consistent with the ATM-7 time series during the mid-1990s, where one can observe the decoupling of an established δ18O-δ13C value covariation pattern during a few years of high local tropical cyclone rainfall.

2.2. Evidence of a Long-Term Contribution of Tropical Cyclone Rain to Speleothem δ18O Values

[9] Residual stormwater can also depress speleothem calcite δ18O values over longer time scales. Correlations between decadal tropical cyclone activity and speleothem average δ18O values have been reported in previous studies from Central America to Bermuda [e.g., Breitenbach et al., 2011; Malmquist, 1997; Schwehr, 1998; Webster, 2000; White, 2002]. These observations hint that it may be possible to reconstruct tropical cyclone activity from decadal variations in speleothem δ18O values, but given the unusual isotopic composition of residual cyclogenic water it also begs the question: How would one distinguish between the effects of persistent climate variability and transient storm events? Close inspection of the continuously micromilled ATM-7 record provides both useful and troubling perspectives on this problem.

[10] First, it is notable that the decadal average δ18O values in ATM-7 for the 1980s exceed those of the 1990s by 1.87‰, and carbon isotope values by 3.19‰ (Figure 1). Although both δ13C and δ18O values in ATM-7 decline dramatically in 1989, this rapid change can be understood as the end of a typical El Niño event rather than any tropical cyclone rainfall effect, because no tropical cyclones rained at this site during those years. The Southern Oscillation Index shows no significant difference for the corresponding decadal periods. Thus, the decadal differences in ATM-7 stable isotope values cannot be attributed simply to the local effects of interannual El Niño variability (Table 1). Local decadal rainfall totals show no differences between the two decades (Figure 1), and no seasonal biases are evident in monthly precipitation data from the nearby Central Farm station. Instead, the observed decadal difference in δ18O values in ATM-7 is likely attributable at least in part to differences in local tropical cyclone rainfall input beginning in 1991 (Table 1). During the 1980s, the cave area was affected by only one tropical cyclone, in contrast to eight tropical cyclones in the 1990s. Even in this relatively humid region, tropical cyclones can deliver substantial rainfall. Rainfall from individual storms in the 1980s and 1990s varied from about 25 to 275 mm at Central Farm [Frappier et al., 2007b]. In 2001, the nearby Belize Zoo recorded 470 mm of rain from Hurricane Keith. Tropical cyclone rainfall in the 1990s contributed nearly 4% of total precipitation, at least four times more than the 1980s (Table 1). For recent years with tropical cyclone rain events in central Belize, the storms contribute an additional ∼2–28% of rainfall above the annual climatology. ATM-7 shows a clear association over two decades between tropical cyclone rainfall amount and the decadal average δ18O values of stalagmite calcite. Although I cannot rule out the contribution of other well-known isotope effects to the observed decadal differences in δ18O values, it is clear that when tropical cyclone events occur more frequently, the average δ18O value of speleothem calcite is also lower.

Table 1. Comparing the 1980s and 1990s for Belize Weather Data and Stalagmite Stable Isotope Values
 1980s1990s
  1. a

    Central Farm and Belize Zoo stations.

  2. b

    SOI indicates prevailing warm conditions during the early 1990s.

Number of TC rain events18
Total TC rainfall (mm)a140600
ATM-7 average δ18O values (‰)−3.05−4.92
ATM-7 average δ13C values (‰)−7.98−11.17
Number of El Niño eventsb2>1
ATM-7 δ18O values (‰) during all three major El Niño events−2.9 (s.d. 1.1)
ATM-7 δ13C values (‰) during all three major El Niño events−6.1 (s.d. 1.8)

2.3. Evidence of the Insensitivity of Speleothem Carbon Isotope Ratios to Tropical Cyclones

[11] Speleothem δ13C values respond indirectly to rainfall, and are affected by factors such as cave ventilation, soil biogenic carbon dioxide production, and prior calcite precipitation [cf. Fairchild and Baker, 2012; McDermott et al., 2006]. Although theoretically possible, there is no evidence that speleothem δ13C values respond to tropical cyclones. In the micromilled ATM-7 record (Figure 1), changes in δ13C and δ18O values over time are often closely correlated, yet subseasonal to decadal δ13C values measured in ATM-7 were insensitive to the effects of transient tropical cyclone rainfall events, and yet closely associated with persistent anomalies such as El Niño events. Figures 1 and 2 show that even during intervals when speleothem oxygen isotope ratios are severely depressed by residual cyclogenic water (e.g., 1997–1998 El Niño event), δ13C values in ATM-7 preserved the same underlying association between the El Niño event and carbon isotopes seen in previous El Niño events.

3. The Tropical Cyclone Masking Effect: Predictions and a Theoretical Test

[12] Some predictions can be made about the tropical cyclone masking effect on the basis of the observations described above. First, following a tropical cyclone rainfall event, the average δ18O values measured in cave dripwater and speleothem calcite will be depressed as a function of the extent to which the stormwater slug is homogenized and attenuated during seepage water infiltration, leading to differences between stalagmite stable isotope time series. Second, temporal covariation in the stable carbon and oxygen isotope ratios of cave dripwater and speleothem calcite may be interrupted temporarily after tropical cyclone rainfall events, such that the δ18O values are lower than predicted by the prevailing covariation pattern. A theoretical test performed on the first prediction is described below. The second prediction is addressed by a case study comparing a long-term speleothem record from Belize with an independent regional paleotempest record.

[13] To simulate the effect of tropical cyclone masking on growing calcite speleothems that are sampled annually for stable isotope analysis, I present results from a two end-member mixing model of cave seepage water at the source drip on an annual basis (supporting information1 ). I assume that the cave temperature is stable, that annual infiltration is isotopically invariant (except in specified cases), that evaporation in the soil and epikarst is negligible, that infiltration is proportional to rainfall, and that there is no seasonal bias in infiltration or calcite deposition. Modeled tropical cyclones add rainfall at 10% above a normal year, and stormwater input occurs over a period of a few days [Frutos, 2006]. Modeled average tropical cyclone rain is 10 per mille lower than background rainwater [Gedzelman et al., 2003; Lawrence and Gedzelman, 1996]. The model assumes that maximum residence time for seepage water is 1 year, and that all tropical cyclone water infiltrates within 1 year. It neglects the effects of complex karst hydrology including isotopically exchangeable water reservoirs. Table 2 gives the essential model parameters.

Table 2. Tropical Cyclone Stormwater Mixing Model Parameters
 Normal YearTropical CycloneEl Niño Year
Rainfall volume (mm)16001701600
δ18O values (‰)−3.85−10.00−2.85

[14] I used this simple mixing model in two experiments that simulate the isotopic composition of speleothem source dripwater on an annual basis, corresponding to typical discrete sampling rates for high-resolution stalagmite stable isotope-based paleoclimate proxy records. First, the model can demonstrate the effect of differences in mixing hydrology during infiltration between stalagmites on the resulting cave dripwater stable isotope signal. Second, the model illustrates the effect of transient tropical cyclone rainfall events compared to persistent climate anomalies such as El Niño events on the resulting stable isotope signal of the cave dripwater.

3.1. Modeled Effect of Hydrologic Mixing on Dripwater and Calcite Stable Isotope Data

[15] The conduit hydrology feeding a stalagmite can vary from very poorly mixed to fully homogenized [Tooth and Fairchild, 2003]. The extent of homogenization of the stormwater is specified in the model as a percentage, from zero to 100%. Low homogenization indicates that little of the stormwater is mixed with other groundwaters, and a majority of the stormwater is delivered to the stalagmite surface as a coherent slug [Frappier, 2008]. The resulting δ18O value excursion is deposited on the speleothem surface over a period of days to weeks, a small feature that can be entirely or partially missed by discrete drill transects that integrate calcite deposited over months to years [Spötl and Mattey, 2006]. Conversely, highly homogenizing conduits attenuate most of the stormwater slug in the epikarst, and little if any of the stormwater remains distinguishable when it arrives at the cave drip site and stalagmite calcite deposition surface. Excursions from individual storm events are not present in the resulting speleothem calcite. I varied the homogenization term from 0 to 1 for a normal year with one hurricane rainfall event, and plot the resulting dripwater composition at 1 year poststorm (Figure 3). As a percentage of annual rainfall, the stormwater slug volume decreases from 10% to 0 as the stormwater homogenization increases from zero to full integration.

Figure 3.

Modeled average cave dripwater oxygen isotope composition illustrating the tropical cyclone masking effect for normal years (black) compared to El Niño years (yellow), both without and with hurricanes (red, orange), assuming complete stormwater homogenization. One tropical cyclone depresses annual dripwater δ18O values, making the entire year appear wetter. In this scenario, El Niño dripwater δ18O values are 1.0‰ higher than a normal year. Dripwater for a year with an El Niño event and a tropical cyclone (orange) is very similar to normal values, an example of tropical cyclone masking that can alter the δ18O value record of interannual climate variability.

[16] Differences in hydrology between stalagmites can affect the δ18O values of annually averaged modeled cave dripwaters by up to 0.69‰ (Figure 3). In the case of a minimally homogenizing conduit, dripwater δ18O values generally reflect background climate, with a brief spike of very low δ18O value water breaks through for a very brief time during the year. This low δ18O value excursion may be missed in drilled transects, and may contribute to low δ18O values or even a hurricane spike detectable in micromilled records. Modeled results show that a substantial low δ18O value excursion may be distinguishable even after significant homogenization. For example, a half-homogenized hurricane spike representing less than a month of rainfall and only 50% of the original stormwater slug volume may retain a δ18O value excursion magnitude of −0.36‰. The resulting calcite δ18O values will generally reflect background climate, but drilled or milled samples containing the stormwater-spiked calcite could be lower than indicated by the “amount effect” of background climate over the course of the entire year. At the other extreme, very well-mixed conduits deliver dripwater with significantly depressed annual δ18O values, resulting in proxy interpretations prone to significant annual rainfall overestimates (e.g., up to 7% for a single storm for sites in Central America and the Caribbean, using amount effect values from Lachniet and Patterson [2009]).

[17] Although the isotopic offset for a single typical tropical cyclone event is limited to about −0.7‰, multiple tropical cyclones have often affected cave sites in a single season during the historical period [e.g., Lambert and Aharon, 2010]. Although the total rain amount might be elevated during active storm years, the isotopic proxy signature recorded in stalagmites would appear to be very wet throughout that period, rather than only over a few days. For multiple storms in a year, speleothem oxygen isotope-based rainfall amount reconstructions from stalagmites with highly homogenizing conduits could overestimate annual rainfall during active storm intervals by a substantial fraction. Without an independent record of tropical cyclone activity, it is difficult to distinguish the isotopic signatures of a wet, storm-free year and a normal year with one tropical cyclone, or between a very wet storm-free year, and a dry year with a few tropical cyclones. Quantitative estimates of the isotopic signatures resulting from such combinations of persistent local climate variability and ephemeral tropical cyclone events will be site specific, varying considerably between sites, and also among stalagmites within cave systems. Conversely, for sites currently prone to frequent tropical cyclone rain events, extended streaks of storm-free years would result in speleothem calcite deposited with higher δ18O values, even when background rainfall rates remained stable. If tropical cyclone masking is not considered in the amount effect proxy transfer functions, researchers could underestimate paleorainfall during periods of quiescent tropical cyclone activity.

3.2. Modeled Effect of Tropical Cyclone Masking on Proxy Records of El Niño

[18] Figure 4 shows the modeled effects of homogenizing seepage water from tropical cyclones and El Niño events alone and in combination. For stalagmites with highly homogenizing conduits, dripwater resulting from a hurricane during an El Niño year is indistinguishable from dripwater from a normal year. Thus, if the mixing time scale of cave seepage water is approximately annual, interannual climate signals with a magnitude of ≤ about 0.7‰ can be entirely masked by typical interannual variations in the amount of residual stormwater. This modeling result is consistent with the tropical cyclone isotope masking effect seen in ATM-7 (Figure 2). While this stalagmite's pattern of interannual climatic and isotopic variability may be unusual, this model experiment demonstrates that the effect of tropical cyclone masking can be large compared to the typical amplitude of stable isotope variations in stalagmite records.

Figure 4.

Modeled average annual cave dripwater oxygen isotope composition during an El Niño event at 1 year posthurricane, as a function of stormwater homogenization. When stormwater is unhomogenized and arrives in the cave as a coherent slug (yellow), the tropical cyclone masking effect is minimized in stalagmites and is confined to the calcite containing the low δ18O value spike. When stormwater from an El Niño year is mostly homogenized, dripwater δ18O values are similar to those for a normal year (gray bar), and the tropical cyclone masking effect is important.

3.3. Speleothem δ13C-δ18O Value Covariation: A Silver Lining?

[19] Stalagmites commonly display some degree of δ13C-δ18O value covariation over time when sampled along the central growth axis [Mickler et al., 2004]. For many stalagmites affected by tropical cyclone masking, approximately annually resolved stable carbon isotope data may provide critical clues for distinguishing between persistent climatic and transient cyclogenic controls on isotopic variability. For speleothems like ATM-7 that display some covariation in δ13C and δ18O values (Figure 1), δ13C values often reflect the structure of background climate variability more consistently than δ18O values during times when tropical cyclone masking is important. In general, tropical cyclone masking can affect stalagmites deposited in equilibrium and nonequilibrium conditions alike, and the underlying factors involved in δ13C-δ18O value covariation need not be understood for periods of independent behavior to be characterized and used to detect instances of tropical cyclone masking.

[20] For speleothem time series from tropical cyclone regions, δ13C-δ18O value covariation is reduced by independent stable oxygen isotope variations due to cyclogenic rainwater; i.e., tropical cyclone rainwater is a source of error in δ13C-δ18O value time series correlations. One can further infer that the temporal structure of δ13C-δ18O value covariation in speleothem time series has climatic meaning: transient breakdowns or changes in the coefficient of covariance may be due to independent factors affecting only calcite δ13C values, or only δ18O values. Comparing temporally averaged correlation coefficients for published speleothem records [Mickler et al., 2006] may obscure useful information about paleoclimate and/or proxy systematics that can be revealed by exploring the temporal patterns of covariance changes [Holmgren et al., 1999]. Various approaches might be used, for example a running index of covariation, derivative, or principle component analysis [Frappier, 2006]. An index which compares the measured δ18O values (δ18OM) to the δ18O values predicted from measured δ13C values (δ18OP) point by point, the Covariation of Stable Isotopes (CoSI) Index (equation (1)), may be a simpler and more direct indicator. Squared differences in the CoSI Index emphasize the largest departures from predicted values, indicating important processes acting independently on C or O isotope ratios.

display math(1)

[21] δ18OP = (correlation coefficient * δ13CM + δ13C intercept), from a least squares linear regression fit model on detrended data

[22] The CoSI Index is near zero when covariation between δ13C and δ18O values is greatest. Because speleothem stable isotope values are usually negative, the CoSI Index is high when δ18O values are lower than expected or when δ13C values are higher than expected. Similarly, the CoSI Index is low when δ18O values are higher or δ13C values are lower than expected. The CoSI Index provides a way to use the more persistent climatic signals preserved in stalagmite carbon isotope ratios to screen stalagmite stable isotope records for time periods when climate signals and patterns of covariation change in specific ways. A CoSI Index could represent different physical factors, depending upon which factors affect the independent stable isotope variability of the stalagmite in question. In the case of tropical cyclone masking, high CoSI Index values that coincide with low δ18OM values point to instances when tropical cyclone rainfall may have depressed speleothem oxygen isotope ratios. Although high CoSI Index values might be generated by other mechanisms, for stalagmites from tropical cyclone-prone regions, the combination of high CoSI Index and low measured δ18O values clearly indicates independent oxygen isotope behavior that is most likely due to times of greater infiltration of low δ18O value waters: snowmelt or tropical cyclones.

[23] Previously published stalagmite records may contain instances of tropical cyclone masking. For example, although stalagmite MC-01 from Belize exhibits covariation in δ18O and δ13C values from ∼300 B.C.E. to ∼2001 C.E. [Webster et al., 2007], a number of subsamples depart from this pattern with distinctly lower than expected δ18O values. If the MC-01 record is interpreted in a tropical cyclone masking framework, one can postulate times when tropical cyclone rainfall may have played an important role (Figure 5a). Interestingly, the periods of enhanced tropical cyclone rainfall shown in Figure 5a also correspond to periods of frequent regional storm strikes identified in independent paleotemest proxy records from Viequez, Puerto Rico [Donnelly and Woodruff, 2007] (Figure 5b). This correlation is consistent with my assertion that tropical cyclone rainwater is an important though poorly recognized contributor to the stable isotope proxy data recovered from many stalagmites.

Figure 5.

Possible instances of tropical cyclone masking in a published speleothem record. (a) Stable isotope time series from stalagmite MC-01 from Belize [Webster et al., 2007], exhibiting moderate covariance in δ18O and δ13C values over the last three millennia. Yellow bars denote times when covariance is low and MC-01 δ18O values are lower than predicted from changes in measured δ13C values, i.e., inferred periods of increased tropical cyclone masking and high regional tropical cyclone activity. (b). Coastal sediment paleotempest record from Viequez, Puerto Rico [Donnelly and Woodruff, 2007], where frequent El Niño events (gray) alternate with periods of coarse grained deposits caused by intense local hurricane strikes. Most intervals of inferred tropical cyclone masking in Figure 5a correspond closely with documented periods of high hurricane activity elsewhere in the Atlantic Basin (Figure 5b).

4. Implications

4.1. Tropical Cyclone Masking Effect Modulates Speleothem Stable Isotope Paleoprecipitation Proxy Accuracy

[24] Residual stormwater from Hurricane Mitch, coincident with the 1997–1998 El Niño event effectively masked the ATM-7 stable oxygen isotope proxy record of that interannual event (Figures 1 and 2). During the 1990s when tropical cyclone events were very frequent in Belize but rainfall climatology was normal, the stalagmite recorded significantly lower average δ18O values than during the 1980s (Figure1). Although these decadal changes may be related in part to established stable isotope effects such as changes in moisture sources [Dansgaard, 1964], it is also likely that shifts in the amount of tropical cyclone rainfall had a depressing effect on ATM-7 calcite during the 1990s. These observations suggest that during intervals of high tropical cyclone activity, residual stormwater can depress the average δ18O values of speleothem calcite to a greater extent than would be predicted from the “amount effect” alone. The oxygen isotope record of background climate variability recorded by speleothems can be depressed for months to years by the isotopically anomalous cyclogenic rainwater from each event. Ultimately, for low-latitude speleothem proxy records of past climate, the masking effect of transient tropical cyclone rain events that last only days produces a bias toward recording low δ18O values that, when interpreted using commonly used proxy transfer functions, indicate months to years of climatic conditions that are significantly wetter than the background climate conditions during that time. Consequently, for speleothems from caves near tropical cyclone basins, interannual to decadal variations in δ18O values are not simply interpretable as a function of rainfall amount. Rather, variable rainfall contributions from tropical cyclones can modulate the accuracy of some tropical and subtropical speleothem oxygen isotope-based paleoclimate reconstructions, leading to biased overestimates of rainfall during periods of locally higher tropical cyclone activity. Because tropical cyclone rainfall changes over time, the resulting speleothem δ18O value data may contain occasionally inaccurate representations of the timing, duration, and intensity of climate variations on seasonal, interannual, and multidecadal time scales. Residual tropical cyclone water may even lead to speleothem proxy records that underestimate the magnitude and duration of droughts punctuated by tropical cyclones unless interpreted in tandem with an independent local paleotempest record. Likewise, tropical cyclone masking does not affect stalagmite δ18O values when tropical cyclone events do not occur, leading to higher δ18O values during storm-free intervals.

[25] Other factors capable of depressing δ18O values on interannual to centennial time scales in speleothem records have been applied in the interpretation of diverse paleoclimatic proxy signals (reviewed in Lachniet [2009]), including: changes in moisture sources and storm tracks, differential seasonal infiltration and recharge, increased rainfall amount or reduced evaporation, changes in cave temperature or cave ventilation, calcite deposition conditions, and periodic snowmelt and glacial meltwater infiltration [Vollweiler et al., 2006]. The temporally variable impact of the newly described tropical cyclone masking effect must be considered when interpreting the paleoclimatic significance of speleothem stable isotope proxy records from karst regions that are affected by tropical cyclone rainfall.

[26] Speleothem rainfall proxies have been developed on the basis of correlations between rainfall amount and speleothem δ18O values, as well as vertical growth rates (annual layer thickness), fluorescence intensity, and δ13C values (see e.g., review in Fairchild and Baker [2012]). Yet, only δ18O values are likely to be affected by isotopically anomalous tropical cyclone events. The tropical cyclone masking effect may thus explain some observed error in correlations between rainfall amount and stalagmite records of growth rate, fluorescence intensity, and δ18O values, particularly for karst regions affected by tropical cyclones where stalagmite annual layer thickness is small or background meteoric water δ18O values are low. The unique sensitivity of δ18O values to tropical cyclone rainfall may make it possible to identify times when calcite δ18O values are depressed by tropical cyclone masking, by comparing independent records of paleotempest activity, rainfall, and changes in stalagmite δ18O-δ13C value covariation. The isotopic composition of tropical cyclone rainfall explains some of the error in correlations between δ18O and δ13C values, and perhaps other hydrologically sensitive proxies such as Mg/Ca, fluorescence intensity, annual layer thickness, and tree-ring width.

[27] By applying approaches such as the CoSI Index, hidden paleotempest activity indicators may be detectable in existing speleothem records from tropical cyclone-prone sites, providing a useful source of hypotheses for paleotempestology. Further comparisons between measures of the temporal structure of speleothem stable isotope covariation and past tropical cyclone activity may prove fruitful in the ongoing effort to understand and reliably use speleothem proxy signals in paleoclimatology. The coefficient of covariation for δ13C and δ18O values in a particular stalagmite time series may be better estimated during storm-free intervals when tropical cyclone masking does not alter δ18O values.

4.2. The Magnitude of the Tropical Cyclone Masking Effect Varies by Region and Speleothem

[28] Although viewing all speleothem data with a tropical cyclone masking lens is not recommended, speleothems from many karst regions may be affected to some degree (Figure 6). The magnitude and importance of the tropical cyclone masking effect is likely to differ by region, based on variations in tropical cyclone rainfall, interannual climate variability, and the stable isotope composition of local background meteoric water. For example, karst areas in or near high-relief terrain, or on the windward sides of tropical islands may experience orographically enhanced tropical cyclone rainfall compared to nearby karst areas in lowlying or downwind areas [Lin et al., 2002]. Because tropical cyclone rainfall can create heavy downpours and flooding for hundreds of kilometers inland [Rappaport, 2000], the tropical cyclone masking effect may even affect speleothems in caves located far from the coast. For speleothems from karst regions in the far reaches of tropical cyclone-affected areas, tropical cyclone masking must be considered as an isotope effect with the capability to generate occasional periods of low δ18O value calcite, which may be indistinguishable from snowmelt in higher latitudes and cooler periods. Further afield, no systematic stable isotope measurements have been made of rainwater from tropical cyclones after traveling far inland, or their derivatives following the transition to extratropical storms, although a project in progress for Superstorm Sandy (G. Bowen, personal communication, 2012). Tropical and formerly tropical East Pacific weather systems have a major impact on summer season rainfall and groundwater recharge the U.S. desert southwest [Corbosiero et al., 2009; Ritchie et al., 2011], where speleothems have been used to reconstruct paleorainfall data [Asmerom et al., 2007, 2010]. Characterizing any isotopic anomalies in precipitation from those tropical cyclones may be important to correctly interpreting the climatic signals recorded in the region's speleothems.

Figure 6.

Distribution of surface carbonate regions (blue, after Ford and Williams [2007]) affected by historical tropical cyclone rainfall (red and orange shading, after 1842–2012 storm track data from IBTrACS [Knapp et al., 2010] and 1998–2007 tropical cyclone rainfall distributions from Williams and Marks, 2009). Regions affected occasionally by low-intensity tropical cyclones are shaded yellow; regions affected frequently by low-intensity to high-intensity tropical cyclones are shaded red. Not shown in this compilation: subsurface karst areas; subtropical and extratropical storm tracks. This map indicates regions where the tropical cyclone masking effect hypothesized here can be tested on speleothem stable isotope and multiproxy time series.

[29] Subtropical regions generally have lower rainfall and higher meteoric water δ18O values compared to the tropics [cf. Rozanski et al., 1992]. Thus, the isotopic contrast between cyclogenic stormwater and “normal” background rainwater may be greater for drier subtropical karst regions. In some subtropical regions, cyclogenic rainwater contributes 10–60% of annual rainfall and a large fraction of interannual variability [Jiang and Zipser, 2010; Williams and Marks, 2009]. Here residual stormwater is likely to constitute an important independent component in the interannual to multidecadal variability of speleothem δ18O values, and the tropical cyclone masking effect must be considered in interpreting local stalagmite records. For regions with frequent storm strikes, anomalies may occur only during rare streaks of consecutive stormless years. In this case, extended periods lacking tropical cyclones may result in high δ18O value anomalies in speleothem calcite that are not necessarily associated with lower rainfall amounts. Because tropical cyclone rainfall is seasonally restricted, discretely sampled stalagmites are subject to aliasing from differing contributions from residual stormwater to each drilled sample. For midlatitude speleothems, it is important to note that tropical cyclone rain and snow have similar isotope ratios, but different seasonal distributions. Wherever tropical cyclone activity is an important component of interannual rainfall variation, local stalagmite records of interannual stable isotopic variability may be subject to tropical cyclone masking, which may substantially interfere with proxy records of interannual climate variability related to climate modes such as El Niño-Southern Oscillation (ENSO). Detection of tropical cyclone masking may be complicated by the well-documented modulation of tropical cyclone activity by major teleconnection patterns such as ENSO, albeit differentially in the Atlantic compared to the Pacific [Landsea, 2000].

[30] Historical tropical cyclone activity is highly variable and little is presently known about the chronology and spatial patterns of paleotempest activity at most sites [Frappier et al., 2007a]. This gap presents an opportunity for paleotempestology to contribute to bringing low-latitude paleorainfall records into sharper focus. As paleotempestology data sets are developed, it will be increasingly possible to correct stalagmite oxygen isotope-based paleorainfall records for the effects of past tropical cyclone activity. The distribution and isotopic composition of tropical cyclone rainfall are domains where advances in fundamental knowledge can benefit paleoclimatology. Regional monitoring of stable isotopes in modern rainwater (Global Network of Isotopes in Precipitation, or GNIP) are often used to calibrate speleothem δ18O value proxy data, in the absence of local monitoring of rainwater and cave dripwater stable isotope composition. GNIP data and its modeled derivative Isoscapes are invaluable for relating seasonal isotopic differences to prevailing circulation patterns [Bowen, 2010]. Yet, because GNIP stations are so widely spaced and data is only available at monthly resolution, the effects of short-lived meteorological phenomena like tropical cyclones have not been systematically incorporated into speleothem analysis. Tropical cyclones have been shown to affect the distribution of oxygen isotopes in tropical water vapor even outside the storms themselves [Lawrence and Gedzelman, 2003; Lawrence et al., 2004], suggesting a systematic and intimate link between tropical cyclone activity and the stable isotope composition of low-latitude rainwater. The isotopic composition of rainwater from remnant tropical cyclone systems is a key unknown relevant to speleothem paleoclimatology. New rainfall isotope data from satellites [Helliker and Noone, 2010; Yoshimura et al., 2011] and cave site monitoring [Beddows et al., 2007] including Cavity Ring-Down spectroscopy (CRDS)-based measurements [Brand et al., 2009] hold promise.

4.3. A Multiproxy Approach to Reducing Tropical Cyclone Masking Effects and Improving Climate Reconstruction

[31] Speleothems contain a wealth of useful paleoenvironmental signals, often measured in multiproxy studies. The speleothem replication test [Dorale and Liu, 2009] is the “gold standard” for developing regional climate histories, because consistent proxy records are most likely produced by a consistent environmental signal. The tropical cyclone masking effect's sensitivity to dripwater conduit hydrology may contribute to some periodic lack of replication between records from nearby areas with different tropical cyclone histories, and to some lack of coherence between speleothem δ18O values and tree-ring width-based dendroclimatology records. When compositing multiproxy data from a single speleothem to improve the quality of a climate reconstruction factor such as rainfall, intervals exhibiting poorer replication in related proxies can fingerprint intervals in which independent behavior is forced by different and potentially identifiable processes. In general, the temporal patterns of covariation and independent behavior among related series in multiproxy speleothem data sets can be used to distinguish amongst climatically important forcing factors, including the divergent effects of persistent climate anomalies and ephemeral storm events.

[32] Widely available speleothem carbon isotope data present an alternative solution to the problem of tropical cyclone masking. Carbon isotope records may provide critical clues for distinguishing between persistent climatic and transient cyclogenic controls on isotopic variability in approximately annually resolved tropical spelean oxygen isotope records, such that temporary decoupling of carbon and oxygen isotopic records may indicate periods of high local tropical cyclone rainfall infiltration. For these time series, δ13C values can record the structure of past climate variations even when δ18O values are depressed due to tropical cyclone masking. The temporal pattern of stable isotope covariation in speleothem time series data is not constant, and its history contains useful information for paleoclimatology. Analysis of measured speleothem data in combination with a covariation series such as the CoSI Index can enable researchers to generate hypotheses about the past behavior of factors, such as tropical cyclones, that affect O or C stable isotope ratios independently. This approach may provide a means to identify specific intervals when independent factors disrupt underlying covariation patterns. For stalagmites with generally correlated trends in carbon and oxygen isotope values, any temporary decoupling when the oxygen isotope values are lower than expected may be a fingerprint of the tropical cyclone masking effect, and a marker for periods of high local tropical cyclone input. Identifying periods of tropical cyclone masking in speleothem data can also indicate times when the paleorainfall proxy may be prone to overestimates and inaccuracies. Existing speleothem records from tropical cyclone-prone regions should be reevaluated in light of nearby paleotempestology data, when available, to assess tropical cyclone masking effects on paleorainfall reconstructions. Where paleotempestology data are not available, and speleothems are affected by tropical cyclone rainfall, incorporating tropical cyclone masking into paleorainfall proxy models may increase error estimates.

5. Conclusions

[33] Tropical cyclone rainfall is an important but previously overlooked factor that affects many low-latitude speleothem paleorainfall reconstructions, particularly in subtropical karst (Figure 6). Instrumental-era proxy data observations and model experiments show that tropical cyclone rain with anomalously low δ18O values that falls over a few days can depress the oxygen isotope ratio of cave dripwater and speleothem calcite by up to ∼0.7‰ for months to years following each storm event. The tropical cyclone masking effect causes a proxy bias in paleorainfall amounts derived from speleothem oxygen isotope data wherever and whenever tropical cyclones occur, producing overestimates of paleorainfall amounts by a few to tens of percent during times when the cave site experienced high storm activity. The same speleothem stable isotope proxy records exhibit no bias during storm-free intervals. When the time scales of cave seepage water mixing time scales and modes of climate variability are similar, the persistent depressing effect of tropical cyclone rain events on vadose groundwater δ18O values can overprint prevailing climate patterns to effectively hide the speleothem oxygen isotope signature of some climate events (e.g., interannual climate variations such as El Niño events). Over time, variable contributions of isotopically anomalous rainwater from tropical cyclones can modulate the fidelity of the stalagmite oxygen isotope proxy signals to recording background hydroclimatic patterns.

[34] Recognition of tropical cyclone masking as a stable isotope effect has a number of implications for the paleoclimatology community. Developing accurate representations of past rainfall variability from speleothems has become essential for understanding the causes of societally important climate extremes, proxy-model intercomparisons, and climate change attribution studies; thus, the newly recognized tropical cyclone masking effect must be considered alongside other established stable isotope effects in interpreting new and existing speleothem records of hydroclimate variability on interannual to multicentennial time scales within affected regions. Perhaps the most problematic effect of tropical cyclone masking is due to wholesale changes to the structure of variability in δ18O value-based paleorainfall proxy records, which can frustrate accurate analyses of past local teleconnection impacts, attempts at replication, and blur frequency spectra. Tropical cyclone masking may affect some speleothem stable isotope records generally, even those deposited in isotopic equilibrium, and those used only for qualitative rainfall indicators. Past interpretations of low-latitude speleothem records without considering the time-transient effects of tropical cyclone masking may have inadvertently led to some underestimates of past local teleconnection impacts, substantial overestimates of rainfall, underestimates of drought intensity, persistence, and duration, as well as inaccurate estimates of interannual to centennial hydroclimatic variability. Earlier, such rainfall proxy data should be reevaluated to assess the effects of tropical cyclone masking on uncertainty, rainfall amount, and the temporal structure of inferred hydroclimatic changes. Speleothem records presently provide the most precisely dated record of past abrupt climate changes [Dayem et al., 2010]; thus, the prospect that local tropical cyclone masking could generate inaccuracies in some reconstructions of the timing of past climate variations is a serious concern ripe for future studies.

[35] The locations of key speleothem paleorainfall records affected by tropical cyclone masking should be targeted in future paleotempestology studies in order to ascertain and/or improve the accuracy of those paleorainfall records.

[36] Although tropical cyclones can mask the stable oxygen isotope proxy structure of interannual to centennial climate variations, some speleothems affected by tropical cyclone masking may retain information about the structure of past climate changes in carbon isotope ratios. The temporal structure in stalagmite C–O isotope covariance has paleoenvironmental meaning that may be revealed by exploring factors associated with independent behavior in each isotope ratio, respectively. Future research along these lines may lead to more accurate reconstructions of patterns of change in paleoprecipitation. This initial study suggests that instances of tropical cyclone masking in speleothem data sets may be identifiable through analysis of covariation, intercomparison of independent rainfall proxy records, or comparisons to nearby paleotempest activity records. If so, then paleorainfall proxy signals can in principle be corrected to more accurately reflect the timing of onset and duration of droughts and wet intervals previously identified on the basis of δ18O value amount effect proxy interpretations, and related frequency spectra of climate variability. Indeed, it is likely that tropical cyclone masking contributes to error in C–O isotope covariation within stalagmite records, and to differences among independent paleoprecipitation proxy time series. Decoding the dual isotope signatures of tropical cyclone masking in affected speleothem records could unveil a valuable hidden source of data about possible past tropical cyclone activity for hypothesis testing in paleotempestology.

Acknowledgments

[37] This work was carried out with the aid of a grant from the Inter-American Institute for Global Change Research (IAI) CRN II 2050 which is supported by the US National Science Foundation (grant GEO-0452325), and financial support from the Charles Lubin Family Chair for Women in Science.

Ancillary