A Unique Long-term Monitoring Transient Electromagnetic Dataset

Abstract

We present the first long-term monitoring dataset collected using a transient electromagnetic (TEM) instrument specifically designed for geophysical monitoring applications. The dataset consists of 762 measurements acquired between January 19^th, 2023, and September 05^th, 2025, from the island of Falster, southwestern part of Denmark, an area where saltwater intrusion is a recurring issue. The dataset displays a high consistency throughout the period, which is the foundation for a reliable interpretation over time. The dataset provides a valuable opportunity for researchers to test and validate advanced TEM inversion algorithms, including deterministic, Monte Carlo or probabilistic inversion algorithms, thereby contributing to the advancement of time-lapse studies using TEM methods.

Background & Summary

The transient electromagnetic (TEM) method is a well-established, noninvasive geophysical technique used for subsurface characterization¹. The method involves pulsing currents through a transmitter (Tx) loop. These currents are turned off as quickly as possible, generating a time-varying magnetic field in the subsurface. This in turn induces electrical currents (also known as eddy currents) that propagate downward and outward from the Tx loop. The eddy currents generate a secondary time-varying magnetic field, whose strength can be measured inductively using a receiver coil (Rx), placed at the subsurface².

The measured decaying voltages, also referred to as dB/dt decays, reflect the electrical resistivity structures of the subsurface, and these can therefore be estimated through inversion methods. The output from such inversion is typically a layered 1D model that describes electrical resistivity as a function of depth. The TEM method is particularly sensitive to conductive units and is therefore useful for differentiating geologic formations in the lower end of the resistivity spectra from formations with higher resistivities. Additionally, it is useful for mapping variations in porewater salinity below the groundwater table and potentially also the groundwater table itself from the contrast in resistivity between saturated and non-saturated units^3–5.

Common TEM modalities include ground-based single-site acquisitions^6,7, surveys conducted using small mobile platforms such as boats and small vehicles (e.g. ATVs)^8,9, and airborne systems¹⁰. TEM surveys typically aim to provide spatial characterization of specific geological or hydrogeological settings. However, dynamic processes are also of significant interest, for example, understanding how contaminants move through hydrogeological systems and the potential effects of such movement¹¹. As a result, there is a growing focus on using TEM for monitoring activities^5,12. For instance, El-Kaliouby, et al.¹³ used the TEM method to study changes in soil during an infiltration process, observing that saturation-driven changes in conductivity produced variations large enough to be detected by TEM.

Because subsurface dynamic processes are typically slow, variations in the time-lapse TEM signal are generally expected to be small over short timescales. This requires a high level of instrument consistency to avoid introducing bias into the measured data¹⁴. To date, most TEM monitoring studies have been short-term^15,16, relying primarily on traditional ground-based systems with only a few measurements, which are insufficient to capture long-term hydrological process. Recently, Zamora-Luria, et al.¹² introduced a monitoring TEM (mTEM) system specifically designed for long-term monitoring applications. A key advantage of the mTEM system is that it is fully automated, making it the first of its kind tailored to this application.

Because the system can be deployed semi-permanently, it enables collection of long-term datasets with low repeatability errors, allowing for noninvasive tracking of subsurface changes. In this work, we present the first long-term monitoring dataset acquired using a semi-permanent TEM instrument. The dataset comprises 701 individual measurements, collected nearly daily from January 19^th, 2023, to September 05^th, 2025.

The dataset enables the detection of resistivity changes with time as a function of depth. In turn, these changes can be correlated with various processes affecting the aquifer, such as saltwater intrusion, groundwater infiltration, and temperature effects¹⁷. Moreover, it provides an excellent opportunity for the academic community to test and develop new inversion algorithms tailored for time-lapse TEM applications.

Methods

The dataset was collected using the mTEM system⁵ (see Fig. 1). The mTEM system consists of a 5 × 5 m Tx-coil and a two 4 × 4 m Rx-coils, arranged in a 14.5 m offset configuration (center to center). A solar panel is integrated into the system to enable automatic battery charging. To ensure a fixed position of the coils and prevent potential damage from surface activity, both the Tx and Rx coils were buried below the plough layer to a depth of 0.4 m. This setup minimizes any potential measurement bias due to changes in geometry². Importantly, the TEM signal is normalized by the measured transmitter current, ensuring that the response is not biased by the coil installation itself. The burial of the coils at 0.4 m depth therefore does not alter the physical signal but simply provides a stable and protected geometry for long-term monitoring. Variations in the measured data can thus be attributed to changes in subsurface resistivity rather than coil setup.

Fig. 1.

Sketch of the monitoring TEM (mTEM) system. Tx-coil: transmitter coil; and Rx-coil: transmitter coil.

The mTEM system is designed to perform automatic measurements at user-defined times and intervals in both Rx coils (Fig. 1). For the purpose of this study, however, the dataset analysis is based exclusively on data from Rx coil 1. Stacking is the standard procedure in TEM measurements, as single transients are strongly affected by noise, thus, by repeating the measurement many times, noise is reduced and the true earth response enhanced^2,18. In our case, measurements were scheduled daily at 02:00 hrs (GMT) for a duration of two minutes, during which 31500 transients were recorded and subsequently stacked to produce a single measurement ensuring high-quality data. The full list of specifications of the mTEM system is provided in Table 1.

Table 1.

Specifications of the monitoring TEM (mTEM) system, see Fig. 1 for a sketch of the system.

Characteristic	Value
Tx-area (single turn)	25 m2
Maximum Tx moment	250 Am2
Pulse repetition frequency	630 Hz
Tx on-time	450 µs
Turn off-time	4 µs
First gate center time	9 µs
Last gate center time	0.98 ms
Gate distribution	21 gates

The mTEM system was installed on the Island of Falster, in the southeastern part of Denmark, on January 18^th, 2023 (see Fig. 2), with the aim of tracking hydrogeological processes undergoing on the site. Previous studies have shown that the eastern part of the area is hydraulically connected to the sea and consequently sea level fluctuations and groundwater recharge influence the freshwater/saltwater balance and potential saltwater intrusion into the drinking water supply¹⁹.

Fig. 2.

Location of the mTEM along with available boreholes, and the coverage of a previous TEM campaign. Lithological description of boreholes is shown in Fig. 3 and resistivity profiles are shown in Fig. 4.

Zamora-Luria, et al.²⁰ conducted a tTEM survey at the site where the mTEM system was installed (see Fig. 2). A comparison between the borehole data and the nearest 1D resistivity profile is presented in Fig. 3. It shows a transition from low resistivity in the uppermost layer (<20 Ωm), which is associated with Quaternary sediments, to higher resistivity values around 20 m depth – likely indicating the transition to the chalk layer. At greater depths, resistivity decreases again, reaching values as low as ~7 Ωm. This lower resistivity zone corresponds to the lower chalk layer (Fig. 3a), where fluid conductivity measurements have shown concentrations like those of seawater^19,21.

Fig. 3.

Comparison of borehole lithology with 1D resistivity profiles from a previous tTEM geophysical survey²⁰. The resistivity values shown correspond to 1D spatially constrained inversion results of the tTEM data shown in Fig. 2. (a) Borehole located along profile 1; (b) Borehole located along profile 2. For locations of the boreholes, see Fig. 2.

Figure 4 shows two resistivity profiles (location seen in Fig. 2) from the tTEM survey with borehole signatures overlain. Both profiles trend from southwest to northeast, capturing the general subsurface geometry of the area. A shallow resistive layer (<20 Ωm) is observed in both profiles, coinciding with the Quaternary layer (see Fig. 3), followed by an increase in resistivity corresponding to the upper chalk, and then a gradual decrease with depth. Notably, both profiles exhibit low resistivity values at approximately 80 m depth for positions where x < 600 m. Since there is no apparent change in geology at this depth¹⁹, the observed decrease in resistivity is likely to be attributed to changes in fluid conductivity.

Fig. 4.

Resistivity sections from a previous tTEM survey. (a) Profile 1; (b) Profile 2. Available lithological descriptions are superimposed on the profiles. The black line in (b) at x = 580 m indicates the position of the mTEM system. The locations of the profiles, boreholes, and mTEM system are shown in Fig. 2.

It should be noted that the purpose of this comparison is not to provide a comprehensive geological interpretation of the area, which has already been described in detail in previous studies (e.g., Knudsen, et al.²¹; Rasmussen, et al.¹⁹), but rather to highlight the hydrogeological units where changes in resistivity are expected during monitoring.

The mTEM system was strategically installed in the southwestern part of the study area (see Fig. 2 for location). The measurements are intended to track potential hydrogeological changes at different depths. In the deeper layers, changes in resistivity are expected due to sea water intrusion¹⁹, while in the shallow layers, resistivity variations may be influenced by temperature and precipitation effects^17,22,23.

Data Records

The dataset supporting this study as seen in Fig. 5, is hosted on Figshare²⁴. It includes nearly daily measurements from January 19^th, 2023, to September 05^th, 2025. Each measurement consists of 21 data points (or gates) measured from 9 µs to 0.98 ms, each with an associated observed value and uncertainty obtained from the data stack and an additional uniform uncertainty. The data is organized in four text files:

1. Data file - a text file containing 762 columns with 22 rows, where each column represents a measurement collected on a specific day. Row 1 specifies the date of the measurement, while rows 2–22 contain the data values, one column for each column of points in Fig. 5.

2. Gate time file - a text file with 21 rows, representing the gate center time at which the data points were measured. The gate center times are identical for all measurements.

3. Uncertainty file - a text file with 762 columns and 22 rows. Each column corresponds to a specific measurement date and contains the associated uncertainty values. Row 1 specifies the date of the measurement, while rows 2–22 contain the uncertainty value for each data point.

4. Waveform file – a text file the piecewise linear waveform.

Fig. 5.

(a) Measured dB/dt responses for each day. Blank spaces indicate periods with missing data due to instrument downtime, typically associated with maintenance. (b–d) Examples of measurements on specific days, plotted as conventional decay curves. The corresponding positions are marked in (a) with vertical black lines.

Figure 5a shows the complete set of measurements, where each vertical position corresponds to the data collected on a given day. Figures 5b-5d present specific examples, illustrating the 21 dB/dt values measured on those days. It is important to note that the gate center times shown in Figs. 5b-5d are identical across all measurements.

The monitoring period extended from January 19^th, 2023, to September 05^th, 2025, totaling 961 days. Data was successfully collected on 762 of those days, leaving 199 days without measurements (~21% of the monitoring period). It is important to highlight that, despite the gaps, this constitutes the longest continuous TEM monitoring reported to date.

Technical Validation

To illustrate the stability of the measurements, the uncertainty associated with each measurement is quantified. In TEM geophysical measurements the total uncertainty on a gate is typically estimated by two components: a uniform (relative) uncertainty, ${\sigma }_o$, that accounts for instrument-related and other non-specific contributions¹⁸, and a measurement-specific (absolute) uncertainty, ${\sigma }_{\mathit{noise}}$. The measurement specific noise is a varying contribution based on the particular noise level at time of acquisition, and it can be estimated from the data stack data. The total uncertainty on gate I is therefore calculated as the sum of

$${\sigma }_i=\sqrt{{\sigma }_o^2+{\sigma }_{\mathit{noise},i}^2},$$ 1

In this work, the uniform uncertainty, ${\sigma }_o$, is set to 0.5% for all data points. This relatively small value was chosen to illustrate the stability of our dataset, which shows very low noise. It is important to note that in practice, uniform uncertainties on the order of 3% are often assumed for TEM measurements, and the methodology apply can readily accommodate such values. The choice of ${\sigma }_o$ therefore does not affect the general conclusions of the study but highlights the robustness of the dataset.

Note that, at early times (i.e., the first data points), the uniform uncertainty tends to dominate, whereas at late times (i.e., the last data points), the measurement-specific uncertainty becomes more significant as the observed data approach the background noise level (see Fig. 5).

Furthermore, the collected data are expected to change consistently over time. However, these changes are also expected to be gradual, as hydrogeological processes in this setting occur at a relative slow rates²⁵. To quantify the data changes, we calculate the percentage change in the zero-normalized (z-score) measured data. As a reference, we use the average values collected between May 27^th, 2024, and June 3^rd, 2024. This interval was selected because it coincides with the period where the normalized data are centered around the zero mean, thus providing a stable and representative baseline before the onset of marked seasonal variations.

Figure 6a shows the uncertainty for each datapoint in the dataset, and we observe that the uncertainty remains stable throughout time, with uncertainties of 0.5% equal to the added uniform uncertainty until 200 µs. After 200 µs the uncertainty starts increasing reaching values higher than 10% at the last gates. This increase in the uncertainty is expected as the background noise starts dominating¹⁸.

Fig. 6.

(a) Total uncertainty for each data; (b) data variation over time, using the period from May 27^th, 2024, and June 3^rd, 2024, as a reference. The variation is computed over the normalized dataset. Normalization was performed using the z-score. Black lines (a) and (b) represent the interval used as a reference.

Furthermore, Fig. 6b shows the change in the measured signal over time. The change is consistent and partly repeats seasonal patterns, which, according to Zamora-Luria, et al.¹⁷, is mainly driven by seasonal temperature effects. To examine this correlation, daily air temperature variations over the area were analyzed and compared with the monitoring dataset (Fig. 5). Figure 7 presents the daily air temperature variations from January 19^th, 2023 to September 5^th, 2025. Clear seasonal changes are evident, with higher temperatures during the summer months and lower temperatures in winter, as expected for temperate climates.

Fig. 7.

Daily air temperature variations from January 19^th, 2023, to September 05^th, 2025.

A Pearson correlation analysis was conducted between the dataset and the daily air temperature variations shown in Fig. 7. Because each measurement consists of 21 points measured at different gate times, the Pearson correlation was performed at each of the gate times. It is also important to note that there is a delay in the temperature penetration in the ground, which depends on the thermal diffusivity of the soil. Thus, a time-lag correlation was also performed as explained by Zamora-Luria, et al.¹⁷. Figure 8a presents the Pearson correlation coefficients between the data measured at each gate time (Fig. 5) and the daily air temperature (Fig. 7). A strong positive correlation of up to 0.91 is observed for gate times up to 100 µs. Beyond 100 µs, the correlation gradually decreases, although meaningful correlations remain detectable up to 300 µs. The lower correlations at later gate times can be explained by increasing noise levels as well as the fact that late-time responses are primarily sensitive to deeper parts of the subsurface, where additional processes such as salinity changes may occur.

Fig. 8.

(a) Pearson correlation coefficients between temperature and the data at each of the gate times of the dataset. It is noted that a strong positive correlation exists between both datasets; (b) Time-lag at which the maximum between temperature and the monitoring dataset is observed.

Figure 8b shows the time lag at which the maximum correlation is observed. Up to 100 µs, the maximum correlation consistently occurs at a lag of about 43 days. Beyond 100 µs, the time lag results become less consistent. From this analysis, it can be concluded that the seasonal changes observed in our dataset are directly related to hydrogeological variables and not due to external noise factors.

Usage Notes

The mTEM instrument was specifically designed for geophysical monitoring applications. The dataset presented in this study can be used to test new inversion algorithms tailored for time-lapse applications. Additionally, it serves as a valuable reference for comparison with datasets collected using similar TEM instruments.

Author contributions

Juan Carlos Zamora-Luria: Conceptualization, Technical Validation, and Manuscript Draft. Anders Vest Christiansen: Conceptualization, Coordination and Final Manuscript Review.

Data availability

The dataset used in our study is publicly accessible on Figshare²⁴ (10.6084/m9.figshare.30464183.v3).

Code availability

The code used to analyze the data uncertainty is available at²⁴ (10.6084/m9.figshare.30464183.v3).

Competing interests

The authors declared no competing interests.