Proxy records of the oxygen isotope ratio of 18O/16O of past precipitation (δ18Op) have played an important role in revealing past hydroclimatic changes, on the basis of global observed relationships between monthly precipitation δ18Op and both precipitation amount and temperature only of a few years as reported by Dansgaard in 1964. It is therefore crucial to systematically re-evaluate the relationships using modern instrumental data. We analysed monthly and annual mean correlations from 108 global stations over the past about 60 years. Consistent with previously reported results, monthly δ18Op values in the high latitudes (≥60°) show a significant positive correlation with temperature (referred to as ‘temperature effect’) and a negative trend with precipitation amount in the low latitudes (≤20°) (‘amount effect’). However, these correlations do not hold true for yearly mean data for more than three-quarters of the stations evaluated. This indicates that the relationships between the different temporal resolutions could be more complicated than previously thought. For the related natural archives, such as ice cores, sediments, and carbonates, further careful evaluation is required to establish the robustness of their paleoclimatic implications.
108 GNIP stations were collected to re-analyse the relationships between δ18Op and climate variables on monthly and interannual timescales.
The relationships between δ18Op and climate variables varied on monthly and interannual timescales.
The results implied that the change of precipitation isotopic values may not be dominated by local climate change, but more likely a response to global atmospheric circulation.
Natural archives enhance the study of prehistoric climate change. The paleoprecipitation data, in particular, inferred from proxy records of the oxygen and hydrogen isotope ratio of 18O/16O (δ18O, Craig 1961), provide important clues for deciphering global hydroclimatic and oceanic circulations over both short-term and orbital timescales. Lines of evidence can be retrieved from various materials, such as the high-latitude and high-altitude ice cores (North Greenland Ice Core Project (NGRIP) Members 2004; EPICA Community Members 2006), the carbonates (Wang et al. 2001, 2005; Yuan et al. 2004; Cruz et al. 2005; Cheng et al. 2016), the tree-rings (McCarroll & Loader 2004; Treydte et al. 2006; Allen et al. 2022), and the sediment cores of lake or peat land (Zhang et al. 2011; Rao et al. 2020). The reliability of these proxies essentially depends on an understanding of modern spatio-temporal precipitation δ18O (δ18Op) variability and its relation to regional and global hydroclimatic processes (Dansgaard 1964; Bowen 2008).
The observed monthly δ18Op (δ18Omp) data has been reported ever since the initial operation of the Global Network of Isotopes in Precipitation (GNIP) in 1961. The analysis of a 2-year (1961–1962 CE) dataset by Dansgaard in 1964 showed a generally negative correlation between δ18Omp values and the monthly precipitation amount (Pm) (‘amount effect’) in the low latitudes. A clearly positive correlation was expressed between δ18Omp values and monthly mean temperature (Tm) (‘temperature effect’) in the high latitudes (Dansgaard 1964). These relationships have been widely used as a fundamental benchmark in paleoclimatic interpretations over the past several decades. The ice-core δ18O data from Antarctica (EPICA Community Members 2006) and Greenland (NGRIP Members 2004), for example, have been taken as an indicator of the past temperature. In the low-latitude regions, paleoprecipitation δ18Op records have been interpreted as a proxy of the monsoonal intensity, and the local or regional precipitation amount (Fleitmann et al. 2003; Tierney et al. 2008).
Subsequently, with the increase of GNIP stations, the spatial distribution of δ18Op and its relationship with climate-related parameters (including the precipitation amount and temperature) on the monthly timescale from 154 stations over the world were re-analysed in 1993 (Rozanski et al. 1993), using a similar approach to Dansgaard (1964). And the results also presented a strong inverse relationship between the mean monthly or annual δ18Op and the amount of monthly precipitation in the tropical marine stations located between 20°S and 20°N. While the correlation of δ18Op and the surface temperature for the stations situated in the northern hemisphere (40°N to 60°N) was different and lower (0.31‰ per °C) than that (0.69‰ per °C) calculated by Dansgaard (1964), and this relationship on the monthly timescale was even non-linear and varied in different stations (Rozanski et al. 1993). In fact, many studies focused on the δ18Op–T relationship in the high latitudes have been conducted, such as Shuman et al. (2001) who presented a temporal δ18Op–T slope by a snow pit approach which is different from the spatial slope of Greenland (Dansgaard 1964); Vinther et al. (2010) provided information on the seasonal aspects of the climate signal in Greenland ice cores, and Sjolte et al. (2011) discussed the δ18Op–T for various timescales in context with the modelling of δ18Op in Greenland. And for the low-latitude areas, based on the observed and model data, Kurita et al. (2009) found the δ18Op–Pm relationship varied substantially from place to place. In short, current research results (Gourcy et al. 2005; Zhang & Wang 2016; Allen et al. 2018; Cai et al. 2018; Yang et al. 2018; Balagizi & Liotta 2019; Falster et al. 2021) all challenged the application of early findings (i.e. ‘temperature effect’ or ‘amount effect’) of Dansgaard (1964) to the interpretation of the paleoclimate records associated with δ18Op.
However, several points could be made regarding the foregoing studies regarding modern relationship between δ18Op and climate parameters (the temperature and precipitation amount). (1) The 2-year spatio-temporal relationships between δ18Omp and Tm or Pm (‘temperature effect’ or ‘amount effect’) is insufficient to decipher the information reserved in the natural archives, like the δ18O (or δD) of ice core, speleothem, sediment of lake, and tree-ring cellulose. (2) This is because the isotopic compositions in these natural archives are not directly inherited from precipitation. For example, the δ18O of speleothem and tree-ring cellulose is indirectly related to δ18Op (Lachniet 2009). Water in the soil and vadose zone could buffer the monthly/seasonal δ18Op variation (Genty et al. 2014; Duan et al. 2016; Sun et al. 2018; Baker et al. 2019; Li et al. 2019). (3) Furthermore, numerous paleo-δ18Op records have been created with resolutions, rather than the monthly to seasonal resolutions (NGRIP Members 2004; Cruz et al. 2005; EPICA Community Members 2006; Cheng et al. 2016). (4) Those subsequent works about δ18Op–T (P) were mostly based on the models or a specific site, which did not have adequately spatially representative. (5) Last but not least, the GNIP data are being updated, which provides an opportunity for re-examining the relationship between δ18Op and local climate-related parameters on the longer timescales or larger data volumes. All of the foregoing factors emphasize the need for re-examination of the relationships between δ18Op and temperature/precipitation on the long-term timescale to better interpret these paleo-δ18O records.
Here, based on the updated data provided by the GNIP, the relationships between δ18Op and temperature in the high-latitude regions and that between δ18Op and precipitation in the low-latitude regions on the monthly and interannual timescales were re-analysed. The objective is to verify whether the ‘temperature effect’ and ‘amount effect’ of monthly scale exist globally with an updated and expanded data size, and whether these relationships hold true for the longer timescale (interannual scale). And it is expected to provide a further understanding of the significance of the paleoprecipitation isotopic records.
DATA AND METHODS
For the analysis method, we used the same method of linear correlation analysis as that of Dansgaard (1964). And we used the MATLAB software to batch process the data to obtain the corresponding slopes, intercepts, correlation coefficients, and significance level values (p), and generate detailed regression graphs. The corresponding maps are generated by the ArcGIS software. All statistical criteria are based on the p values. Significance levels for linear correlations are classified as: significant (p < 0.05), moderately significant (0.05 < p < 0.1), less significant (0.1 < p < 0.2), and non-significant (p > 0.2). Considering the length and readability of this paper, the corresponding slopes, intercepts, correlation coefficients, and significance level values are presented in Supplementary Table S1.
RESULTS AND DISCUSSION
Relationships between δ18Op and temperature at high-latitude regions
Furthermore, among the six stations (five in NH, one in SH) with a significant positive relationship of δ18Op/T, two stations (NH17 and NH43) only had 5 years data (Figure 2(b)), which may influence their results' reliability due to the limitation of the data size. Consequently, there are only four stations from the high latitudes that showed a robust significant positive relationship of δ18Op/T on the interannual timescale. Interestingly, these four stations all present a larger δ18Op/T gradients on the interannual timescale than that on the monthly timescale (Supplementary Table S1). For instance, the slope of δ18Op/T in NH25 station was 0.15‰/°C at the monthly timescale, but 1.12‰/°C at the interannual timescale. Besides, the SH12 station close to the SH13 station (Figures 1 and 6), but their δ18Op/T relationships on the interannual timescale displayed a marked difference. Specifically, the SH13 station showed a significant positive δ18Op/T relationship, with a gradient of 0.69‰/°C. The SH12 station, however, presented a negative δ18Op/T relationship (−0.08‰/°C), although it did not pass the significance test at 5% confidence (Figure 3(b)).
Obviously, the relationship between δ18Op and Ta in the high latitudes was more complicated on the interannual timescale. In practice, there is increasing evidence that the T/δ18Op relationships in polar region vary at multiple timescales and do show spatial heterogeneity (Jouzel et al. 1997; and references therein). For instance, it has been found that there are significant differences between the temperature records from the east coast and the west coast in Greenland (Dansgaard et al. 1975). However, the behaviour of isotopic records from the interior of Greenland did not follow consistently the changes of temperature reserved at either the east or west coast stations (Robin 1983). Furthermore, based on the GCM model simulations, it was demonstrated that the estimated temporal slopes varied over specific regions (Jouzel et al. 1994), which is also presented in this study. And this phenomenon may be attributed to the fact that the variation of δ18Op could be controlled by other factors (Allen et al. 2018; Balagizi & Liotta 2019; Falster et al. 2021), in addition to the local temperature. Interestingly, some studies suggested that the behaviour of δ18Op/T slopes may be associated with the alternating modes of the North Atlantic Oscillation, which is associated with temperature anomalies of opposite sign to the east and west of the crest (Barlow et al. 1993). Besides, some results based on GCM demonstrated the shift of moisture source (e.g. North Atlantic vs. North Pacific) for a Greenland site could generate a local δ18Op anomaly of ∼7‰, under a local climate condition without significant change (Charles et al. 1994).
Relationships between δ18Op and precipitation amount in low latitudes
In the low latitudes, the δ18Omp values in 70 of the 77 stations showed a significant negative correlation with the Pm, although six stations do not show a significant negative correlation, and even one station (KOZHIKODE) at the south of Indian peninsula shows a significant positive correlation (Figures 1 and 7(a); Supplementary Table S1). As can be seen from Figure 1, these abnormal stations are all located in the coastal areas of the mainland, which may be affected by the tropical cyclones (typhoons and hurricanes) in the surrounding seas. The source of typhoon generation, different paths, the location of the typhoon where the precipitation is located and the frequency of typhoons, will all affect the regional δ18Op variability (Lawrence & Gedzelman 1996; Ohsawa & Yusa 2000; Lawrence et al. 2004; Fudeyasu et al. 2008; Xu et al. 2019; Sun et al. 2022). Moreover, the coasts of the continent will also be affected by the marine environment, such as the ocean currents. On the whole, mostly stations δ18Omp in the low latitudes present a negative correlation with the Pm. This indicates the δ18Omp values are mainly governed by an ‘amount effect’ in the low latitudes on the monthly timescale, supporting the previous conclusions with 2-year observations reported in 1964 (Dansgaard 1964).
Interestingly, the correlation between δ18Omp and Pm of KOZHIKODE station in south India shows a significant positive correlation (Figure 1). Contrary to the results on the monthly timescale, the δ18Owa is a significant negative correlation with the Pa on the interannual timescale. Similarly, WELLAMPITIYA at Sri Lanka, which is near KOZHIKODE, is also non-significantly negative on the monthly timescale, but significant in the recheck results on the interannual timescale (Figure 8), although the interannual results of raw data (Figure 6) show an insignificant negative correlation, which may be caused by some years in the statistical raw data missing some months data. In fact, the correlation between δ18Op and P at other sites in the Indian subcontinent affected by the Indian monsoon is also more significantly negative on the interannual scale than on the monthly timescale (Li et al. 2015). It may be a regional feature of the Indian subcontinent and requires further exploration for the underlying mechanism.
In summary, in the high latitudes, the relationship between δ18Op and T is more spatially consistent and significantly positively correlated on the monthly timescale, but it does not show such consistency and significantly positive correlation on the interannual timescale. Similarly, in the low latitudes, the relationship between δ18Op and P on the monthly timescale is more significant than that on the interannual timescale. So, it would not be fully appropriate to use the δ18Op records as indicators of paleotemperature in the high latitudes and paleorainfall amount in the low latitudes, at least on the interannual timescale.
Perspective from the paleo-δ18Op records
Over the past decades, the δ18O records of high-latitude ice cores have been used as a paleotemperature indicator (NGRIP Members 2004; EPICA Community Members 2006). However, this has been widely challenged. Modern observations (Johnsen et al. 1989), model simulations (Charles et al. 1994), and deuterium excess evidence (Masson-Delmotte et al. 2005a, 2005b) have highlighted the influence of the moisture sources and atmospheric circulation on ice core δ18O data. Previous studies also revealed the impacts of the tropical temperatures (Boyle 1997), precipitation seasonality (Werner et al. 2001), North Atlantic Oscillation and sea surface temperature (White et al. 1997), and shifts in vapour sources (Cole et al. 1999). In addition, inferred temperature variation discrepancies between the high-latitude ice core δ18O and other proxies, such as the borehole paleothermometry (Cuffey et al. 1995), and δ40Ar and δ15N of gases trapped in the ice (Landais et al. 2004; Huber et al. 2006; Buizert et al. 2014), have also been reported.
In the low latitudes, the stalagmite δ18O records from southern America (Wang et al. 2007) and leaf-wax compound-specific δD records from the lacustrine sediments from lacustrine (Tierney et al. 2008) and the marine sediments (Schefuß et al. 2005, 2011) from tropical Africa have been used in the reconstructions of the past precipitation amounts for the last climatic cycle. Some studies, however, showed that not only the precipitation amount, but also the influences from atmospheric circulation (Vuille et al. 2003) and water vapour sources (Lewis et al. 2010) controlled the precipitation isotopes in the low latitudes. For the Asian monsoon region, some simulations (Caley et al. 2014; Zhang & Wang 2016; Cai et al. 2018; Yang et al. 2018) suggest the effect of the hydrologic processes and the atmospheric circulation on the Asian stalagmite δ18O records.
Finally, the discrepancy among monthly (Figure 1), interannual (Figures 6 and 8), and multidecadal to centennial/millennial correlations from the paleo-records (Figure 9) directly challenges the hypothesis that precipitation isotopes are mainly governed by a ‘temperature effect’ in the high latitudes and an ‘amount effect’ in the low latitudes. This could be attributed to the scale difference of controlling factors between climatic variables and isotopic value (Sjolte et al. 2011). Therefore, it might not be fully appropriate to interpret the paleoprecipitation isotope records just based on the relationships at the monthly timescale or even shorter timescale (like daily and hourly).
An open question then is at what timescale the modern relationship is the most suitable for interpreting the geological records, and whether it is feasible at such long timescale to implement modern analysis? The classic geological principle of ‘The present is the key to the past’ is commonly used in the studies of past earth environmental reconstruction, and we support this research method without any doubt. However, when it is not that powerful to interpret proxies by using modern relations, the model simulation and replication test between paleo-records could complement the rationality of the interpretation (Dorale & Liu 2009; Liu et al. 2014). That is, based on the relationships obtained by the modern monitoring, a model simulation can be preliminarily constructed. When the reliability of reconstruction records is assured by the replication test, the paleoclimate reconstruction records could be used to verify the accuracy of the model simulation results and improve the designment of the models. And then the model simulations could provide more information and mechanistic explanations, which modern monitoring and paleoclimate reconstruction could not provide or solve. However, the importance of the modern monitoring and the paleoclimate reconstruction cannot be denied, and should be subject to the same attention as the model simulations, because they are a basic, critical component of the relevant research item. In conclusion, more modern observations, model simulations, and paleoclimatic reconstructions (but not confined) may help to clarify the enigmatic nature of precipitation isotope ratios in different natural archives and their potential applications in the paleoclimatic studies.
108 stations at the high latitudes (≥60°) and the low latitudes (≤20°) in both hemispheres covering the past about 60 years have been selected from GNIP. And the linear correlations between δ18Op and climatic variables (temperature and precipitation amount) from these stations, on the monthly timescale and interannual timescale, respectively, have been conducted. The results show: (1) The monthly δ18Op values show a significant positive correlation with Tm for all stations in the high latitudes (≥60°), and a negative correlation with precipitation amount in the low latitudes (≤20°), which is consistent with previous results. (2) Whereas on the interannual timescale, the relationships between δ18Owa and Ta in the high latitudes, and between δ18Owa and Pa in the low latitudes, are not as strong and spatially uniform as those on the monthly timescale. (3) And the paleo-records associated with the precipitation isotopic composition presented broadly global similarities and were synchronous during the abrupt climate events since the last glacial, implying the change of precipitation isotopic values may not be dominated by the local climate change, but more likely a response to the global atmospheric circulation. (4) Lines of evidence indicate the relationships between the different timescales could be different and more complicated than previously thought. To establish the robustness of paleoclimatic implications of the related natural archives (e.g. ice cores, sediments, and carbonates), further modern observations, model simulations, and paleoclimatic reconstructions are required and should complement one another.
This work was jointly supported by the National Natural Science Foundation of China (No. 42001080, 42171156 and 42106228), the Natural Science Foundation of Hunan Province China (No. 2021JJ40349), Research Foundation of the Department of Natural Resources of Hunan Province (No. 20230148ST), and Hunan Education Department Project (No. 20B351). The original data provided by GNIP are gratefully acknowledged.
DATA AVAILABILITY STATEMENT
All relevant data are included in the paper or its Supplementary Information.
CONFLICT OF INTEREST
The authors declare there is no conflict.