Tropical cloud forest climate variability and the demise of the Monteverde golden toad

Abstract

Widespread amphibian extinctions in the mountains of the American tropics have been blamed on the interaction of anthropogenic climate change and a lethal pathogen. However, limited meteorological records make it difficult to conclude whether current climate conditions at these sites are actually exceptional in the context of natural variability. We use stable oxygen isotope measurements from trees without annual rings to reconstruct a century of hydroclimatology in the Monteverde Cloud Forest of Costa Rica. High-resolution measurements reveal coherent isotope cycles that provide annual chronological control and paleoclimate information. Climate variability is dominated by interannual variance in dry season moisture associated with El Niño Southern Oscillation events. There is no evidence of a trend associated with global warming. Rather, the extinction of the Monteverde golden toad (Bufo periglenes) appears to have coincided with an exceptionally dry interval caused by the 1986–1987 El Niño event.

The demise of the Monteverde golden toad (Bufo periglenes) in the montane cloud forest of Costa Rica in 1987–1988, as well as subsequent amphibian extinctions throughout the American tropics, has been believed to be a consequence of the interaction of global warming (1, 2) and the introduced chytrid fungus Batrachochytrium dendrobatidis (3–5). Analysis of the limited available weather data from Monteverde has suggested that a trend toward decreasing immersion cover since the late 1970s reflects the influence of increasing tropical air and sea surface temperatures (SST) (1, 2). Pounds and coauthors (1) argued that this pattern of increasing dry days implicated rising global temperatures due to anthropogenic greenhouse gas emissions. Nair et al. (6) and Lawton et al. (7) used a regional atmospheric model to demonstrate that deforestation in the tropical lowland forests upwind from the Monteverde Cloud Forest could also potentially result in elevated cloud base height and drier conditions.

Subsequent examination of a larger dataset of the last year of observation (LYO) of extinct neotropical amphibians in the genus Atelopus found a positive, lagged correlation between mean annual tropical (30 °N to 30 °S) temperatures and the timing of Atelopus extinctions (2). The authors hypothesized that temperatures at many montane locations in the Americas were shifting towards an “optimum range” for the growth of the chytrid fungus. This “chytrid-thermal-optimum hypothesis” suggested that tropical temperature trends associated with anthropogenic global warming were responsible for widespread amphibian extinctions due to chytridiomycosis.

This “climate-linked epidemic hypothesis” (2) postulates that global warming contributes to the development of optimal environmental conditions for the fungus. Upward trends in global temperatures could therefore already be fundamentally altering the suite of climatic conditions that maintain mountain forest ecosystems in Central America. General circulation model simulations of climate under doubled CO₂ conditions predict higher lifting condensation levels and reduced cloud contact for tropical montane cloud forests as a result of increasing temperatures (8). Imprinted on the apparent instrumental drying trend at Monteverde, El Niño events also cause local increases in temperature and reductions in cloud cover and moisture (9). However, the role of climate change in neotropical anuran extinctions has been questioned on statistical grounds (10, 11). Without the context provided by long-term climate records, it is difficult to confidently conclude whether the extinction of the golden toad and other changes in cloud forest ecology are the consequence of anthropogenic climate forcing (2), land-surface feedbacks due to deforestation (6, 7), or natural variability in tropical climate (9).

Approach

We developed annual proxy records of climate variability from tropical montane cloud forest trees without annual rings at Monteverde (10.2 °N, 85.35 °W, 1500–1800 m) over the last century using stable isotopes (12, 13). The advantage of this approach is that it does not rely on the formation of yearly morphological growth hiatuses in trees, which in the tropics can often be absent (Fig. 1A), but rather takes advantage of the annual variation in the stable oxygen isotope ratio (δ¹⁸O) of water used by plants over the course of a year. This technique can also be applied where posited “growth rings” cannot be shown to be annual. Annual oxygen isotope cycles have been identified in lowland tropical forests (12, 14), and at Monteverde we have previously demonstrated that variation between the summer wet season and the cloud-dominated winter dry season is sufficient to induce an annual δ¹⁸O cycle along the radial xylem growth of cloud forest trees (13). The δ¹⁸O of cellulose in cloud forest trees reflects the seasonal change in the δ¹⁸O of source water as determined by the amount of rainfall and the ¹⁸O-enrichment of cloud water. On interannual time scales, departures from the mean annual cycle amplitude will result from anomalous rainfall and changes in temperature, relative humidity, and evapotranspiration (13, 15). At Monteverde, these are further related to changes in the intensity of moisture advection over the continental divide into the Pacific slope forests and the amount of cloud cover. Interannual changes in the annual δ¹⁸O cycle maxima can therefore be interpreted as moisture changes during the boreal winter dry season (approximately January to April).

Fig. 1.

Oxygen isotope cycles in Pouteria at Monteverde. (A) Unlike most temperate region trees (for example, the Pinus strobus from North Carolina shown here), Pouteria from Monteverde lack any identifiable annual rings. However, stable oxygen isotope measurements on two Pouteria trees reveal consistent annual cycles in the δ¹⁸O data sampled at constant increments from the cambium in samples MV12 (B) and MV15 (C). Seven radiocarbon dates are shown by red triangles. Further ¹⁴C calibration details are available in SI Text. Note that the two cores do not span the same exact time period (see Mathods and Materials). (D) Monthly age modeled δ¹⁸O series from MV12 (red) and MV15 (blue) as a function of time, with lighter colors showing ± 2σ analytical error.

We made high-resolution (200 μm increment) stable oxygen isotope ratio measurements along the radial growth axis of two Pouteria, mature canopy trees in the Sapotaceae family growing in the cloud forest on the Pacific slope below the continental divide at Monteverde, using a unique online induction pyrolysis system (16). The resulting δ¹⁸O data series have coherent annual oxygen isotope cycles we use for chronological control (Fig. 1 B and C) in the absence of rings. We developed an age model for both series by assigning the maxima of each discrete cycle to the dry season of the corresponding year and interpolating the δ¹⁸O data series between these tie points to monthly resolution (13) Fig. 1(D). Radiocarbon assays confirm our age modeled δ¹⁸O time series, which spans the period 1900 to 2002, with an estimated overall age model error of ± 2 years (details of the radiocarbon analysis are available as SI Text).

Results

Annual δ¹⁸O cycles show a mean amplitude of 2.16‰ (Fig. 1 B–D; Fig. 2A), with an average magnitude of year-to-year variability of 50% of this annual cycle (σ = 1.08‰). We observe significant correlations consistent with pilot studies and the interpretive hypothesis for annual and interannual variability described above. Collectively, dry and wet season temperature and precipitation, dry season zonal wind speed, and dry season relative humidity explain 40% of the annual cycle amplitude variability in our δ¹⁸O time series over the instrumental period at Monteverde. These empirical correlations are also consistent with expectations from independent, mechanistic modeling (15, 17) of expected cellulose δ¹⁸O as a function of climate (Fig. S1 and SI Text). Dry season amplitude anomalies also show a clear relationship to interannual climate variability in sea surface temperature anomalies (18) and zonal wind velocity (19) (Fig. 2 B and C). El Niño Southern Oscillation (ENSO) years in the δ¹⁸O chronology show a isotopic cycle enrichment 0.8‰ (greater than 1σ) above the mean (Fig. 2A), with a strong mean positive anomaly of 2.0‰ for 1983, 1987, and 1998 alone (greater than 2σ above the mean). These years of discernible strong positive anomalies in the monthly δ¹⁸O chronology (Fig. 3A) correspond to periods of dry season drought (greater number of consecutive dry days), negative dry season precipitation anomalies, and temperature anomalies of a degree Celsius or more (1). ENSO years have significantly higher δ¹⁸O values in the dry and early wet season (January through July, one-tailed paired t test, p < 0.01). In addition to dry season variability associated with these strong El Niño events, there are larger mean positive δ¹⁸O anomalies of ∼1.8‰ in years of substantially warmer tropical Atlantic and Western Hemisphere warm pool (WHWP) temperatures (20). The anomalous peak (greater than 2σ above the mean) δ¹⁸O associated with warm WHWP events indicates a later onset of the rainy season and the effective moisture deficit persists for several months. The end of the dry season and early wet season (May through July) are significantly higher when WHWP events (n = 4) are considered alone (one-tailed paired t test, p < 0.01), consistent with the association of the largest magnitude WHWP warm events with El Niño events that have persistent boreal spring SST anomalies (20).

Fig. 2.

Association between Monteverde δ¹⁸O records and regional/global climate. (A) Annual δ¹⁸O cycles are clearly enriched during warm ENSO events (red line; 1905, 1926, 1942, 1952, 1958, 1966, 1969, 1978, 1983, 1987, 1992, and 1998) and extensive WHWP (blue line; 1958, 1969, 1987, 1998) years (20) (A), relative to the overall mean (black line). ± 1σ confidence intervals are shown by same-color shading. (B) The 10 most enriched years in the isotope chronology are associated with an average dry season (January-April) SST anomaly (18) in the eastern Pacific of greater than 1 °C. (C) Similarly, positive dry season (January-April) δ¹⁸O anomalies are associated with weaker 850 mb tradewinds in both the tropical Atlantic and Pacific (19).

Fig. 3.

Composite isotope time series and inferred moisture anomalies at Monteverde. (A) The composite monthly δ¹⁸O time series shows coherent annual isotope cycles back to 1900. The composite series was developed by averaging the two δ¹⁸O series after first dilating the variance of MV15 to that of MV12 as estimated over their common 253 month overlap. Differences in the long-term mean and common period variance of the two series are only slightly higher (0.35 and 0.41‰) than analytical precision (0.28‰). Gray shading indicates the full envelope of additive uncertainty due to analytical measurement precision and the observed differences in variance and mean between the overlapping section of the cores. Since these three different elements of total uncertainty are not independent, this envelope is likely conservative. δ¹⁸O enrichment associated with the 1986–1987 ENSO/WHWP event is highlighted, with horizontal dashed lines in (A) indicating the most likely range of dry season isotopic maxima for 1987, given uncertainties. Radiocarbon dates and their uncertainty are shown by filled triangles and their associated horizontal bars. (B) Monthly δ¹⁸O anomalies from the mean annual cycle emphasize that the time period associated with the 1986–1987 warm ENSO event was one of the driest of the last century. (C) Multidecadal variability in δ¹⁸O (SSA filtered, M = 20 years) shows changes in the magnitude of δ¹⁸O seasonality. (D) Deforestation rates in Costa Rica’s Atlantic forests upwind from Monteverde (30).

A weakening of local and equatorial Pacific 850 mb (∼1,500 m) easterly winds appears to consistently accompany the δ¹⁸O anomalies over the length of the reanalysis period (19) (Fig. 2C). Those years with anomalously high maximum δ¹⁸O values occur simultaneously with dry season westerly zonal wind anomalies. Because the slackening of the northeasterly tradewinds at Monteverde will influence moisture advection over the continental divide, the relationship between the dry season amplitude and local zonal wind anomalies from reanalysis data also accounts for the largest portion of the total variance in δ¹⁸O amplitude of any single local climate variable (r = 0.53, p = 0.01) (Fig. 2C). Higher temperature and lower relative humidity display similar associations with the isotope chronology and likewise covary with ENSO and WHWP events. Anomalous δ¹⁸O peaks that occur during warm ENSO and WHWP events coincide with increased Monteverde temperature, weaker zonal winds, and reduced relative humidity and precipitation; however, smaller δ¹⁸O dry season anomalies can also occur in ENSO neutral years when only one or two of these climatic factors is anomalous.

A distinct change in the amplitude of the annual δ¹⁸O cycle occurs in the middle or late 1960s (Fig. 1D; Fig. 3 A and C) and is related to a change in both the wet season minima and the dry season amplitude, suggesting an overall damped seasonality, reduced summer rainfall, and slightly increased dry season moisture for the period from the mid-1930s through the 1960s. There is support for this interpretation from the limited climate station data from the Pacific side of northwestern Costa Rica that spans the period of the transition, which shows an increase in precipitation between 1960 and 1970 (21, 22). The previous Monteverde station, downslope from the cloud forest (1,460 m elevation), shows an increase in average summer precipitation of approximately 200 mm over the 1960s, which could translate into a change in the δ¹⁸O of rainy season precipitation of as much as ∼1.2‰ (13, 23), approximately the same as observed in our isotope chronology. However, the association between the δ¹⁸O time series amplitude and the concurrent zonal wind field retains its spatial pattern and magnitude, reinforcing our interpretation that the most consistent and proximal control on dry season moisture at Monteverde is the strength of the easterly winds. The change to high amplitude δ¹⁸O cycles in the late 1960s coincides with a shift in the sign of the summer North Atlantic Oscillation (NAO) index (24), an increase in interannual band (ENSO) variance in the tropical Pacific (25–27), a change in the seasonality of ENSO (28), and with other rapid changes in patterns of global atmosphere circulation, particularly in the tropical and Atlantic regions (24). Positive phases of the NAO, such as observed after the 1960s, are associated with positive precipitation anomalies in the early wet season in Central America (29). At the same time, deforestation rates in the Atlantic coast forests of Costa Rica increased (Fig. 3D) (30) and could have contributed to mesoscale alterations of boundary layer interactions sufficient to influence downwind dry season cloud formation (6, 7).

We empirically identify temporal patterns in our δ¹⁸O time series using singular spectrum analysis (SSA) (31–33). The identification of interpreted spectral components is robust to the choice of embedding dimension. Leading modes contain annual and 2–6 yr periodicities and explain 34% and 12% of total monthly signal variance, respectively. ENSO band frequencies contain 34% of the power at annual resolution. Only one significant low order eigenvector is dominated by multidecadal variance, with a prominent spectral peak between 20 and 30 yr (Fig. 3C). This reflects changes in the annual isotope amplitude, or seasonality, since we determined that removing the annual cycle prior to SSA causes this component to disappear entirely. A monthly anomaly series (Fig. 3B) is recovered by calculating the difference between each monthly δ¹⁸O value and the overall mean (n = 102) δ¹⁸O “climatology” for the corresponding month.

Discussion

Causes of the Demise of the Monteverde Golden Toad.

The early study of the relationship between the hydroclimatology of Monteverde and ENSO events by Pounds and Crump (9) noted the strong drying which occurred during the 1986–1987 El Niño event and identified higher temperatures and lower seasonal rainfall as well during the prior (1982–1983) El Niño. In that case, warmer Pacific SSTs were interpreted to correspond to warmer, drier conditions in the cloud forest. Based on statistical hypothesis testing, Pounds and coauthors (34) subsequently concluded that the abnormal dry conditions of 1987 were likely to have caused the multispecies population crash, which included the extinction of the Monteverde Golden Toad. Both of these papers speculated at the time that disease may have been the proximate cause of the extinction of the golden toad (9, 34). Chytridiomycosis, caused by the chytrid fungus Batrachochytrium dendrobatidis, was eventually identified as a major cause of amphibian extinction throughout the Americas, Europe, Australia, and New Zealand (3–5).

The “chytrid-thermal-optimum hypothesis” suggested that tropical temperature trends associated with anthropogenic global warming were responsible for widespread amphibian extinctions due to chytridiomycosis (2), and the earlier work (1) associated the extinction of the golden toad with dry conditions at Monteverde. Lips et al. (11) tested the robustness of the statistical comparison between tropical temperatures and Atelopus LYO data, concluding that uncertainties in the choice of anuran observational data were sufficient to call into question the relationship. These authors instead sought to demonstrate that spatial patterns of extinction due to chytridiomycosis could be related to the spread of the fungus from a few centers of initial infection. Rohr et al (2008) (10) questioned both conclusions and stated that “Almost all of our findings are contrary to the predictions of the ‘chytrid-thermal-optimum hypothesis.’ ” They also found that there was no stable linear relationship between the diurnal temperature range and modeled growth of the chytrid fungus. Significant correlations were found, however, between modeled chytrid fungus mortality and decreases in cloud cover (10).

The significant drying trend in the number of days without rainfall identified by Pounds and coauthors (1, 2) is a result of the rapid increase in the number of annual dry days beginning with the strong 1982–1983 El Niño and the occurrence of long periods of drought during the 1998 dry season, also associated with ENSO. This trend is sensitive to the range of years and the threshold of consecutive days utilized (see Fig. S2 and SI Text). We have also determined that, due to the limited length of the record, there is no statistically significant difference [p≥0.26 (35)] in the correlation between the LYO for Atelopus and annual tropical temperatures (r = 0.76, p = 0.001), the Atlantic Multidecadal Oscillation index (r = 0.54, p = 0.04) (18, 36), or average spring-summer sea surface temperatures in the WHWP (r = 0.48, p = 0.06) (20).

Contrary to interpretations of the short instrumental record (1), no long-term trend in dry season hydroclimatology can be inferred from our δ¹⁸O time series at Monteverde (1900–2002, p = 0.50; 1973–2000, p = 0.31). Rather, variability at the interannual scale dominates the isotope signal, particularly during the period of increased ENSO variance since the late 1960s. The years 1983, 1987, and 1998 are the three driest winters in the last three decades (Fig. 3A). Collectively, the period from 1986 to 1988, associated with a strong El Niño event and an extensive WHWP that persisted into boreal late spring-summer, appears to be one of the driest in the last three decades and is amongst the driest of the last century (Fig. 3B). The strong warm ENSO event of 1997–1998 was also followed by demographic shifts in anuran populations at Monteverde (1). There is currently no consensus on how anthropogenic climate change will influence the El Niño Southern Oscillation (37, 38) and ENSO anomalies in the most recent decades are not beyond the range of natural variability during the instrumental period (39). This suggests that the cause of the specific and well-documented extinction of the Monteverde golden toad was the combination of the abnormally strong ENSO-forced dryness (9) and the lethality (3) of the introduced chytrid fungus, but was not directly mediated by anthropogenic temperature trends, a finding from paleoclimatology that is in agreement with statistical reanalysis (10, 11) of the “climate-linked epidemic hypothesis.”

In light of these results, the early hypotheses by Pounds and coauthors about the potential role of warm ENSO events in driving recent ecological change at Monteverde now appear prescient (9, 34). Our δ¹⁸O chronology from Monteverde allows us to conclude that one of the longest driest periods in the last 100 yr occurred during the 1986–1987 El Niño. It is possible that Batrachochytrium dendrobatidis was already present in Monteverde prior to that year (40). Dry conditions culminating during the reproductive season in 1987 would have caused amphibian populations to coalesce around a few remaining wet microhabitats (1, 34, 41), where the fungus would have been able to spread rapidly through the population and ultimately result in extinction.

Sources of Uncertainty.

While high-resolution oxygen isotope measurements allow us to resolve the annual cycle of source water δ¹⁸O and isotopic enrichment in cloud forest trees, and to therefore establish chronological control even in the absence of annual rings, there are a number of uncertainties that necessitate cautious interpretation of the chronological and climatic information in our δ¹⁸O time series. Our approach to developing records of past climate from tropical trees (12) has a closer methodological and procedural affinity to paleoclimate analysis using speleothems or corals than to classical dendrochronology. Although chronological uncertainty in this study is minimized through the combination of the identification of annual stable isotope cycles (13) and radiometric dating, we do not achieve the age model precision available from traditional approaches to dendrochronology that use large sample sizes and robust statistical pattern matching (42), nor can we be certain we have identified “missing” years. While the δ¹⁸O signal of cellulose from tree rings will be dominated by the enrichment due to transpiration and the source water signal associated with the amount effect (43), the latter accounts for only a portion of the total variability in precipitation δ¹⁸O (13) and the former could be influenced by several additional environmental and physiological factors; only the effects of precipitation, temperature, humidity, and wind are accounted for directly in our analysis. We have found that the observed annual and interannual δ¹⁸O variability in our Pouteria trees match those expected from both our hypothesis and mechanistic simulations. However, additional replication is necessary to place more confidence in estimates of multidecadal variability.

On the Role of Climate and Disease in Extinction Events.

Epidemic-driven extinction may be the result of a series of biotic and abiotic interactions across spatial and temporal scales (11, 44). Drought or pluvial conditions associated with strong ENSO events are known to be associated with localized extinction (45, 46). In Yellowstone National Park, amphibian declines have been linked to the combination of decreasing precipitation, increasing temperatures, and wetland desiccation, which affects populations by contributing to increasing mortality, decreased migration, and reduced opportunities for wetland colonization (47). The decline of amphibians in El Yunque Forest in Puerto Rico is believed to be a consequence of a change in their behavior during dry periods, with populations moving from a dispersed distribution to a few protected microsites on the landscape, increasing their vulnerability to contagion (48). Similar patterns were observed at Monteverde prior to the multispecies population crash and extinction of the golden toad in 1987 (1, 9, 34, 41). Extinction events are an omnipresent feature of the history of life (49). Hence, while climate anomalies are not a feature of all extinctions (11, 44), it is not unreasonable to expect that natural climate variability can interact with species life history and ecological community and population dynamics to contribute to extinctions, even as anthropogenic climate change continues to develop. Indeed, future increases in tropical temperatures, in conjunction with interannual or decadal variability, may exacerbate stresses on tropical montane ecosystems.

Materials and Methods

Stable Oxygen Isotopes.

Increment cores from two individual mature Sapotaceae (Pouteria) were collected at an elevation of 1,560 (MV12) and 1,580 m (MV15). MV12 was struck by lightning and subsequently felled in 2002. Radiocarbon measurements on MV15 revealed that it had an outer wood date more recent than 1960. The cores were subsampled in the laboratory at continuous 200 μm increments using a rotary microtome. The raw wood samples were chemically processed to α-cellulose using the Brendel method (50, 51) as modified for high-resolution sampling (12), and 300–350 μg of α-cellulose were wrapped in silver capsules and converted online to CO in a Costech High Temperature Generator/Elemental Combustion System (HTG/ECS) (16). At the time these analyses were conducted, this was the minimum sample size necessary to achieve acceptable measurement precision, and therefore determined the sampling increment on the 5 mm diameter cores collected from our Pouteria trees from Monteverde. The HTG uses a 1 MHz radio frequency induction heater to rapidly raise the temperature of a thin molybdenum crucible susceptor packed with graphite to > 1,500 °C for highly efficient pyrolysis (16). Isotopic measurements were made using a Costech induction heater coupled to a ThermoFinnigan Delta XP continuous flow isotope ratio mass spectrometer (16). Measurement precision on a total of 320 interspersed Sigma Alpha Cellulose solid standards was 0.28‰.

Radiocarbon and Chronology Validation.

Seven samples from the two cores were selected for radiocarbon analysis to test the δ¹⁸O-based age model. We estimate that maximum age model error for the earliest part of our chronology to be ± 1 yr, based on the range of realistic age models as constrained by the 2σ resolution of the post-1955 radiocarbon dates. The older (1900–1949) portion of the chronology has an estimated maximum age model error of ± 2 yr due to slower growth rates and lower radiocarbon precision. The actual calendar dates that could be realistically associated with the radiocarbon assays, particularly on MV15, can be better constrained using Bayesian probability estimates between the radiocarbon calibration curve and the sample measurements (52–54). The likely dates are restricted foremost by the growth orientation of the tree—that is, samples from near the bark must be more recently formed than those near the center of tree—and secondly using the number of annual cycles between each Δ¹⁴C measurement, in order to determine the most probable interval spanned by the core. One radiocarbon age from MV15 is post-bomb (AA776775) and provides an anchor for constraining the remaining three prebomb dates. Within these temporal, self-consistent constraints, the radiocarbon measurements and their individual probability density functions are then matched against the calibration curve and a posteriori probability functions calculated. Applying Bayesian analysis and these additional restrictions reveals that the core from MV15 spans the period 1900–1969 (Table S1). As an independent check on chronology development from both cycle counting and radiocarbon dating, we cross-correlated the “floating” MV15 sequence with the securely dated MV12 chronology. Even without additional age control constraints from Δ¹⁴C, a span of 1900–1969 for MV15 has the largest cross-correlation with MV12 for the monthly δ¹⁸O values [r = 0.43, p = 0.016, (55)], the monthly mean (r = 0.44, p < 0.01), and the monthly minimum (r = 0.65, p < 0.01) δ¹⁸O, which mirrors the independent high-resolution radiocarbon dating described above.