arising from: P. Piña-Varas et al. Scientific Reports https://doi.org/10.1038/s41598-023-43326-0 (2023).

Tracking changes in electrical resistivity structure at volcanoes as a monitoring method using magnetotelluric (MT) data is an important goal which has been shown to have significant benefits for understanding magmatic systems1,2,3,4,5. However, as with any developing method, care must be taken to ensure robust results. We believe that the recent study by Ref.6 (herein referred to as PV23) lacks sufficient evidence to support the claim that observed temporal variations in MT data are due to active processes within the crustal magmatic system. Below, we argue that the temporal variations shown in PV23 are not due to deep magmatic processes related to the Dec. 2021 eruption on La Palma and are instead due to problems with the geoelectric field data quality likely related to electrode instability along a single dipole. The fact that the observed temporal variations are not due to deep magmatic processes significantly undermines the interpretation and conclusions of PV23.

As shown in previous temporal MT studies, we would expect changes to have a relatively subtle effect on MT data over a limited range of frequencies (or periods)1,2,3,4,5. This is because electromagnetic waves propagate diffusively at the frequencies MT uses, and sample bulk average resistivity values. However, the data presented in PV23 show rapid and dramatic changes to the MT data across all periods with apparent resistivity varying by over an order of magnitude (Fig. 1). This fundamental observation raises many questions about the results, methodology, and conclusions and suggests that further scrutiny is warranted.

Fig. 1
figure 1

Comparison of apparent resistivity and phase data from the four intervals examined in PV23: LP-6, LP-10, LP-21, LP-27. First column shows off-diagonal components (XY and YX); second column shows diagonal components (XX and YY). Data have been rotated by 15° to match PV23.

Full size image

An additional concern with the observations is that the vast majority of the observed changes only occur in the XY and XX components, with limited changes to the YX and YY components (Fig. 1). The components are related to the measurement coordinate system: the XY and XX components are related to electric fields measured in the north–south direction, while the YX and YY components are related to electric fields measured in the east–west direction. In PV23, it is argued that the dramatic difference in the temporal variations between the XY and YX components can be explained by the fact that the geology of the region has roughly north–south striking features in reference to shallow fault systems. This may be true, but the large magnitude of the variation in the XY component still must be explained. In addition, the unrotated data have extreme temporal variations in the XY phase (including out of quadrant phases > 90° and < 0°) which may also point to a deeper underlying issue with the data. It is difficult to explain why the unrotated data would have such large temporal variations in the phase (including different quadrants > 90° and < 0°) while data rotated by just 15° have no such temporal variations.

One possible explanation for the odd data characteristics is a problem with the north–south geoelectric field dipole measurement. Long-term geoelectric data are notoriously difficult to measure due to electrode drift from temperature effects, self-potentials, electrode polarizations, poor ground contact, precipitation events, and other instabilities1,7,8,9,10. In particular, historical weather data from La Palma airport11 indicate rainfall events at various times throughout the measurement period and one possibility is that variations in soil moisture content due to, e.g., rainfall, might negatively impact data quality especially if an electrode is already unstable for other reasons (i.e. poor ground contact). It is also telling that the XY component is consistently noisier (e.g. larger error bars) than the YX component in nearly every 10-day batch process. This persistent increase in noise could be indicative of problematic or variable ground contact along the north–south dipole due to e.g. site-specific soil conditions. Problems with the north–south electric dipole would elegantly explain the enormous temporal variations which largely impact the XY and XX components. Unfortunately, PV23 does not provide any plots of the geoelectric time series to assess data quality, particularly during intervals of large variations in apparent resistivity, nor do they provide any information about electrode ground contact resistance or electrode installation procedures.

Regardless of the data quality, another important issue with PV23 is the use of three-dimensional inversion modelling to deduce changes in resistivity structure using a single continuous MT site. PV23 uses an existing model of the magmatic system based on a ~ 50 site array12 which is then updated based on the inversion of a repeated measurement at a single location. Issues with this methodology are discussed in detail for the Tongariro magmatic system where it was shown that a small number of repeat measurements did not provide sufficient data coverage to update the pre-eruption model based on a larger array3. Forward modelling studies which properly account for the sensitivity kernel for the limited data space is arguably a better way to test for changes in the data. If the model space is poorly constrained by the data, then algorithms are prone to generating artefacts, especially as the distance from the single site increases. PV23 manually constrain the model space by fixing parts of the model space a priori, but such a constraint may inadvertently bias the inversion, and may still ultimately lead to artefacts without proper sensitivity testing to validate which features in the model are responsible for the observed variations in the data.

Our hypothesis is that the large variations in the MT data are due to electrode instabilities in the north–south geoelectric field measurement, rather than deeper (> 1 km) variations in the magmatic system. If this hypothesis is correct, we might expect these issues with the geoelectric field measurement to manifest as near-surface artefacts in the inversion model near the MT measurement site. In fact, the inversion model presented in PV23 supports this interpretation (Fig. 2). At LP10 and LP27, there is a strong conductor (< 1 Ωm) in the near-surface (~ 200 m.a.s.l.) which appears immediately adjacent to the single MT measurement site. This same feature is not present in LP6 or LP21 and represents a structural change in > 2 orders of magnitude in resistivity from model to model. This feature is not acknowledged in PV23, and this represents a significant oversight, as we believe the feature likely accounts for the majority of the change in the predicted data as the inversion tries to fit the distorted XY and XX components. This model result is consistent with our interpretation that small, near-surface distortion related to, e.g., electrode instability on a single dipole best accounts for the change in the observed data. Deeper changes in the PV23 models are more subtle, and further model sensitivity testing is required to validate the impact of these deeper variations on the predicted data fit.

Fig. 2
figure 2

Modified version of PV23 Supplementary Fig. SM6 showing the shallow modelled resistivity structure. Large changes in near-surface structure near the measurement site (black dot) are highlighted with the black arrows.

Full size image

To illustrate how a small near-surface feature can have a strong impact on the predicted data in the inversion, we constructed a model of La Palma island and the MT response was calculated at the location of the long-period site in PV23 for four different scenarios (Fig. 3). The first was a reference model which included the subsurface and the ocean with uniform values of 100 Ωm and 0.3 Ωm, respectively. The resulting MT data computed at the measurement site for this model varies with frequency due to the inductive and galvanic effects of the topography and the conductive ocean (Fig. 3c). The other scenarios can be compared to this reference model.

Fig. 3
figure 3

Synthetic modelling study showing the influence of a shallow feature (C1) and a deeper “magmatic” feature (C2) beneath La Palma island with topography and ocean bathymetry15. Both features are 1 Ωm, background resistivity is 100 Ωm, and ocean resistivity is 0.3 Ωm. (A) Horizontal slice through the model at 0 km below sea level (b.s.l.) showing the outline of La Palma island. Dashed rectangle shows the spatial footprint of C2 at depth. Black triangle is the MT site location. Pink box shows a zoomed view of C1 at 1.07 km above sea level (a.s.l.). White line from P-P’ shows the location of the cross-section. (B) Cross section through the model along P-P’ showing the relative size and depth of C1 and C2. (C) Apparent resistivity and phase data calculated for four different scenarios: 100 Ωm reference model, C1 included, C2 included, and both C1 and C2 included.

Full size image

The second scenario includes a small 1 Ωm conductor (C1) directly below the MT measurement located at 200 m depth (Fig. 3). With a thickness of 40 m, the total volume of this feature is < 0.005 km3. This relatively small and shallow feature results in a decrease in the observed resistivity at all periods (30—10,000 s) by a factor of ~ 4 in the XY component but limited change in the YX component. The reason for this behavior is due to galvanic charge build-up at conductivity boundaries which is a well-studied phenomenon in the MT literature13.

A third model was constructed with a significantly larger 1 Ωm conductor (C2) oriented north–south at a depth from 1 to 4 km below sea level, located ~ 2 km west of the measurement location (Fig. 3). This is analogous to the distance between the La Palma MT site location and the Cumbre Vieja vent, and similar to the depth of the inferred resistivity changes of magmatic origin presented in PV23. It is important to note that the size and resistivity of this feature are far more extreme than the inferred changes described in PV23. This feature would represent ~ 54 km3 of pure melt injected into the 100 Ωm background resistivity. For reference, it is estimated that the La Palma eruption produced < 0.1 km3 of material (VEI 2–3)14. However, even in this extreme scenario, there is relatively limited change in the MT data compared to the data from the reference model (Fig. 3c).

Finally, the fourth scenario included both C1 and C2 in the model and the computed data show that the effect of C1 dominates the signal. It is important to reiterate that this is true even when the volumetric change at C2 is enormous and physically implausible. This simple modelling exercise supports the idea that the dramatic changes observed at the La Palma MT site are better explained by local, near-surface distortions near the measurement site, rather than deeper, magmatic sources related to the Cumbre Vieja eruption.

It is worth noting that even these relatively extreme cases could not reproduce the scale of the observed variations in XY apparent resistivity in PV23 (Fig. 1). Furthermore, given the long-period measurements > 30 s employed in PV23, all the scenarios considered produced largely galvanic effects on the synthetic data (i.e. frequency-independent shifts in the magnitude of apparent resistivity, and limited changes to the phase data). Higher frequency measurements are required to resolve the inductive effects that would be associated with temporal changes to the crustal magmatic system, which are not possible given the choice of instrumentation in PV23 with 4 Hz sampling rate. All previous temporal MT studies at volcanoes that we are aware of have used broadband MT instrumentation able to sample frequencies > 100 Hz1,2,3,4,5 and hence are sensitive to the depths where the crustal magmatic system is expected. This raises the question of whether the long-period MT instrumentation used by PV23 is appropriate for crustal volcanic monitoring studies, when broadband MT instrumentation is better suited. We acknowledge that the MT dataset presented in PV23 may indeed be sensitive to some variations at depth due to changes in the magma plumbing system, especially given the volcanic activity during the time of measurement. However, changes in the data due to magma movement are likely far more subtle than the changes implied by PV23. It is likely that cleaner geoelectric data, and/or more temporal MT sites using a more suitable frequency bandwidth (i.e. higher frequencies) are necessary to definitively detect changes within the La Palma magmatic system.

In summary, there are several significant issues with the interpretations and conclusions presented in PV23. The most significant oversight is that alternative explanations for the large variations in observed data are not considered such as electrode instabilities on the north–south dipole. Our arguments and analysis above suggest that the observed changes in the MT data are not due to deep (> 1 km depth) magma movement. Firstly, it is difficult to explain how magma movement at depth could lead to such dramatic changes in the MT data at all frequencies; a clear sensitivity test which demonstrates the physical plausibility of such an observation is required. Secondly, the model presented in PV23 shows significant near-surface variations immediately adjacent to the measurement site (Fig. 2), but the influence of these variations on the data fit is not discussed in PV23. As shown in our synthetic example (Fig. 3), these near-surface variations in the model likely play a large role in the predicted data fit and further sensitivity testing is required to demonstrate otherwise. Near-surface variations in the model may be due to real variations in shallow hydrologic structure near the measurement site, but, in this case, are best explained as inversion modelling artefacts because of poor geoelectric data along the north–south electric dipole. A problem with a dipole-specific geoelectric field measurement due to, e.g., electrode instability, would elegantly explain why the observed variation is so pronounced in the XY and XX components compared to the YX and YY components. We believe that careful consideration of electrode stability is necessary to rule out the possibility of variations in electrode ground contact or electrode drift related to e.g. precipitation events, temperature effects, self-potentials etc. We hope that additional data collection and analysis can continue to show the value of MT monitoring at volcanic systems.