Speleothems are remarkable recorders of paleoclimatic conditions as long as their thin calcite laminae can be precisely dated. Conventional geochronological methods used in speleothems include 14C and U-Th radiometric dating1. The radiocarbon method is limited to the last 0.5–50 ka (i.e., about 9 half-life of 14C of ~ 5700 year) and requires corrections for the proportions of dead carbon issued from the dissolution of the carbonate bedrock40–42. Due to recent analytical advances in mass spectrometry, U-Th dating provided accurate chronological constrains of speleothem for the past 500 ka, even with low 238U concentrations1,43,44. However, U-Th dating of "dirty" speleothems, those containing significant detrital components incorporated in the matrix, is problematic due to the contribution of detrital U-series isotopes20,21,45. Diagenetic alterations, including aragonite-to-calcite transformation and calcite-to-calcite recrystallization46, may also severely compromise the accuracy of the U-Th chronology by mobilizing U from the site of diagenesis and leading to an increase in the 230Th/238U isotopic ratio47–49. For these reasons, many dirty speleothems are discarded despite they may provide important paleoclimate and paleoenvironmental information recorded in the detrital fraction. Paleomagnetism offers an interesting alternative for dating dirty speleothems, but such an approach has not been explored hitherto. The principle is based on the capability of magnetic minerals, principally magnetite, issued from the overlying soil/sedimentary cover and carbonate bedrock to settle down in the calcite laminae and align according to the direction of the Earth’s magnetic field. Once calcite precipitation is completed and trapped the magnetic minerals, magnetization locks in almost instantly50. This process provides several advantages in using paleomagnetic directions as a dating tool over U-Th methods, including the fact that magnetic minerals are commonly more resistant to diagenetic alteration than calcite; magnetic measurements are cost-effective, allowing a much more complete and detailed database; and is particularly suitable for dirty speleothems, which exhibit high remanent magnetization due to the relatively high concentration of magnetic particles contained in the detrital fraction. Unlike sediments, the magnetization recorded in speleothem is not affected by inclination flattening due to compaction, as long as samples are taken in the central part of the speleothem, in the horizontal layers, where particle rolling due to gravity is minimum16. Because U-Th is usually expensive, many age-depth models of speleothem are built on a few dating points, with age interpolation in function of the depth obtained by using the algorithm from StalAge software51, assuming a nearly constant growth rate between each U-Th datings. However, growth rates are rarely constant, and this can result in poorly constrained age-depth models (Fig. 4A). Conversely, paleomagnetism can fill this gap by providing a higher resolution dataset. The thickness of the samples depends on how dirty the speleothem is, i.e., on the number of magnetic particles trapped in the calcite laminae to be measurable in a standard magnetometer. Pioneer paleomagnetic studies of speleothem usually used 2 cm cubic samples11,12,14,17. Advances in magnetometer sensitivity now allow to measure the magnetic remanence of speleothem at a millimetric scale, typically of some millimeters (6 mm in this study) up to the centimeter 10,16,52−54, providing high-resolution records of the past Earth’s magnetic field.
The main objective of this study is to test whether a speleothem can be accurately dated based on the comparison of the recorded paleomagnetic directions with a well-known paleomagnetic database. This approach requires that: i) the magnetization is primary and well-defined by principal component analysis; ii) the time window in which the speleothem grew is approximatively known; and that iii) well constrained PSV curves do exist for this time window (coming from global or local paleo-reconstructions).
Analysis of the isothermal remanent magnetization (IRM) curves of the studied samples indicates that detrital/pedogenetic magnetite is the main magnetization carrier (Fig. 2). A small contribution of hematite is observed in the IRM curves and in the demagnetization intensity decay curves, where only 80% of the remanence is cleaned after AF demagnetization at 100 mT (Fig. 3). Thermal demagnetization up to 700ºC shows that the magnetic directions carried by hematite are comparable to those carried by magnetite, suggesting that this hematite is detrital as well. Well-defined and stable demagnetization patterns pointing to origin (Fig. 3) also attest to the primary character of the remanent magnetization.
Because most of the available Holocene paleomagnetic models are limited to the last 10,000 to 14,000 year (the CALS10k.2, the ARCH10k.1, the pfm9k, and the SHA.DIF.14k models), only speleothems younger than 14,000 year can be dated with our approach. The first step is to define the time window into which paleomagnetic data are compared to the reference PSV curve. This can be done by dating the base and the top of the stalagmite, but more U-Th dating can be added to confirm and calibrate the final age-depth model obtained by using the two different approaches (Fig. 4). Here we obtained the age model illustrated in Fig. 4A based on the seven U-series and calibrated 14C anchored ages indicated in Table 1. Based on these results we consider the time interval of 2,000 CE to 7,000 BCE for both SHA.DIF.14k and pfm9k models provide alternative age-depth models based on paleomagnetic data only (Fig. 4B-C), following the approach described in section 4.4. The resulting age-depth model from SHA.DIF.14k provides slightly different but overlapping age intervals than those obtained with U-series and calibrated 14C ages, except for the base and the top of the stalagmite (Fig. 4B). We obtained minimum and maximum ages of 6307 ± 312 BCE and 1443 ± 84 CE, respectively. The shape of the age-depth model using paleomagnetic data exhibits more details and inflections due to a higher number of datapoint (43 paleomagnetic samples against 7 anchored radiometric ages). These inflections should reflect changes in the stalagmite growth rates. The misfit between the age of 1920 CE measured for the more recent calcite layer at the top of the stalagmite and the age of ~ 1440 CE provided by our approach can be explained by the presence of a significant gap in the calcite precipitation rate at the most external (younger) part of the stalagmite, as shown in Fig. S1 of the Supplementary Information. The age of 1920 CE is obtained from the most superficial calcite layers, whereas the last (younger) paleomagnetic specimen has been sampled just below (some mm). The misfit between the ages of the base of the stalagmite remains, however, unresolved.
Our method provides an interesting and promising approach to date dirty speleothem based on paleomagnetic data but has its limitations. First, we assume that the master PSV curve used for comparison is correct and robust, which is not necessarily the case for all time intervals. The differences observed in the PSV models proposed in the literature depend on several factors: the type of data used for their construction (i.e., igneous rock, sediments, archeological artifact) and their spatial and temporal distributions; the dating and measurement uncertainties associated with the different samples and how they are treated in the modeling approach; the statistical analysis used to develop the models; etc. As mentioned above, the CALS10k.237 and the pfm9k38 models include all kinds of paleomagnetic data around the world, including paleomagnetic data from sediments. This results in PSV curves with smoother magnetic inclination than the SHA.DIF.14K model, which includes paleomagnetic data from archeological objects and volcanic rocks only. This smoothness is reflected in the local inclination minimum presented in the SHA.DIF.14k curve around 4000–5000 BCE that is not recorded by the pfm9k model (see inclination curves in the bottom panels of Fig. 4B-C). It is important to note that the time interval from 5000 BCE to 1900 CE is relatively well-document in terms of paleomagnetic data55, whereas some time intervals lack critically reliable paleomagnetic data (ex. from 7000 BCE to 5000 BCE).
We also tested our approach with the pfm9k family models 38 to evaluate how different the results are depending on the choice of the PSV master curve. We used the pfm9k.1a model with the error bars given by the model version 1b (version 1a does not provide error bands). We first synthesized a directional PSV curve for the speleothem coordinates and obtained a new age model by fitting its paleomagnetic profiles to the pfm9k curves illustrated in Fig. 4C. The resulting age-depth model is closely similar to those calculated with the SHA.DIF.14k model and with radiometric data, with a minimum and maximum age of 5388 ± 198 BCE and 1472 ± 84 CE, for the base and the top, respectively. Interestingly, the younger age provided by using the SHA.DIF.14k or the pmf.9k model are strikingly comparable within the 95% probability interval. The older age (base) is younger and still statistically distinct from the age calculated based on radiocarbon analysis, but with lower residual values than those calculated when using the SHA.DIF.14k model (Fig. 5).
Figure 5 illustrates the correlation and residual values from the mean (y = x line in Fig. 5) between the age-depth model calculated with radiometric ages and the age-depth model calculated based on paleomagnetic data using the SHA.DIF.14k and the pmf.9k PSV curves. We discarded the 14C age of 1920 CE of the most external layer due to the presence of the hiatus (Fig. S1) interpreted to be responsible for the misfit with the age of ~ 1440 CE calculated with our approach. We observe a striking correlation between the when using both the SHA.DIF.14k and pmf.9k models. The exception is the age determined at the base of the stalagmite, where the correlation gives a residual value of ~ 500 year for the SHA.DIF.14k model and a residual value of ~-400 year for the pmf.9k model. We interpreted this discrepancy since both models are poorly defined before 4000–5000 BCE.
We also test our approach in the case of another stalagmite from the Algarve area16,52. This stalagmite (called SPAIV) is a middle-Holocene dirty stalagmite that formed during the ~ 4,100–3,300 BCE interval, based on U-series isochron ages (Fig. 6). The age model of this stalagmite was calculated based on 6 U-series isochron ages, and intermediate ages corresponding to each paleomagnetic site-based. The paleomagnetic site-based data corresponds to the mean of the directions obtained from several specimens collected in the same calcite layers. The interpolated ages were obtained using the StalAge algorithm51. In this case, we have used the SHA.DIF.14k paleo-reconstruction to constrain its archaeomagnetic age model. As in the case of the Soprador do Carvalho stalagmite, the age model calculated with paleomagnetic data is closely comparable to the U-Th age model, within their overlapping 95% confidence intervals (Fig. 6). Residual values between both set of data are lower than 200 year (Fig. 5C).