Extended Cave Drip Water Time Series Captures the 2015–2016 El Niño in Northern Borneo

Abstract

Time series of cave drip water oxygen isotopes (δ¹⁸O) provide site-specific assessments of the contributions of climate and karst processes to stalagmite δ¹⁸O records employed for hydroclimate reconstructions. We present ~12-year-long time series of biweekly cave drip water δ¹⁸O variations from three sites as well as a daily resolved local rainfall δ¹⁸O record from Gunung Mulu National Park in northern Borneo. Drip water δ¹⁸O variations closely match rainfall δ¹⁸O variations averaged over the preceding 3–18 months. We observe coherent interannual drip water δ¹⁸O variability of ~3‰ to 5‰ related to the El Niño–Southern Oscillation (ENSO), with sustained positive rainfall and drip water δ¹⁸O anomalies observed during the 2015/2016 El Niño. Evidence of nonlinear behavior at one of three drip water monitoring sites implies a time-varying contribution from a longer-term reservoir. Our results suggest that well-replicated, high-resolution stalagmite δ¹⁸O reconstructions from Mulu could characterize past ENSO-related variability in regional hydroclimate.

Plain Language Summary

Cave stalagmites allow for the reconstruction of past regional rainfall variability over the last hundreds of thousands of years with robust age control. Such reconstructions rely on the fact that differences in the isotopic composition of rainwater set by regional rainfall patterns is preserved as the rainwater travels through cave bedrock to feed the cave drip waters forming stalagmites. Long-term monitoring of rainwater and cave drip water isotopes ground truth the climate to stalagmite relationship across modern-day changes in regional rainfall. Twelve years of monitoring data presented in this study identify individual El Niño–Southern Oscillation events in rainfall and cave drip water isotopic composition, providing a strong foundation for stalagmite-based climate reconstructions from this site.

1 Introduction

Stalagmite oxygen isotope reconstructions provide key insights into past terrestrial hydroclimate variability globally on seasonal to orbital time scales. In some tropical and subtropical regions where rainfall is dominated by strong vertical convection and infrequent year-round temperature variability, the “amount effect” framework aids interpretations of stalagmite oxygen isotope records. Modern empirical observations classify the “amount effect” as a correlation between high (low) rainfall rates on monthly and longer time scales with depleted (enriched) rainfall δ¹⁸O (hereinafter δ¹⁸O_R) (Craig, 1961; Dansgaard, 1964; Rozanski et al., 1992). Using the amount effect framework, overlapping δ¹⁸O_stal records from monsoon-vulnerable regions in South America (Cheng et al., 2013; Cruz et al., 2005), Australia (Griffiths et al., 2009), Oman (Burns et al., 1998), and India (Sinha et al., 2005) demonstrate interhemispheric antiphasing in δ¹⁸O_R variability from precession-driven (~19–23 kyr) orbital forcing. North Atlantic millennial-length Dansgaard-Oeschger (Dansgaard et al., 1993) and Heinrich events (Heinrich, 1988) are mirrored in stalagmite δ¹⁸O records utilizing the amount effect framework from far-off tropical-subtropical locations such as central South America (Wang et al., 2006), China (Wang et al., 2001), and northern (N.) Borneo (Partin et al., 2007), suggesting a link between high-latitude temperature variations and changes in monsoons (Cheng et al., 2012). Other reproducible stalagmite δ¹⁸O records from N. Borneo (Carolin et al., 2013, 2016; Meckler et al., 2012; Partin et al., 2007) suggest forced changes in Walker circulation properties in the west Pacific warm pool, a major source of heat and water vapor for the interannual climate phenomenon the El Niño–Southern Oscillation (ENSO) (Rasmusson & Wallace, 1983). However, to tease apart external and internal climate forcings over high-frequency natural climate phenomenon, such as ENSO, requires subannually resolved δ¹⁸O_stal records.

Over the last 20 years, the combination of improved high-resolution sampling techniques together with increased temporal resolution, precision, and accuracy of stalagmite U/Th dates have allowed for the generation of absolutely dated subannually to annually resolved stalagmite δ¹⁸O records in stalagmites with relatively fast growth rates. Successful efforts from high-resolution stalagmite δ¹⁸O records spanning the late Holocene have identified individual tropical cyclones in Belize (Frappier et al., 2007) and Australia (Nott et al., 2007), intraseasonal variation of monsoonal rainfall in Thailand (Cai et al., 2010), annual-length droughts in the Yucatan Peninsula (Medina-Elizalde et al., 2010), and the 1997/1998 El Niño event from a Chinese stalagmite (Liu et al., 2018). In regions where monsoon strength is linked to ENSO extremes, high-resolution stalagmite δ¹⁸O records characterize past precipitation-related anomalies in SE China (Zhang et al., 2018) and N. Thailand (Muangsong et al., 2014), as well as in Panama (Lachniet et al., 2004), Belize (Akers et al., 2016), and Mexico (Lachniet et al., 2012). In Borneo, where ENSO exerts a dominant influence on δ¹⁸O_R (Moerman et al., 2013), a subannually resolved stalagmite δ¹⁸O record resolves a reduction in interannual variability during the mid-Holocene (Chen et al., 2016). Despite the drive toward subannually to annually resolved stalagmite δ¹⁸O records, the links between large-scale climate, local rainfall δ¹⁸O, and cave drip water δ¹⁸O remain poorly constrained.

Long-term rainfall and cave drip water monitoring is critical to the identification of large-scale versus local influences on cave drip water δ¹⁸O variability and, by extension, stalagmite δ¹⁸O variability through time. Such data form an integral constraint on proxy system models for stalagmite δ¹⁸O (Dee et al., 2015; Partin, Quinn, et al., 2013), which allow for the transformation of physical climate variables to stalagmite δ¹⁸O variability. For δ¹⁸O_R variability, the amount effect remains an interpretative framework for tropical stalagmite δ¹⁸O records (Lachniet, 2009; Lachniet et al., 2004), although many studies document a range of additional local-scale controls on modern rainfall δ¹⁸O_R. In the tropics, influencing hydroclimate processes over δ¹⁸O_R composition include upstream rainout (Konecky et al., 2019), moisture source convergence (Cai & Tian, 2016), and precipitating cloud type (Aggarwal et al., 2016). Once the rainwater falls to the ground, evaporative enrichment may alter the δ¹⁸O_R signal as it infiltrates into the karst zone (Cuthbert et al., 2014). Because stalagmite δ¹⁸O reconstructions reflect amount-weighted drip water δ¹⁸O variations, it is important to quantify the high frequency (daily to intraseasonal) as well as lower frequency (interannual to subdecadal) controls on rainfall and cave drip water δ¹⁸O variability at a stalagmite δ¹⁸O reconstruction site. Additional factors that govern the rainfall-to-drip water transformation include the size of the vadose zone (Ford & Williams, 2007), the degree of interaction between different reservoirs across the vadose zone (Fairchild et al., 2006), and the preferred recharge flow pathway (Fairchild & Baker, 2012). Karst water routing can vary appreciably through time, with a wide range of water transit times implied by cave drip water δ¹⁸O (hereinafter δ¹⁸O_dw) variability within the same cavern (Partin, Cobb, et al., 2013; Treble et al., 2013; Zhang & Li, 2019). Indeed, long-term monitoring of rainfall and cave drip waters remain crucial to quantify the robustness of stalagmite δ¹⁸O-based climate reconstructions at individual paleoclimate sites.

In this paper we present a daily rainfall δ¹⁸O time series (2006–2018) paired with the longest biweekly cave drip water δ¹⁸O time series (2007–2018) from Gunung Mulu National Park, home to numerous stalagmite δ¹⁸O-based reconstructions (Carolin et al., 2013, 2016; Chen et al., 2016; Meckler et al., 2012; Partin et al., 2007). This study builds on similar work presented by Moerman et al. (2013, 2014), in that it extends Mulu δ¹⁸O_R and δ¹⁸O_dw time series through the very strong 2015/2016 El Niño event. Using the δ¹⁸O_R as input to simulate the observed δ¹⁸O_dw variations at three sites, we constrain the residence times of karst waters and assess the linearity of the δ¹⁸O_R to δ¹⁸O_dw transformation at this key site.

2 Methods

This study extends one rainfall and three cave drip water time series first presented by Moerman et al. (2014) from Gunung Mulu National Park in N. Borneo (4°06′N, 114°53′E) (supporting information Figure S1). For a detailed description of the geologic and climatic setting of Gunung Mulu, the reader is referred to previous studies (Carolin et al., 2016; Cobb et al., 2007; Moerman et al., 2013; Partin, Cobb, et al., 2013). Rainfall samples were collected by Mulu Airport Meteorological staff using a splayed-bottom, copper rain gauge (Casella model M1144003), following the sampling protocol outlined in Moerman et al. (2013). Rainfall and cave drip water samples were collected and stored in 3 ml glass vials and sealed with rubber stoppers and aluminum crimp-tops to reduce evaporation prior to analysis. All samples were measured for δ¹⁸O and δD using a Picarro L2130-i cavity ring-down water isotope analyzer, with a long-term precision better than ±0.1‰ and ±0.5‰ (1σ, N > 500), respectively, following analytical procedures outlined in Moerman et al. (2013).

Cave drip water samples reflect two distinct sampling strategies: (1) quasi biweekly collection at three established drip sampling sites by park staff (Figure S2) and (2) collections spanning most of the Gunung Mulu formation conducted during large field expeditions (Figure S1). For each drip water collection, corresponding drip rates were recorded in drips per minute (dpm). The Wind Fast (WF) and Wind Slow (WS) drip sites are located ~75 m from the entrance to Wind Cave, roughly ~20 m apart dripping at 32 ± 7 dpm (1σ) and 7 ± 1 dpm (1σ). The L2 drip site is located ~140 m from the entranced to Lang's Cave, which is approximately 5 km south of Wind Cave, and drips at 15 ± 3 dpm (1σ). All three drip water time series contain two significant sampling hiatuses: February–July 2014 and September 2014 to August 2015. Spatial surveys of both stalagmite-forming and non-stalagmite-forming drip waters were collected during field expeditions in August 2008 (N = 63), February/March, 2010 (N = 128), October/November 2012 (N = 291), February/March 2013 (N = 37), May 2016 (N = 92), May 2017 (N = 180), and March/April 2018 (N = 124) (Figures S3–S5).

Two proxy system models were used to model the transformation of δ¹⁸O_R to δ¹⁸O_D, generating estimates of karst residence times (τ) given one or two drip water reservoirs at Mulu following Moerman et al. (2014). The autogenic recharge model (hereafter ARM) generates an ensemble of modeled drip water time series' through a backward projected running mean of averaged daily amount-weighted local δ¹⁸O_R for different averaging intervals, or residence times. The bivariate mixing model (hereafter BMM) assigns δ¹⁸O values for two reservoirs (A and B) that mix to form cave δ¹⁸O_dw. The equation for BMM is as follows:

where X_A and X_B are the isotopic composition of reservoirs A and B, respectively, and f_A and f_B are the reservoir A and B mixing ratios, respectively. Throughout the manuscript, the significance of observed Pearson's correlation coefficients is assessed using a Student's t test, where the degrees of freedom have been modified to reflect serial autocorrelation in the data following Bretherton et al. (1999).

3 Results

3.1 Rainfall δ¹⁸O Variability Across ENSO Extremes

Daily δ¹⁸O_R values vary appreciably across the 12-year time series, ranging from −1.6‰ to −24.3‰ (Figure 1b) with an average of −7.3 ± 3.7‰ (1σ, N = 2,664), and δD ranges from −85.4‰ to −153.8‰ with an average of −45.6 ± 29.6‰ (1σ, not pictured). Daily rainfall amounts range from 0 to 300.3 mm/day (Figure 1e) averaging 13.8 ± 22.4 mm/day. Seasonal variability is absent in both rainfall amount and in δ¹⁸O_R, consistent with previous studies in N. Borneo (Kurita et al., 2018; Moerman et al., 2013).

Figure 1

N. Gunung Mulu rainfall and cave drip water oxygen isotope time series. (a) NIÑO 3.4 SSTa index (ERSSTv5; Huang et al., 2017), plotted as a 3-month running average. Red (blue) coloring indicates warm (cold) ENSO anomalies based on a threshold of +0.5 °C (−0.5 °C). (b) Mulu daily rainfall δ¹⁸O (gray circles, non-amount-weighted) plotted with a 7-month running mean (black line). Non-amount-weighted average δ¹⁸O_R for the entire time series is indicated by a dotted line (−7.3 ± 3.7‰), while the amount-weighted average δ¹⁸O_R is indicated by a dashed line (−8.4 ± 2.4‰). (c) Mulu cave δ¹⁸O_dw from Wind Fast (WF; maroon circles), Wind Slow (WS; blue circles), and Lang's Cave (L2; green circles). The solid line indicates the δ¹⁸O_dw averaged across all three drips (−8.0 ± 1.2‰), while the dashed line represents the amount-weighted δ¹⁸O_R average plotted in (b). Box (25th–75th quartiles) and whiskers (total range, excluding outliers; see Figure S5) represent δ¹⁸O_dw values from seven spatial drip surveys. (d) Drip rate in drips per minute (dpm) for the three time series drips plotted in (c), plotted in the corresponding color. (e) Daily Mulu rainfall amount (gray bars) plotted with the 7-month running average (solid black line). y axes in panels a, b, and c are inverted. Monthly ERSSTv5 data found online (at https://www.cpc.ncep.noaa.gov/data/indices/ersst5.nino.mth.81-10.ascii).

Significant correlations between rainfall amount and rainfall δ¹⁸O exist on weekly to interannual time scales at Gunung Mulu. Daily Mulu δ¹⁸O_R values are significantly correlated to local rainfall averaged over the previous 5–8 days (R = −0.39, p < 0.05) and to outgoing longwave radiation averaged over the preceding 5–8 days (R = −0.45, p < 0.05; Figure S6a). Our results suggest a water vapor residence time of roughly 1 week, in line with previous estimates from the site (Moerman et al., 2013). We observe the strongest relationship between Mulu δ¹⁸O_R and Mulu rainfall when both variables are averaged over monthly or longer time scales (Figure S6b and Table S1). For annually averaged data, the correlation between rainfall amount and rainfall δ¹⁸O reaches −0.67 (p < 0.05).

ENSO is the dominant driver of rainfall δ¹⁸O variability in N. Borneo, whereby El Niño and La Niña events drive decreases and increases in regional precipitation, respectively, that influence daily to interannual δ¹⁸O_R variability. Over the entire data set, δ¹⁸O_R is significantly correlated to ENSO indices such as the NIÑO4 index of western equatorial Pacific sea surface temperature (SST) anomalies (R(δ¹⁸O_R, NIÑO4 SST) = 0.64, p < 0.05; Figures 1a and 1b and Table S2). Local rainfall amount exhibits weaker but still significant correlations to ENSO indices (e.g., R(δ¹⁸O_R, NIÑO4 = −0.38); p < 0.01) (Table S2). The 2009/2010 and 2015/2016 El Niño events caused ~50 and 90 days of significantly enriched daily δ¹⁸O_R values up +1.0‰ in January-February-March (JFM) (with respect to mean JFM values; −6.1 ± 2.2‰; Figure S7). Conversely, the 2007/2008 and 2010/2011 La Niña events were associated with significant decreases in daily δ¹⁸O_R of up to −18.5‰ during JFM, albeit with a larger isotopic spread (Figure S7).

3.2 Cave Drip Water δ¹⁸O Variability as a Function of Local Rainfall δ¹⁸O

Drip water δ¹⁸O values for WF and WS δ¹⁸O_dw range from −3.7‰ to −11.6‰ and −3.4‰ to −11.7‰, respectively, while values for Lang's Cave range from −5.9‰ to −10.0‰ (Figure 1c). WF and WS δ¹⁸O_dw values average −7.9 ± 1.3‰ (1σ, N = 253) and −7.7 ± 1.3‰ (1σ, N = 254), respectively, and Lang's Cave (L2) δ¹⁸O_dw average −8.1 ± 0.7‰ (1σ, N = 253), statistically indistinguishable from the amount-weighted average of Mulu δ¹⁸O_R (−8.4 ± 2.4‰, Figure 1c). Mulu amount-weighted δ¹⁸O_R overlaps all three cave δ¹⁸O_dw time series mean values, a unique trait not observed at other tropical cave δ¹⁸O_dw sites whose amount-weighted precipitation is generally more positive than cave δ¹⁸O_dw (Baker et al., 2019). There is no clear relationship between δ¹⁸O_dw and drip rate, nor local rainfall and drip rate, on any time scale (Figure 1d).

All three cave δ¹⁸O_dw time series display coherent interannual fluctuations driven by ENSO variability, consistent with our analysis of δ¹⁸O_R variability at the site. WF and WS drips reflect a higher range of interannual δ¹⁸O_dw variability (~3‰ to 5‰) in comparison to L2 (~1‰ to 2.5‰). All three δ¹⁸O_dw time series become more enriched (depleted) during El Niño (La Niña), with anomalies that persist for months following the peak of each event (Figures 1a and 1c). Indeed, δ¹⁸O_dw anomalies across all three caves scale with the size of ENSO-related SST anomalies, with the largest excursions observed during the 2015/2016 El Niño event (Figure S8). We observe a strong linear relationship between cave δ¹⁸O_dw and NIÑO 3.4 SST anomalies at both Wind cave sites (R² = 0.94 and R² = 0.92 for WF and WS, respectively), whereby El Niño (La Niña) events correspond to enriched (depleted) cave δ¹⁸O_dw values (Figure S8). Lang's cave δ¹⁸O_dw values are also significantly correlated to NIÑO 3.4 SST values (R² = 0.54, p < 0.05) but are overall lower than the correlations observed for Wind Cave (Figure S8).

Seven spatial surveys of stalagmite and non-stalagmite-forming cave δ¹⁸O_dw from nine cave systems reflected system-wide shifts in δ¹⁸O_dw associated with ENSO variability. During the largest events (2010 and 2015/2016 El Niño), cave-wide δ¹⁸O_dw values exhibited the most spread, as drips with a faster residence time reflected large δ¹⁸O_R anomalies during these times, while some reflected little if any shift from long-term average δ¹⁸O_dw, possibly indicative of multiyear residence times (Figure S5). We observed no significant difference in δ¹⁸O_dw mean values for stalagmite-forming versus non-stalagmite-forming drips (−7.8 ± 0.8‰ (1σ, N = 356) and −7.5 ± 1.1‰ (1σ, N = 577), respectively (Figure S4).

3.3 Estimates of Mulu Karst Residence Times

Modeled drip water δ¹⁸O time series derived from local rainfall δ¹⁸O allow for the quantification of karst residence times for the three drip water sites and reproduce up to 85% of the observed drip water δ¹⁸O variability. Karst residence times are estimated using an ARM, where the daily amount-weighted δ¹⁸O_R values are back-averaged over different time intervals (reflecting different karst residence times) to generate a suite of modeled δ¹⁸O_dw time series. A best fit between modeled and observed δ¹⁸O_dw variability is identified as a maximum in Pearson correlation coefficients and a minimum in the sum of residuals calculated between the modeled and observed δ¹⁸O_dw time series. It is important to note that the best fit estimates of residence times are only approximate, as residence times within ±10% of the best fit yield similarly high correlations between observed and modeled δ¹⁸O_dw time series (Table S3). For drips WF and WS, the best fit is obtained for a residence time of 4.5 months (R_WF = 0.89 and R_WS = 0.93, respectively, p < 0.05; Figures 2a–2c), while the best fit for drip L2 is achieved with a residence time of 10.5 months (R_L2 = 0.85, p < 0.05; Figure 2d and Table S3). For drip L2, the modeled drip water δ¹⁸O time series overshoots the observed δ¹⁸O_dw variations by ~ +1‰ during El Niño events and by ~ −1‰ during La Niña events.

Figure 2

Observed versus modeled Mulu drip water δ¹⁸O time series. (a) The top panel is NIÑO3.4 SSTa index (ERSSTv5; Huang et al., 2017), plotted as a 3-month running average with a 1-month overlap. Red (blue) coloring indicates warm (cold) ENSO anomalies based on a threshold of +0.5 °C (−0.5 °C). (b) Observed (open circles) and modeled (black line) δ¹⁸O_dw for drips Wind Fast (WF; maroon) and (c) Wind Slow (WS; blue), where both panels b and c are plotted with a residence time of 4.5 months from the autogenic recharge model. (d) Observed (open circles) δ¹⁸O_dw for Lang's Cave (L2; green) with modeled δ¹⁸O_dw (black line) using a residence time of 10.5 months from the autogenic recharge model. (e) Observed (open circles) δ¹⁸O_dw for Lang's Cave (L2; green) with modeled δ¹⁸O_dw from the bivariate mixing model (dual-reservoir). For pink and purple (maroon) lines, residence time selected is 10.5 months (16.75 months). Reservoir A = 10.5 or 16.75 months from the autogenic recharge model, and Reservoir B is the Mulu amount-weighted rainfall δ¹⁸O mean (−8.4‰). An A:B mixing ratio of 40:60 (purple line) and an A:B mixing ratio of 80:20 (pink and maroon lines) provide an optimal data fit. Note that y axes are inverted in all panels and different y axes scalings.

A BMM that employs two reservoirs—a fast-responding and a slow-responding reservoir—provides a better fit to the L2 δ¹⁸O_dw observations relative to the single reservoir model employed in the ARM used above. For the BMM, we define a short-term Reservoir A that reflects back-averaged rainfall as utilized in the ARM, that mixes with a longer-term Reservoir B that we assign a value of −8.4‰, reflecting the amount-weighted mean δ¹⁸O_R at Mulu over our entire time series. In this model, we simulate the mixing of newly recharged waters (Reservoir A) with older karst waters (Reservoir B). In using the entire 12-year-long time series to estimate a long-term average value for amount-weighted δ¹⁸O_R at Mulu, we make the assumption that the longer the time series utilized for this purpose, the more accurately the estimate reflects the true long-term average of karst waters at Lang's Cave. That said, the long-term average of Mulu amount-weighted δ¹⁸O_R is poorly constrained and can only be approximated by the 12-year average calculated from our time series. We employ two mixing ratios to simulate the observed Lang's Cave L2 δ¹⁸O_dw time series: (1) 40% contributions from Reservoir A and 60% from Reservoir B (hereafter referred to as “40:60”) and (2) 80% contributions from Reservoir A and 20% from Reservoir B (hereafter referred to as “80:20”), following Moerman et al. (2014). The 40:60 mixing scenario yields a significantly better fit to the observed L2 time series than the single-reservoir ARM model (Table S3; R = 0.86, p < 0.05), particularly during the 2009/2010 El Niño event and during an ENSO-neutral period from 2012–2014 (Figure 2e, purple line). The 80:20 mixing scenario also improves on the ARM (Table S3; R = 0.86, p < 0.05), particularly the strongly enriched (depleted) isotopic excursions associated with large El Niño (La Niña) anomalies (Figure 2e, pink and maroon lines).

Analyses of modeled versus observed drip water δ¹⁸O variations for the Lang's Cave drip uncover evidence for a potential change in the karst water residence time across the 12-year drip water δ¹⁸O time series. We find an optimal fit between modeled and observed δ¹⁸O_dw values for L2 when the residence time for Reservoir A changes from 10.5 months during 2007–2014 to 16.75 months during the period 2015–2018 (Figure 2e and Table S3). The most enriched δ¹⁸O_dw values at L2 occur during the very strong 2015/2016 El Niño; however, this period is marked by a prolonged hiatus in the three cave drip water time series (September 2014 to August 2015) that precludes an investigation of recharge effects during the early phase of the 2015/2016 El Niño event. However, we note that the distribution of rainfall intensity during relatively wet (with depleted δ¹⁸O_dw) versus dry (with enriched δ¹⁸O_dw) periods is similar (Figure S9), implying that recharge rates are not the primary driver of the inferred change in mixing scenarios (from 40:60 to 80:20) and residence times (10.5 months to 16.75 months) across the Lang's Cave time series.

4 Discussion

Interannual variations from ENSO-driven rainfall δ¹⁸O are evident in all three Mulu cave drip water δ¹⁸O time series, reflecting rainfall δ¹⁸O variations with karst residence times between ~3 and 18 months. Indeed, all three cave δ¹⁸O_dw time series are consistent with a relatively simple transformation of δ¹⁸O_R to cave δ¹⁸O_dw, supporting previous descriptions of diffuse-seepage flow at Mulu (Cobb et al., 2007; Moerman et al., 2014; Partin, Cobb, et al., 2013). The shorter karst residence times at Wind cave (4–5 months) implies less homogenization of δ¹⁸O_R and/or less vadose zone mixing with other reservoirs of karst waters, resulting in higher amplitude δ¹⁸O_dw variability (−3.7‰ to −11.7‰). This is especially evident during individual ENSO extremes. The longer karst residence times documented at Lang's Cave (10–17 months) result in smaller δ¹⁸O_dw variations (−5.9‰ to −10.0‰). The residence time differences between Wind and Lang's Caves are somewhat proportional to the karst overburden, whereby larger overburden at Lang's Cave relative to Wind Cave (~200 versus 100 m) correlates with longer residence times. It is difficult to compare our residence time estimates with other tropical or ENSO-influenced cave δ¹⁸O_dw sites given that (i) most tropical cave δ¹⁸O_dw monitoring sites are characterized by high rainfall seasonality and seasonal recharge (Beal et al., 2019; Fleitmann et al., 2004; Jones et al., 2000; Kennett et al., 2012; Lases-Hernandez et al., 2019; Mickler et al., 2004; Partin et al., 2012) and (ii) there are no other multiyear tropical cave δ¹⁸O_dw time series that capture ENSO extremes in δ¹⁸O_dw (Chen & Li, 2018; Sun et al., 2018; Zhang & Li, 2019). In tropical caves, high rainfall seasonality combined with short karst residence times generate a seasonal bias in cave δ¹⁸O_dw variability skewed toward the δ¹⁸O_R of the dominant recharge period (Jones et al., 2000; Lases-Hernandez et al., 2019; Partin et al., 2012). Indeed, Baker et al. (2019) documents this behavior across many tropical karst sites with mean annual temperatures >16 °C. However, given that Mulu is not characterized by significant differences in seasonal rainfall with no seasonal bias evident in our cave δ¹⁸O_dw time series, we conclude that our observations represent a departure from the patterns presented in Baker et al. (2019).

We provide evidence that cave δ¹⁸O_dw values represent a linear transformation of δ¹⁸O_R values, with some evidence for variable input from a longer-term reservoir at select drip water sites. Nonstationary behavior in the Lang's cave drip water δ¹⁸O time series reflects variable karst residence times and/or changes in mixing ratios between a fast-responding and long-term reservoir. Nonstationary behavior in the L2 cave δ¹⁸O_dw time series was previously observed by Moerman et al. (2014), potentially from a hydrological extreme. Hydrological extremes impact the hydraulic pressure of karst waters, potentially causing a change in drainage routes and/or aquifer storage (Ford & Williams, 2007). This type of nonstationary behavior is well documented across a variety of latitudes and climates, including rapid infiltrations of tropical cyclone related rainfall (Lases-Hernandez et al., 2019), changes to drip rate in SE Australia from rapid infiltration events (McDonald et al., 2007), and in N. England associated with rapid snowmelt events (Baker & Brunsdon, 2003). In the case of the 2015/2016 El Niño event, four consecutive seasons of enriched δ¹⁸O_R (≤ −5.0‰) from July 2015 to April 2016 period may have contributed to protracted enriched δ¹⁸O_dw values for drip L2s. Alternatively, a significant increase in karst water residence times, perhaps linked to relatively dry conditions during this time, may explain the extended L2 δ¹⁸O_dw enrichment.

The fact that ENSO-related changes in Mulu rainfall δ¹⁸O are recorded in Mulu cave drip water δ¹⁸O time series has important implications for the reconstruction of ENSO in Mulu cave stalagmite δ¹⁸O records. While ENSO anomalies can be resolved in drips fed by sources with residence times as long as ~18 months (Chen et al., 2016), our results demonstrate that the amplitude of ENSO-related cave δ¹⁸O_dw variations is inversely correlated to karst water residence times. Indeed, larger spatial surveys of Mulu cave δ¹⁸O_dw values during ENSO extremes demonstrates that while the distribution of cave δ¹⁸O_dw values shifts significantly during ENSO extremes, the amplitude of the ENSO-related cave δ¹⁸O_dw anomalies differs up to ~6.0‰ across the system during any one time. These conclusions hold true for the subset of Mulu cave drips that feed actively accreting stalagmites (Figure S7). Other studies that capture individual ENSO events in rainfall, cave drip water, soil water, and/or stalagmite δ¹⁸O time series also document responses of different amplitudes (Chen & Li, 2018; Sun et al., 2018). As such, single stalagmite δ¹⁸O records from tropical (Chen et al., 2016; Lachniet et al., 2004) and subtropical (Zhao et al., 2015) regions that aim to reconstruct changes in ENSO variability through time may plausibly reflect changes in karst residence times. Our results indicate that accurate stalagmite δ¹⁸O-based reconstructions of past ENSO-related hydrological variations at a given site require the generation of records from multiple stalagmites at a given site, to rule out the influence of drip site specific changes in residence time. This strategy follows on the success of well-replicated stalagmite δ¹⁸O records of centennial to millennial scale at the Gunung Mulu (Carolin et al., 2013, 2016; Partin et al., 2007; Partin, Cobb, et al., 2013). Unfortunately, fast-growing samples capable of resolving ENSO variability are quite rare (Orland et al., 2014), such that it might be more practical to focus on assessing common trends in ENSO-related variability between reconstruction sites, and between proxy types, than to pursue multiple high-resolution stalagmite δ¹⁸O records from the same site. In that case, proxy system forward models (Dee et al., 2015; Evans et al., 2013) of the rainwater-to-drip water δ¹⁸O transformation at stalagmite δ¹⁸O reconstruction sites would accelerate progress toward the identification and intercomparison of ENSO-related signals in a network of stalagmite δ¹⁸O reconstructions. Likewise, such forward models are critical to the intercomparison of ENSO reconstructions derived from multiple proxy types such as corals (Cobb et al., 2003, 2013; Grothe et al., 2019; McGregor et al., 2013; Tudhope et al., 2001), lake sediments (Conroy et al., 2008), and single forams (Leduc et al., 2009; White et al., 2018).

5 Conclusions

Long-term monitoring of rainwater and drip water δ¹⁸O at Gunung Mulu National Park enables the quantification of karst residence times ranging from ~3 to 18 months. Simple linear transformation of amount-weighted rainfall δ¹⁸O to drip water δ¹⁸O capture ~75–85% of the variance expressed at three drip sites in our study. ENSO extremes are associated with local rainfall δ¹⁸O anomalies of ~6–8‰ and drip water δ¹⁸O anomalies of ~3–5‰, where the magnitude of the peaks correspond to ENSO strength. One of the three long-term drip monitoring sites exhibits nonstationary behavior in karst residence times, implying time-varying contributions from a well-mixed longer-term karst water reservoir. As such, our results suggest that while stalagmite δ¹⁸O records from Gunung Mulu are prime candidates for ENSO reconstruction, multiple stalagmite δ¹⁸O records are required to assess the robustness of such reconstructions. Lastly, our work demonstrates the utility of site-specific, long-term monitoring of rainfall and cave drip water δ¹⁸O to inform the climatic interpretation of stalagmite δ¹⁸O records.

Acknowledgments

We wish to thank Gina Moseley, Nele Meckler, Stacy Carolin, Syria Lejau, Stein-Erik Lauritzen, Robbie Shone, and Jan Berndorff for field work assistance and all of the staff at Gunung Mulu National Park World Heritage for their dedicated assistance in keeping us safe during field expeditions. We also are extremely grateful to the Mulu Meteorological Station staff for faithfully overseeing the collection of daily rainfall samples for the last 12 years. We also acknowledge the Mulu Caves Project for their invaluable information on characterizing the Mulu karst system. We would also like to thank Hussein Sayani for his assistance in water isotope analyses. Permits for this work were granted by the Malaysian Economic Planning Unit, the Sarawak State Planning Unit, and the Sarawak Forestry Department. This work was supported by NSF Grants 0645291 and 1502830 awarded to K. M. C. All data are publicly available on PANGAEA (https://www.pangaea.de), all rainfall data are publicly available on IAEA-WISER (https://nucleus.iaea.org/wiser/index.aspx), and all data are contained in supporting information files associated with this article. The authors acknowledge no conflicts of interest.