Romy Meyer

Our planet is surrounded by a magnetic field, acting like a huge shield against harmful solar and cosmic radiation that, if it would not be deflected by the Earth’s magnetic field, may damage the atmosphere and disrupt communication systems (Boteler et al., 1998; Courtillot et al., 2007). The Earth’s magnetic field can be described with a direction and a strength, and both are constantly changing through time. The research field studying the past variations in the Earth’s magnetic field is known as paleomagnetism. By using archaeological materials, volcanic rocks or sediments one can determine the past field value at that location at a specific moment in time. Currently, the Earth’s magnetic field is characterised by a region of weak field strength above South America, named the South Atlantic Anomaly (SAA) (Fig. 1). This low intensity anomaly is one of the most intriguing features of today’s geomagnetic field. Its present-day behaviour is studied using satellite data but much about the past behaviour of the anomaly is unknown (Campuzano et al., 2019; Engbers et al., 2022; Nilsson et al., 2022). In this thesis, I first focus on studying the reliability of volcanic rocks as paleomagnetic recorders using data from Mt. Etna and La Palma, and then I study the behaviour of the SAA by obtaining paleomagnetic data from three different locations on Earth: the Revillagigedo Archipelago, La Réunion and the island of Bali in Indonesia (Fig. 1).

When averaged over a sufficient amount of time the Earth’s magnetic field can be approximated as a geocentric axial dipole (GAD), yet it is continuously changing through time and space. These variations, known as paleosecular variation (PSV), operate on timescales ranging from seconds to millions of years. On the longer timescale, the most prominent expression of field variability is the alternation between periods of stable polarity (chrons or superchrons) and geomagnetic reversals. For example, during the Cretaceous the field maintained a normal polarity for 37 million years before switching polarity in a geomagnetic reversal. Even during such long intervals of stable polarity, the field still exhibits secular variation in direction and intensity. During a geomagnetic reversal the magnetic north and south pole of the Earth switch. Magnetic reversals have occurred very irregularly in the past but on average around every 200,000 years (Ogg, 2020). The last reversal, the Matuyama-Brunhes reversal, took place about 780,000 years ago. The time for a complete reversal to take place may be in the range of hundred or thousands of years (Mahgoub et al., 2023; Sagnotti et al., 2014, 2016; van Grinsven et al., 2025).

In contrast to full reversals, Earth’s magnetic field also shows excursions, which are mostly only a temporary collapse of the field with directions deviating >45◦ degrees from the geographic poles. They did occur frequently since the last reversal, an example is the Laschamps excursion around 41,000 years ago when Earth’s magnetic field weakened and the poles briefly switched (Guillou et al., 2004). On even shorter timescales geomagnetic jerks occur: sudden, brief changes of the field lasting hundreds of years, or even less. On top of this there is the everlasting secular variation: constant variations in declination, inclination and intensity that make the poles wander around the geographic poles. Secular variation also leads to the variations in declination, which is the angle between the magnetic north and the geographic north, that requires corrections to compass readings depending on where on Earth you are. Secular variation of directions and intensity is measured by magnetic observatories worldwide and by satellites in space. These direct measurements reveal a very interesting feature of today’s Earth magnetic field: the South Atlantic Anomaly (SAA).

The SAA is a region of anomalously low field strength located above South-America and the South-Atlantic, and is typically defined as the area where the surface magnetic intensity is below 32 µT (Pavón-Carrasco & De Santis, 2016). In this region, the Van Allen radiation belts are located at a lower altitude than usual (Domingos et al., 2017). The belts contain trapped charged particles from the solar wind, which can pose risks to spacecrafts passing through the area (Heirtzler et al., 2002). The SAA is currently observed to be expanding and seems to split into two minima, one over South America and the other over southern Africa (Fig. 1), while the entire low intensity region is drifting westwards (Domingos et al., 2017). This westward movement is thought to result from the convection of liquid iron in the Earth’s outer core. Estimates of the average westward drift rate of the centre of the anomaly vary approximately between ~0.18◦ to ~0.3◦ per year (Fürst et al., 2009; Ye et al., 2017). Concurrent with the growing area of the SAA is an overall decrease of the Earth’s dipole moment, showing a decrease of 9% between 1840-2015 CE (Finlay et al., 2016; Olson & Amit, 2006). As reversals are expected to happen when the Earth’s magnetic field is weak, there is a scientific debate about whether a continuation of this decrease in dipole moment will lead to a geomagnetic reversal. Moreover, it remains a pressing question whether the SAA represents a precursor to an upcoming reversal.

The present-day variations in the Earth’s magnetic field are closely monitored with satellites. However, understanding the changes in the Earth’s magnetic field and the evolution of geomagnetic anomalies prior to the direct measurements is equally important. For this, we rely on a different type of data, which is paleomagnetic data.

The first uses of magnetism go back to the sixth century BCE, when Thales of Miletus described the properties of magnetic lodestone, and to the fourth century BCE, when the first compasses were invented in China. Direct recordings of magnetic declination in Europe go back to the 16th century CE. During sea voyages, declination measurements from around the world were recorded in ship logs, used for navigational purposes. Inclinations were only measured occasionally, systematic records of inclination became more common from the 18th century onwards. The very first measurements of the Earth’s magnetic field intensity began later. In 1798 CE, Alexander von Humboldt started with systematic observations of relative magnetic intensity. True measurements of the absolute magnetic intensity started around 1832 CE, when Carl Friedrich Gauss developed a magnetometer which was capable to measure the direct field strength. This marked the beginning of absolute systematic intensity measurements. The research field of rock magnetism went through innovations in the early 20th century. Bernhard Brunhes discovered that the natural remanent magnetisation in a rock was oriented in a direction antiparallel to the ambient field in 1906 CE. Additionally, in 1926 CE Motonori Matuyama showed from a stack of lava flows that the younger lavas were normally magnetised, and the older lavas were reversed. Both of them discovered the process of a geomagnetic reversal: the magnetic signals in the rocks had a reversed magnetic polarity which indicated that the Earth’s magnetic poles had flipped in the past. The last reversal, Matuyama-Brunhes reversal, was named after them. Laboratory studies on the remanent magnetisation in volcanic rocks were further developed by Johannes Koeningsberger (1930s), Emile and Odette Thellier (1940s) and Takesi Nagata (1940s). They also laid the foundation for paleointensity experiments.

The basic principle of how a rock acquires a magnetisation is relatively easy. When a rock is heated above its Curie temperature, all magnetisations of the mineral particles, such as magnetite, are removed. Upon cooling below the Curie temperature, the magnetisation of the magnetic particles align themselves with the surrounding magnetic field. Thereby they acquire a natural remanent magnetisation (NRM), which reflects the direction and strength of the Earth’s magnetic field at that specific moment in time. Sediments may also acquire a magnetisation: the magnetic particles in a water suspension of sediments can rotate freely, and they turn into the direction of the field to acquire a depositional remanent magnetisation when they are deposited. However, this magnetisation is generally much weaker than the magnetisation acquired by igneous bodies. Paleodirections can be obtained from igneous rocks, archaeological material and sediments. The declination and inclination of a sample is determined by taking oriented samples in the field, and later demagnetising them in the laboratory. This can be done by subsequently heating the samples in steps, or by using alternating fields. After each demagnetisation step the remaining magnetisation in the sample is measured, the results are plotted in an orthogonal vector diagram named the Zijderveld diagram (Zijderveld, 1967). Absolute intensity measurements are not possible for sediments, only for rocks or archaeological material that acquired their magnetisation upon cooling. Furthermore, whereas measuring paleodirections is relatively straightforward, obtaining paleointensity data is much more difficult.

The most common method to obtain a paleointensity from a rock is to subject it to a series of heating experiments (Thellier, 1959). In the IZZI-Thellier method (Tauxe & Staudigel, 2004), the strength of the ancient geomagnetic field in a rock is estimated by comparing the natural remanent magnetisation (NRM) to a thermal remanent magnetisation (TRM) acquired in a known lab field. Samples are heated in a series of temperature increments, alternating between zero-field, in-field (ZI) steps (Coe, 1967) and in-field, zero-field (IZ) steps (Aitken et al., 1988). The NRM remaining is measured during the zero-field steps, and the TRM gained is measured in the in-field steps. The NRM remaining vs TRM gained is plotted in an Arai diagram (Nagata et al., 1963) (Fig. 2), and the slope of the linear segment of this diagram times the laboratory field is the paleointensity. Unfortunately, not all samples behave ideally for paleointensity experiments. Often the specimens suffer from thermal alteration in the higher temperature steps. To check for this a pTRM check is usually included in the measurement protocol: after each four ZI and IZ steps a previous in-field step is repeated to check whether the pTRM of the lower temperature step can be reproduced (Coe, 1967). Another form of non-ideal behaviour is related to the grain size of the magnetic carrier in the sample. The Thellier type methods are based on three laws; Additivity, Reciprocity and Independence of partial thermoremanent magnetisation. In 1949, Néel (1949) showed that these laws were valid for single domain (SD) grains. However, a bulk volcanic sample is often not purely SD-sized, but is a mixture of SD, pseudo-single domain (PSD) and multi-domain (MD) grain sizes. SD grains are very small, so all spins are aligned into one direction and therefore their remanence is very stable. On the other hand, MD grains are large in size, and the grain is split into several domains with a more unstable remanence. PSD and MD grains may cause the three laws underpinning thermal paleointensity experiments to fail and result in non-linear Arai plots.

To detect such adverse magnetic behaviour, several selection criteria have been proposed to check the quality of paleointensity results. MD behaviour can be detected using the curvature criteria k’ and there is a maximum difference a pTRM check may give. Specimens that do not pass the selection criteria are routinely rejected and not included in further interpretations. Applying these criteria, the IZZI-Thellier protocol often results in success rates of only <20%. Therefore, to increase the success rate, various other methods have been proposed. For example the microwave method (Hill & Shaw, 2000; Walton et al., 1993), the multispecimen technique (Dekkers & Böhnel, 2006), the pseudo-Thellier technique (De Groot et al., 2013; Tauxe et al., 1995; Yu et al., 2003), and recent advances in end-member modelling (van Grinsven et al., 2023). Nonetheless, the IZZI-Thellier technique remains the most widely applied method for paleointensity determinations and is used in four chapters of this thesis. In contrast to previous studies obtaining low success rates, we achieved success rates well above 20%. Despite volcanic rocks typically being reliable recorders of the Earth’s magnetic field, there are often biases reported in paleomagnetic datasets. Even the samples with linear Arai plots that all pass selection criteria may yield biased paleointensity results (L. V. de Groot et al., 2013). A bias in the paleomagnetic data can be detected when the data is from recent historical flows, emplaced after 1850 CE. Their declination, inclination and intensity result can be compared with the expected field values according to the International Geomagnetic Reference Field (IGRF, (Alken et al., 2021)), or for flows prior to 1900 CE the gufm1 model (Jackson et al., 2000). One location for which the paleomagnetic dataset often does not match with the expected field is Mt. Etna, a volcano on Sicily, Italy. Mt. Etna is a very active volcano with many recent lava flows, and has been extensively studied. In Chapter 1 we investigated a possible reason for the consistent underestimation reported in paleomagnetic intensity and inclination data for historical flows from Mt. Etna. First, we compiled an overview of all available paleomagnetic data from Mt. Etna, and include new paleomagnetic directional data from seven different historical flows. The paleomagnetic dataset shows no bias in declination, but it does give consistently lower than expected inclinations and intensities. Second, the ambient geomagnetic field was measured at five sites above the surface of the lava flow with a three-axial fluxgate magnetometer. This device measures the magnetic field that a hypothetically new lava flow would record. The field was measured at each site along the lengths of three paths perpendicular to the presumed flow direction at two different heights above the surface of the lava flow. The paths had different topography, and consisted of at least one ridge and one gully. This allowed for detailed mapping of changes of the ambient geomagnetic field above an irregular topography. The measurements show that inclination and intensity values are lower above gullies, and higher above ridges. Furthermore, deviations were larger for measurements closer to the surface. Therefore, variations in the ambient magnetic field on Mt. Etna seem to be caused by the magnetised terrain below: the irregular topography influences the surrounding magnetic field, creating local magnetic anomalies. Lastly, we simulated what the effect would be on paleomagnetic statistics when a hypothetical new flow would cover the surface. We randomly sampled both the complete fluxgate measurement dataset and only the measurements from the gullies. These tests showed that a high k-value, a precision parameter for the clustering of directional data, does not necessarily predict accurate results. To the contrary, a high k-value might actually indicate that local magnetic anomalies were not averaged out. We highlight the importance of taking samples spread out over a large area, especially in a rugged volcanic terrain, to minimise the effect of local magnetic anomalies. Lastly, we advise future studies to always report sampling strategies in detail. Taking samples spread out over a larger area was done in Chapter 2, where we sampled the very young 2021 lava flow on La Palma at three different locations. By visiting the island in October 2021 while the eruption was still ongoing, the sampled rocks were only 3.5 weeks old when they were measured in the laboratory. We tested whether the intensity of the magnetic field at the time of cooling was reliably recorded by the new flow, how successful the IZZI-Thellier technique was on these extremely young samples, and whether there are changes in the intensity over time. The declination, inclination and intensity were measured from the samples of the 2021 flow immediately upon returning from the field. The results varied per site, however, the average result is close to the expected field value. This confirms the sampling strategy recommended in Chapter 1, and implies that this basaltic flow is a good recorder of the Earth’s magnetic field. Then, we stored two batches of intensity samples in the Earth’s magnetic field, and two batches in a shielded room with a residual field <300 nT. For each storage condition, one batch was measured two years after sampling and the others were measured three years after sampling. Across all batches the average paleointensity result remains approximately similar and in agreement with the known paleofield, although it seems to slightly decrease after three years of storage time. However, the success rate of the paleointensity experiments is lower for some sites after two or three years of storage, so fewer samples pass the quality criteria. Nevertheless, all batches had exceptionally high success rates with the IZZI-Thellier method (>48%). Lastly, there appears to be a small tendency for samples stored out-field to have higher success rates than those stored in-field. During a second fieldwork in 2025, new samples were obtained from the same 2021 lava flow, these samples were thus ’stored’ in the natural field for more than three years. Also, Calvo-Rathert et al. (2024) obtained samples from the flow in 2022. Interestingly, both our new samples and those from Calvo-Rathert et al. (2024) were largely unsuccessful with the IZZI-Thellier technique. The decline in success rate for samples stored in a magnetic field may be due to a combination of the ’fragile curvature’ process (Tauxe et al., 2021) and the acquisition of viscous remanent magnetisation (L. V. de Groot et al., 2014a), although differences between sampling the inside or outside of the flow could also play a role.

Current geomagnetic field models that capture the evolution of the SAA have contrasting hypothesis about its emergence. The anomaly may be related to a reverse flux patch at the edge of the Large Low-Shear Velocity Province (LLSVP) below Southern Africa (Tarduno et al., 2015), or migrated westwards from the Indian Ocean (Campuzano et al., 2019; Di Chiara & Pavón-Carrasco, 2022; Nilsson et al., 2022). Also the timing of initiation is unknown, possibly as early as 860 CE (Trindade et al., 2018), since 950 CE (Campuzano et al., 2019), at 1250 CE (Tarduno et al., 2015) or as late as 1800 CE (Gubbins et al., 2006). A recurrence of the SAA has also been proposed (Engbers et al., 2022; Nilsson et al., 2022; Shah et al., 2016; Trindade et al., 2018). The main problem with current geomagnetic models is that there is an unequal data coverage: paleomagnetic data from the Northern Hemisphere is overrepresented in the data kernel of the models, while data from the Southern Hemisphere remains scarce. To describe the recent behaviour of the Earth’s magnetic field and the development of the SAA, paleomagnetic data from understudied regions, particularly in the Southern Hemisphere is needed. At the same time, even in the Northern Hemisphere there are remote regions without Holocene paleomagnetic records, and filling such gaps is also valuable. We therefore present paleomagnetic data from a remote Northern Hemisphere location in Chapter 3, followed by two Southern Hemisphere locations in Chapters 4 and 5.

In Chapter 3 we present paleomagnetic data from the remote Revillagigedo Archipelago, Mexico. These islands lie around 400-700 km south-east of the tip of the coast of Baja California. We visited two islands, the uninhabited island of San Benedicto, and the slightly larger island of Socorro. Little research has been done on the islands: no paleomagnetic data is available from San Benedicto and only one paleomagnetic study has been done on Socorro (Sbarbori et al., 2009). For San Benedicto, we present paleomagnetic data from the 1953 CE lava flow. The three sampling sites of the 1953 CE lava flow show different results in paleodirections, but combining the results of the sites comes very close to the expected reference value. This again supports the sampling strategy recommended in Chapter 1. The paleointensities of the sites, however, underestimate the expected field value. This might be due to multi-domain effects or arise from local magnetic anomalies as proposed in Chapter 1. From the sampling sites on Socorro, one site was dated using radiocarbon and was assigned an age between 4052-2846 BCE and 4950-2846 BCE (Farmer et al., 1993). This site yielded an interestingly low inclination of 4.2◦ , whereas 35◦ is expected at this latitude according to a geocentric axial dipole. This might suggest a previously unknown inclination anomaly in this period, as it is not predicted by geomagnetic models. All other sampling sites on the island of Socorro are older, likely of Pleistocene age. Successfully obtained paleodirections and intensities from these latter lava flows are consistent with expected GAD-values at this latitude, and therefore recorded no anomalous fields directions or intensities.

One hypothesis on the evolution of the SAA is that the feature migrated from the Indian Ocean westwards. To investigate this hypothesis, we visit the island of La Réunion for Chapter 4. La Réunion is an island east of Madagascar in the Indian Ocean, and therefore ideally located to study whether the SAA has influenced this region in the past. Paleomagnetic data was obtained from 18 sampling sites around the frequently erupting Piton de La Fournaise. Most sites yielded successful paleodirections: 10 sites gave reliable paleointensity results. We combined our paleomagnetic data with the results from two previous studies (Béguin, 2020; Tanguy & Le Goff, 2004) to construct a full-vector PSV curve. The dataset of La Réunion consists of a variety of dating techniques, including radiocarbon dated lava flows, locations dated using pioneer trees, and historically documented lava flows. Additionally, there are periods when paleomagnetic data coverage is low due to periods of low volcanic activity. To deal with these two challenges, we used a novel Bayesian method to create PSV curves which we developed in Schanner et al. (2026). This method takes all kinds of age distributions into account, and gives the possibility to use only a simple prior or a global field model as prior for the Bayesian modelling. Using a global field model as prior is especially useful in regions or time periods with low paleomagnetic data coverage. The PSV curves from La Réunion captures rapid directional changes around 1600-1750 CE, and reveal an intensity high around 1400 CE, after which the intensity drops to ~29 µT at 1550 CE. This low intensity might be due to the presence of the SAA. Combined with paleomagnetic data from southern Africa, the data from La Réunion allow for two possible scenarios. The first is that the SAA originated below southern Africa and affected the field at La Réunion before moving westwards. The second option is that it migrated from the Indian Ocean underneath La Réunion prior to 1300 CE, before shifting slightly eastward again to influence the field at La Réunion at 1550 CE.

La Réunion is on the western side of the Indian Ocean, on the eastern side is another underrepresented region in the paleomagnetic dataset: Indonesia. Indonesia lies east of the proposed emergence area of the SAA and within a region where another geomagnetic anomaly has been reported between 1620 CE to 1820 CE, known as the West Pacific Anomaly (WPA) (He et al., 2021; Yue et al., 2024). As there is no paleomagnetic data from volcanic rocks available, and therefore absolute paleointensity data also lack, Indonesia is a key location to investigate both of these anomalies and thereby improve global field models. In Chapter 5, we present the first paleodirectional and paleointensity data from the volcanoes of Gunung Batur and Gunung Agung on Bali, Indonesia. The sampled lava flows on Gunung Batur were of historical age, and were suitable for paleodirectional and paleointensity experiments. By comparing the results from these historical lava flows we find that, contrary to the results from Chapter 1, inclinations and intensities are higher than expected. Models of inclination anomalies in gullies indicated the effect might be reversed in the Southern Hemisphere (Baag et al., 1995), which could be the reason for this reversed effect. The ages of the sampled volcanics on Gunung Agung were older, approximately from 5000 BCE until the recent lava flow of 1963 CE. We construct a PSV curve based on a simple and a global model prior, which reveal rapid directional variations with, depending on the prior, lower intensities of 32 - 36 µT at ~1000 CE or ~1300 CE. In Chapter 4, we observed similar field behaviour at La Réunion, only a few hundred years later. The data from Bali might reveal the initial phase of the South Atlantic Anomaly, or otherwise might be related to the West Pacific Anomaly. We stress that to distinguish between these two scenarios improved age constraints are needed on the lava flows of Gunung Agung, as well as a more comprehensive global field model. Nonetheless, the results from Agung and Batur demonstrate that these rocks are highly suitable for paleomagnetic methods. A future PSV curve may also assist in dating volcanic deposits in Indonesia and thereby map the eruption history, which is particularly important in regions with hazardous volcanic activity such as Indonesia.