Monitoring spatiotemporal soil moisture variability in the unsaturated zone of a mixed forest using electrical resistivity tomography

European forests are suffering considerably from the consequences of the droughts of recent years, and the exact reasons and influencing factors for this are still not fully understood. This study was conducted to characterize the changes and dynamics of soil moisture in a mixed forest in northern Bavaria within 1 year. Since electrical resistivity correlates well with soil water content, we used two‐dimensional electrical resistivity tomography (ERT) monitoring and time‐lapse analyses to supplement punctual measurements by sensors and soil analyses to show soil moisture changes throughout a whole year (2020–2021). While the topsoil dries out significantly from summer to autumn down to a depth of about 3 m, a clear increase in soil water content and a decrease in resistivity below 3 m can be observed during winter period. Anomalies in the topsoil (0–1 m) showing lower resistivities than the surrounding substrate could be related to tree positions by additional terrestrial laser scans. A significant relationship could be found between tree crown projection area and resistivity in 1–2 m depth. We found a trend that mean resistivity below pine is lower as below beech. ERT data were also used to estimate the soil water content via Archie's law and the results correlate strongly with the measured values, but the degree of correlation varies depending on the depth level. ERT as a noninvasive method, in combination with additional data, for example, on the vitality status of individual trees, could help to better understand root water uptake and water supply to trees, especially during periods of drought.


INTRODUCTION
Climate change and the resulting exceptional drought in the spring and summer months of recent years showed clear effects in Central European forests (Schuldt et al., 2020). Scots pine (Pinus sylvestris L.), which is after Norway spruce [Picea abies (L.) Karst] the second most common tree species in Bavaria, covering an area of 17%, showed a mean needle loss of about 35% in 2020, the highest value since the inventory records. Overall, the proportion of defoliation has increased to about 60% and the condition of the tree canopy has worsened, especially in northern Bavaria (STMELF, 2020). Since the severe drought periods in 2018/19 when unexpectedly high tree mortality could be observed, the detailed study of various factors that can influence tree vitality have come into focus (Schuldt et al., 2020). Tree vitality in drought stressed forests can be related to soil water storage capacity, but also light and nutrient availability, neighborhood interactions, and tree species diversity (Chakraborty et al., 2017). These factors cannot just be seen as site specific because there also occurs a high heterogeneity within forest sites. The reason for this could be the complex interaction of different abiotic and biotic factors on a smaller scale. Therefore, a better understanding of soil distribution and heterogeneity is needed (Cailleret et al., 2014).
Soils can store different amounts of water depending on their characteristics. In general, the water storage capacity of soils depends on granulometry, soil texture, soil depth, skeleton content, and percentage of organic matter. When considering the soil water balance at forest sites, it should also be noted that, in addition to the various soil parameters, the variability of water content is also proportionally influenced by stand structures, that is, age, height, spatial distribution, and tree species composition, as these affect the stand climate, throughfall, stemflow, and evapotranspiration (Hartmann et al., 2016).
Due to climate change and the associated challenges for plants and living organisms, there are several studies investigating the soil water balance in different areas. The topic is of interest to a number of sciences, such as hydrology, physical geography, biology or forestry, because of the various factors that influence the water balance and the resulting effects on plants. Most forestry studies involve investigations on soil physical and hydrological parameters in the field and laboratory (Hartge et al., 2014;Morgenstern et al., 2011;. Most of the data are supplemented by long-term observations of soil moisture changes and meteorological surveys. Besides these methods, there is the possibility of using electrical resistivity tomography (ERT) to show or quantify soil moisture changes. This method is based on the fact that the electrical conductivity of subsurface materials, waters, or organogenic deposits varies and is also sensitive to temperature or saturation. Stainless steel rods serve as electrodes and are used to inject electrical cur-

Core Ideas
• In situ soil moisture monitoring for realistic assessment of tree water supply. • ERT monitoring can help understand root water uptake in a mixed forest. • Resistivity of subsurface anomalies correlates with the crown projection area of trees. • Resistivity anomalies show descriptive differences between tree species. • Volumetric water content can be estimated via Archie's law, calibrated with in situ data.
rent into the ground and to measure the potential difference, from which the electrical resistivity is calculated. Since the lithology can be considered consistent, soil moisture is one of the most important variables left to describe temporal changes in resistivity. In general, there are various studies using ERT for soil moisture monitoring in different ecosystems. Alamry et al. (2017) and Nijland et al. (2010), for example, studied the soil moisture variability in Mediterranean soils via ERT. Uhlemann et al. (2016) characterized hydrological processes in wetlands with ERT monitoring. Compared with data from punctual measurements of the soil water content using probes, ERT allows to detect the 2D-or 3D-spatial and temporal dynamics of soil water content via time-lapse analyses through repeated measurements. Several studies already showed that soil water content and respective resistivities of ERT measurements are influenced by precipitation events (Carrière et al., 2020;de Jong et al., 2020;Ma et al., 2014;Jayawickreme et al., 2010) and the composition of the soil but also by the vegetation itself and by the distribution of the root system (Al. Hagrey et al., 2007;Jayawickreme et al., 2008;Mulyono et al., 2019;Ursino et al., 2014). Dynamics in the main root zone or root water uptake (RWU) have been investigated in different studies using 2D or 3D ERT, for example, at vineyards (Mary et al., 2018(Mary et al., , 2020. Whalley et al. (2017) showed how changes in soil water within the root zone can be related to changes in resistivity or conductivity, determined by combining ERT with electromagnetic inductance (EMI) and penetrometer measurements. So far, there are relatively few studies investigating soil moisture variability in forests, including the study by Ma et al. (2014) and Fäth et al. (2022) in temperate deciduous or mixed forests or Fan et al. (2015) investigating a subtropical mixed forest compared with a meadow showing in each case quite different developments in the subsoil based on infiltration and water uptake by roots. Peskett et al. (2020) studied across-slope forest strips and their impact on hillslope subsurface hydrological processes by using time-lapse ERT. A study by Carrière et al. (2020) demonstrates the impact of soil conditions on variations in tree response to drought by using ERT combined with ecophysiological traits. ERT data can show heterogeneities and specific soil moisture patterns in the forest depending on vegetation distribution and canopy structure (Dick et al., 2018). Furthermore, there are some studies using ERT data for the quantitative estimation of soil water content based on Archie's law (1942) or transformations of it, including terms describing the influence of different soil properties (Dick et al., 2018;Jayawickreme et al., 2008Jayawickreme et al., , 2010Nijland et al., 2010;Schwartz et al., 2008;Shah & Singh, 2005). Besides Archie's law, there are other pedophysical models relating electrical resistivity to influencing factors, such as the Waxman and Smits model (Waxman & Smits, 1968), which has been applied in the context of monitoring RWU and dynamics by Garré et al. (2011Garré et al. ( , 2012. However, to make use of these models, more details about the soil at the study site need to be known, for example, particle-size distribution or porosity. To analyze a forest area in more detail, vegetation should also be included, as trees and their roots affect soil moisture and hence, can also have an impact on changes in resistivity (e.g., Jayawickreme et al., 2008;Mulyono et al., 2019). Tree structures can be mapped in three dimensions by terrestrial laser scanning (TLS) and analyses can be performed on individual trees. This method has been used since the early 20th century to obtain quantitative information on forest structures and composition or to quantify individual tree competition (Dassot et al., 2011;Metz et al., 2013). In the course of the aforementioned questions on soil water balance and forest development, TLS plays an increasing role alongside airborne LiDAR data to monitor forest structures and the physical development of ecosystems under climate change (Calders et al., 2020). Furthermore, it can be used to help for the interpretation of the ERT data and topsoil resistivity anomalies that may be linked to the roots of trees.
The aim of this study was to investigate the small-scale differences of the soil water distribution in a mixed stand with Scots pine (P. sylvestris) and European beech (Fagus sylvatica L.) and the soil moisture changes during a measurement period of 1 year. For this purpose, soil physical investigations were combined with ERT and complemented by meteorological and hydrological monitoring data. Terrestrial laser scans served as a further basis of interpretation for the evaluation of the detected resistivity anomalies.

Study site
The study area is located in the "Oberhübnerwald" in the Lower Main Plain close to Aschaffenburg in Lower Frankonia, a government district in the state of Bavaria in southern Germany (49˚59′ N, 9˚4′ E, 129 masl). Geologically, the area borders on the Spessart region and is framed by crystalline rocks and mostly covered by Quaternary sediments such as aeolian sands or fluvial terrace sands of the river Main. Below these sediments red sandstone and crystalline rocks occur, mainly muscovite-biotite gneiss (Streit & Weinelt, 1971). The deposits of the Late Glacial accumulated to mainly longitudinal and parabolic dunes, partly up to 10 m high. These dune forms, which are difficult to recognize as such in the field due to dense forestation, show up very clearly in topographic terrain modeling. Figure 1 shows the Digital Elevation Model for the "Oberhübnerwald" generated from LiDAR data with a spatial resolution of 0.5 m that was provided by the LDBV (2013). The terrain in the study area is relatively flat, with low slopes that are only steeper in the area of leeward slopes of dunes. The surveyed stand is located in a flatter area near dunes (cf. Figure 1). This site is characterized by a temperate warm climate with mean annual air temperature of 10.0˚C and mean annual precipitation of 648 mm (data from the reference period for Schaafheim/Schlierbach: 1981DWD, 2021). The soils in the study area are formed mainly from aeolian/alluvial sands with Dystric Cambisol being the dominant soil type. The chosen study site is a mixed forest stocked with Scots pine (P. sylvestris, age approximately 140 years) and European beech (F. sylvatica; age approximately 40 years). The tree heights within our plot range from a few meters up to 32 m, most beech trees have heights between 20 and 25 m while the Scots pine trees range between 25 and 32 m. The mean diameter at breast height (DBH) is 21.25 cm for beech and 44.88 cm for pine and the tree density is about 550 trees ha −1 (proxy data based on TLS scans described in Section 2.2.3).

2.2
Soil, meteorological, and soil moisture data and tree positions

Soil sensors
At our selected forest site, instrumentation for measuring air temperature, air humidity, precipitation, soil water content, matric potential, and soil temperature has been in place since summer/fall 2020. Precipitation, air temperature, and humidity are measured automatically every 10 min via a weather station. The main root zone is assumed to be between 0 and 1 m depth, so the volumetric water content (VWC) is determined via a total of three ECH 2 O probes installed at 20, 40, and 80 cm depth in the soil. To determine the soil temperature as well as the soil matric potential, which provides information on water availability for plants, three Tensiomark sensors (GeoPrecision GmbH) were installed at each of the three depths mentioned. Additionally, the Tensiomark sensors were placed in different distances to a tree with 0.5-4.5 m distance to the trunk. The matric potential is presented in this study as pF value, which describes the magnitude of the potential or the decadal logarithmized pressure in hPa, respectively. This means the value of the energy with which the water is held in the soil against gravity. Simplified, a low pF value stands for wetness, a high value speaks for dryness (Amelung et al., 2018).

Soil composition
Information on the soil composition was obtained by analyzing soil samples from a soil profile close to the ERT monitoring. The classification of the soil horizons was carried out according to IUSS Working Group WRB (2022). We used soil sample rings with 100 cm 3 volume for undisturbed sampling to measure the soil dry bulk density (ρ t ) by the quotient of the dry mass of the sample by the cylinder volume. Additionally, the total pore volume (TPV) was determined using a pressure membrane apparatus. Further, disturbed samples were analyzed in the laboratory to determine the grain size distribution using standard sieve-pipette method, as well as the content of organic matter. All these soil parameters were investigated for the same depth levels as the ECH 2 O probes in 20, 40, and 80 cm. In order to obtain information on the soil material and soil moisture changes over greater depths below 80 cm, additional percussion drillings were performed to a depth of approximately 6 m within and close to the monitoring plot.

Tree positions
To determine the exact position of the trees and the measuring instruments on our experimental plot, as well as to obtain information on tree parameters, five terrestrial laser scans of the plot were conducted in December 2020 with a Leica Nova MS60 MultiStation using the dome-scan method with a resolution of 5 cm at a distance of 40 m and a maximum range of 300 m. The scanner was positioned at five different locations, one in the center and the other four at the edges of the plot, to capture the trees within the monitoring site from all sides (cf. Gollob et al., 2019). Afterwards, the scans were registered to one 3D-point cloud within the software Leica Infinity. We used the open-source program 3D Forest (Trochta et al., 2017) based on C++ to analyze the scan. The 3D-point cloud was divided into ground and vegetation points and the automatic segmentation process was used to obtain individual tree point clouds. We calculated the height, DBH, and the crown projection area (CPA) of the trees within the forest monitoring plot.

ERT monitoring
ERT measurements are generally carried out by injecting direct current via two emission electrodes (A and B) into the subsurface, where a potential field is subsequently built up depending on the current electrode position and the different conductivities of the soil material. Via the measurement of the potential difference by two measuring electrodes (M and N), the apparent resistivity can finally be calculated (Equation 1) (e.g., Reynolds, 2011). With the help of the potential difference, it is possible to determine the spatial arrangement of the resistivity (or its inverse: electrical conductivity) in the subsurface. The apparent resistivity (ρ a ) can be calculated via Ohm's law using the known current (I), the measured potential difference (∆V), and a geometry-dependent correction factor (k), which describes the geometry of the electrode configuration (e.g., Reynolds, 2011): (1)

Data acquisition
In order to investigate the spatiotemporal changes in resistivity as a response to changes of soil moisture and temperature at different depths, an ERT monitoring profile was established in the forest stand. For this purpose, a cable with 36 electrodes each 1 m spaced was installed. We chose this electrode spacing because we already knew from previous geophysical soundings (ERT, ground-penetrating radar [GPR] and core drilling), that the area is characterized by intermittent waterlogging at depth below 2 m, and we therefore wanted to achieve a good compromise between vertical resolution and the greatest possible coverage of the test area with 36 m profile length. Measurements are performed almost weekly with an IRIS Syscal Pro Switch with 2-4 stacks (measurement replicates per quadrupole), followed by a quality check with a maximum of 5% stack deviation according to Loke (2004). Measurements were performed using Wenner-Schlumberger as well as the dipole-dipole array, as the two complement each other well in terms of horizontal and vertical resolution (Loke, 2004). The electrodes and cables stayed in place the whole year to ensure good comparability of the datasets over longer periods of time.

Temperature correction and inversion scheme
A first evaluation of the data was performed using the software Prosys II where the Wenner-Schlumberger and Dipole-Dipole datasets of the same measurement date were merged to do a combined inversion of both datasets. For interpreting the results of the time-lapse analyses only in terms of water content change over a longer period, it is advisable to subject the resistivity data to a temperature correction beforehand, since the seasonal variations in ground temperature may show a relevant influence on the resistivities depending on the geographical location of the research site (e.g., Hayley et al., 2010). For this purpose, Equation (2) according to Keller and Frischknecht (1966) was used, which was already applied in different studies (cf. Brunet et al., 2010;Chambers et al., 2014;Fan et al., 2015;Jodry et al., 2019;Wicki & Hauck, 2022). Usually, the data are corrected to a standard temperature of 25˚C (Brunet et al., 2010). However, we used a standard temperature T ref of 9.88˚C in this study because it is the annual mean air temperature of the study site, which was measured directly in the stand using the weather station data.
Besides this, the equation remains the same, as well as the α of 0.025˚C −1 (Pellicer et al., 2012).
We follow Hayley et al. (2010), who showed that a temperature correction before inversion can lead to better results. According to Wicki & Hauck (2022), we applied the approach in which the values of apparent resistivity were directly corrected with weighted temperature fields. The temperature used for the correction at each level of apparent resistivity was calculated as a weighted average of the temperature of all overlying depth layers, with increasing weight closer to the surface.
Soil temperatures of the upper depth layers at 20, 40, and 80 cm depth are available for the entire period of measurements by the Tensiomark sensors. To derive estimated temperature data for all depths, we used a model derived from Musy and Soutter (1991), which was already applied, for example, by Brunet et al. (2010) and Wicki and Hauck (2022). We used Equation (3) to estimate the soil temperature: where T(z,t) is the temperature at day t and depth z. T mean(air) is the mean annual air temperature, A is the yearly amplitude of the air temperature variation, d the damping depth, ω is the angular frequency (2π/365), and is the phase offset (− *t 0 ). Based on Hillel (2003) the damping depth d was set to 2.88 m typical for sand. For processing these data and for statistical analysis, the program R (R Core Team, 2021) was used. The correlation of the existing temperature data in 20, 40, and 80 cm and the modeled temperature shows that R 2 ranged between 0.94 and 0.97 and RMS between 2.0 and 2.14, where statistical correlation becomes stronger and RMS increases with increasing depth down to 80 cm.
For correction of the apparent resistivity at different depth levels, we used the weighted mean temperature of the overlying depth levels as applied by Wicki and Hauck (2022). The weight of layers closer to the surface increases as they represent a higher relative volume due to electrical current flowing in half space. This weighted mean temperature was used as temperature "T" within Equation (2).
The time-lapse inversion was performed in ResIPy (Blanchy et al., 2020a) by combining the datasets of the two array types with temperature-corrected apparent resistivities. We filtered data points with a stack deviation of >5%. The time-lapse inversion was done using regularized inversion with linear filtering and logarithmic data type and normal regularization according to time-lapse inversion example by Blanchy et al. (2020b). According to ResIPy sota for all depths, we used a model derived from Musftware, we used the iterations where convergence criteria were reached (mostly 5th or 6th iteration) with a weighted RMS varying between 1.0 and 1.2 (mean: 1.09). Based on the first survey (T1, September 2020), the percentage changes of the resistivities can then be determined to enable a qualitative description of the soil water content change over time.

VWC estimation via Archie's law
For a quantitative estimation of the water content, it is attempted to use the inverted electrical resistivities for deriving the VWC. There are different approaches to derive the VWC via ERT data, but there is no universally applicable formula with constants, because the soils have different properties depending on grain size distribution, pore size distribution, and bulk density (cf. Shah & Singh, 2005;Glover, 2010;Dick et al., 2018). All these approaches are based on Archie's law (1942), which was originally developed for petroleum exploration. It describes the relationship between the conductivity or resistivity of a rock and influencing factors of the material such as the porosity ( ) or the resistivity of the pore water ( w ). The equation is intended for nonconductive porous materials, such as pure sand. Because conditions in the soil usually change with depth, inverted resistivities from ERT measurements are sorted by depth. We used the following Equation (4) by Archie (1942), which refers to the unsaturated zone and has the following variables: the volume fraction of pores filled with water s, cementation exponent m, saturation exponent n, and the tortuosity factor a: The saturation s can take values between 0 and 1, where 0 means dry and 1 means water saturated (Brovelli & Cassiani, 2010) and can be calculated using the ratio of VWC and porosity (Amidu & Dunbar, 2007;Jayawickreme et al., 2008;Verruijt, 2018). For this purpose, the respective VWC measured by the ECH 2 O-sensor per depth level is divided by the TPV of the respective depth determined in the laboratory. According to Liu and Kitanidis (2013), there are different ways to calculate the tortuosity a. In this case (Equation 5) it is determined by the VWC and the cementation exponent m, using the following formula based on Archie's law: Since soils usually have a certain amount of clay content that affects the soil conductivity, the equation can be adjusted by adding a surface conductivity term, but we neglected this term due to lack of sufficient information on the cation concentration per pore volume unit and the average mobility of ions. To fit the parameters for different depth layers, the relationship between water content and total conductivity can also be derived as follows (Equation 6), where is the bulk conductivity, w the pore water conductivity, the VWC, and c T A B L E 1 Parameters used to estimate volumetric water content (VWC) via electrical resistivity tomography and Archie's law and R 2 and root mean square (RMS) of the measured versus estimated VWC at the depths of 0.4 and 0.8 m. and m are constants depending on the grain size distribution (Schwartz et al., 2008;Shah & Singh, 2005). We used this equation to fit the parameter c within our calibration: For calibration, we used the inverted resistivity data from datasets of the combined inversion (November 2020-November 2021; 24 measurements), as well as the measured VWC from the ECH 2 O probes and the soil data to calculate the variables in Archie's law. We could not use all the ERT datasets for calibration as the measurement of VWC started with a time delay of some weeks. We related the inverted resistivities (median) in 0.35-0.45 and 0.75-0.85 m depth to the VWC sensors in 0.4 and 0.8 m depth. We tried different combinations for variables m and n and calculated the median of the variables a, and c for the depths 0.4 and 0.8 m. The parameters that were used to estimate the VWC are shown in Table 1.

Soil analysis
The results of the soil analyses (Table 2) show that the upper two soil layers (20-40 cm) consist mainly of medium and fine sand, typical for aeolian sand. The overall clay content of the upper meter is low (0.83%-9.35%) and decreases significantly with increasing depth up to 80 cm. Below 0.8 m, the substrate can be considered as homogeneous sand up to about 5.3 m what we have determined through percussion drillings. In contrast, the lower depth layers, starting from 5.3 m, differ significantly and show higher clay and silt contents. These deeper layers consist of weathered gneiss with an increased mica content. The TPV in the upper soil decreases with depth, while for the deeper soil layers, no undisturbed samples in soil rings could be taken to determine the density or volume. Figure 2 shows the soil profile of the Stagnic Dystric Cambisol with a soil classification according to WRB (2022). Additionally, we marked the respective sensor depths at 0.2, 0.4, and 0.8 m to match with the respective soil properties.

F I G U R E 2
Soil profile close to the electrical resistivity tomography (ERT) monitoring. Soil classification according to WRB. Ah, mineral horizon of the topsoil with accumulation of organic matter; Bw, mineral subsoil horizon with development of color or structure; Bwg, with loss of Fe and/or Mn by lateral subsurface flow (pale colors), transport in reduced form; Cg, transition to mineral horizon that is less affected by pedogenic processes.

Meteorological analysis
An overview of the meteorological data from the weather station and soil sensors in Figure 3b shows  Figure 3a.
In contrast to precipitation, VWC and matric potential have distinct minima and maxima. The soil water content decreases with depth and has its minima in fall. The course of the soil matric potential (pF value) runs in the opposite direction but also varies depending on depth. Especially at 80 cm depth, the pF values have less fluctuations and are generally higher than in the remaining depths. This is also confirmed by the VWC, which is lowest at 80 cm depth. The pF values at a depth of 20 cm are generally lower, especially in winter, also corresponding to the higher VWC. Down to 80 cm, the pF values T A B L E 2 Textural fractions of the soil close to the electrical resistivity tomography monitoring, bulk density, and total pore volume of the soil up to a depth of 80 cm; last two samples/depths from percussion drillings close to profile.  The permanent wilting point of pF 4.2 is not reached in this measurement period, but the course already indicates a clearly drier phase from August till November. In general, high VWCs lead to a decrease of the pF curve, and high temperatures and a lack of precipitation lead to an increase of the pF curve due to high evaporation. Regarding the median inverted resistivities of the ERT measurements, the resistivities are higher when VWC is low and the highest values are reached in late summer till fall, following the trend of the pF values. While the median resistivity at about 20 cm depth (dark green dots) and 40 cm (light green dots) have only small distance between values, the median resistivity at about 80 cm (turquoise dots) deviates more strongly and shows a slightly different course. To demonstrate the variability of resistivities or soil moisture over a whole year, we have chosen measurement data at fixed points in time with equal intervals of 2 months, although the whole dataset is available with a greater temporal density of measurements. Additionally, we added the position of the trees to help interpreting the anomalies in terms of RWU.

2D ERT monitoring
Orange trees indicate trees standing in 0-1 m distance to the profile, blue ones are in 1-2 m distance. More detailed information on tree characteristics, extracted from the laser scans, can be found in Table 3. The monitoring profile shows a quite horizontal stratification in all tomograms, characterized by higher resistivities in the upper soil, and lower resistivities at greater depth.
The interface runs at about 2-4 m depth, although it varies throughout the year. The specific resistivities for sand can usually range between 100 and 5 × 10 3 Ωm (cf. Reynolds, 2011), so the resistivity within our plot is comparatively high. But previous measurements at another plot within this forest already showed that the resistivities are much higher compared with other forest sites in northern Bavaria (cf. Fäth et al., 2022), which is due to the aeolian and fluvial sands that form the solum of the soil.
The lower resistivities below 4 m depth are related to the change in subsoil material, as the clay content increases at about 5 m depth. But it also indicates a higher water content as the interface shifts vertically over time. This saturation could be confirmed through percussion drillings. Since this boundary is very clear and almost straight, it is most likely the interface to the subsurface zone which is influenced by waterlogging. The sandy soil beginning from 80 cm depth is also bleached and gray (cf. Figure 2) caused by a reduction T A B L E 3 Parameters of the trees close to the electrical resistivity tomography (ERT) monitoring derived by terrestrial laser scanning data.  in the water saturated area. The prominent moisture differences, as well as relatively uniform sandy substrate down to about 5.3 m depth and a change to more loamy material could also be confirmed by percussion drillings close to the profile (cf. Table 2). The study site is not directly influenced by high groundwater levels (HLNUG, 2016), so it is relatively certain to be waterlogging which might be typical for this relief position. From September to November 2020 (T2), an increase in resistivity can be seen and the upper high-resistivity area spreads downward, respectively the interface to the watersaturated zone moves downward, indicating a drying process. This is also illustrated by the positive percentage change (red color) of the resistivities in the lower half of the profile. But not the whole deeper subsoil segment is characterized by drying, there are two anomalies (blue) in T2/T1 that might indicate a wetting process.

Resistivity (Ohm m) 1-2 m
From January 2021 (T3) on, the resistivities decrease significantly in the entire depth profile (blue within difference tomograms), especially the deepest layer, possibly due to the higher precipitation in December and a significantly lower water consumption of the trees. Some anomalies indicating an increase in resistivity are clearly visible, for example, just below the soil surface at Tree ID 4 (cf. Table 3) within the difference tomograms at T3 and T4 (compared with T1). Similarly, such anomalies are visible below the group of trees in the eastern part of the profile (Tree ID 8-11) at about 1 m depth and are most prominent at T6 compared with T1. This could be related to RWU and resulting drying of the soil areas below the trees.
At the end of the measurement period (T7, September 2021), the resistivity is mostly higher as compared with 1 year before (T1). Especially the lowest layer seems drier, except for one anomaly between 5 and 15 m along the profile. In the upper meter at the tree locations, the subsoil is drier as the year before but in 1-3 m depth, there is a heterogeneous pattern with positive and negative percentage changes in resistivity.
Regarding the average resistivity of the anomalies (cf. Table 3), we calculated the median of the inverted resistivity over the whole measurement period, in 0-0.8 m depth, relating to the main root zone and to tree positions (x-location along profile) as well as in 1-2 m depth for comparison. For testing the relationship between resistivity and the tree parameters, we calculated the Spearman rank correlation coefficient, while Welch's t-test was used to test for differences of means in resistivity between the two tree species. For all the parameters, no significant relationship was obtained in 0-0.8 m depth, which can be due to different reasons. One problem is the low number of trees (n = 12) or that, for example, Tree ID 1 and 2 share the same x-location along the profile (standing left and right at almost the same electrode position). We also used different depths for calculating the median resistivity of the anomaly (0-0.5 and 0-1 m) for the main root zone but in all cases, there was no significant relationship.
For comparison, we also tested the relationships between the tree parameters and the median resistivity per tree position within a depth of 1-2 m. Interestingly, there is a significant strong correlation (p = 0.01, r = 0.70) between median resistivity and CPA. This supports the assumption that trees with a high CPA have a higher demand in water and therefore water is absorbed from deeper soil layers as a result of a more branched deeper reaching root system, also resulting in higher resistivity due to lower water content within the soil. Welch's t-test does not show significant differences in mean resistivity for the species, but at least a trend (p: 0.1, mean res. for F. sylvatica: 14,657.28 Ωm, mean res. for P. sylvestris: 11,025.65 Ωm), indicating that resistivities are higher in 1-2 m depth under beech trees compared with pine trees. Considering anomalies under trees at 0-1 m depth that have lower resistivities than the surrounding substrate, it could be that pine trees have deeper reaching roots which lower the resistivity in 1-2 m depth. This assumption would fit to the different root shapes of the two species (cf. Pretzsch, 2019).

Deriving VWC via ERT
To examine the relationship of the water content and the median inverted resistivities in more detail, we compared them at the two depths 0.4 and 0.8 m. The slope and sensitivity of the relationship between electrical resistivity and soil moisture vary with soil depth and the specific characteristics of the substrate. The correlation between the measured VWC and the median resistivities per depth level is strong ( Figure 5), especially at higher VWCs at 0.4 m depth, but the relationship seems to become weaker at low VWCs and higher resistivities. Figure 5 also shows the matric potential (pF) data from the sensor at 0.5 m distance to the trunk. The correlation between pF and median resistivity is particularly strong at 80 cm. In dry soil (at higher resistivities and high pF), the values scatter, possibly due to the location of the sensor and the stronger influence of the tree root system, or the pF values respond with a delay to changes in water content. The estimation of the volumetric soil water content (VWC) was done using the ERT data from all measurement dates (starting from November 2020) and the parameters shown in Table 1. To test the correlation between the measured and the ERT derived VWC at different depths, the data of the ECH 2 O probes at depths 0.4 and 0.8 m were compared with the respective depth levels 0.35-0.45 and 0.75-0.85 m of the ERT measurements.
When considering the VWC derived via Archie's law and the punctual sensor-derived VWC (Figure 6), it becomes clear that in both depth levels there is a strong correlation and relatively minor deviations with a root-mean-square error of 0.01-0.02. Best results with R 2 of 0.88 were found in 40 cm depth while in 80 cm depth R 2 is 0.81. In both cases, VWC is overestimated at very low or in some cases at high water contents, while VWC is underestimated at medium water contents. This may be due to the weaker correlation between low VWCs and high resistivity values, as already shown in Figure 5, or the Archie parameters could not be determined representatively enough via the calibration data, or this misfit could also be caused by not accounting for the surface conductivity in our model.
Another way of showing the relationship between resistivity and VWC is to compare the change in resistivity with the change in VWC (cf. Figure 7). It shows that a positive change in VWC is related to a negative change in resistivity. As soil texture can be assumed to be constant and resistivity data are already corrected regarding temperature changes, mostly the conductivity of the pore water and changes in VWC influence changes in resistivity. The strong correlation indicates that the changes in resistivity are dominated by changes in VWC.

Time-lapse data
The geophysical investigation of the "Oberhübnerwald" site has provided important findings in several areas, for example subsurface stratification and soil structures, and it showed evidence of water influence at depths below 2 m. The highest resistivity values regarding the time-lapse analyses are found in September to November (T2/T1) and the lowest in March to May (T4/T3), a bit earlier than usually reported for the northern hemisphere (cf. Samouëlian et al., 2005). There is also a higher variability of moisture or resistivity differences in the upper 2 m of the soil, typical for forests compared to grassland where it is more homogeneous (Jayawickreme et al., 2008). The resistivities also show a high seasonal variation, which is characteristic for coarse grained soils in comparison with clayey soils (Aaltonen, 1997). The fact that ERT time-lapse analyses can be used to qualitatively characterize the distribution and variability of soil moisture is evidenced by the very strong significant relationship between resistivities and measured VWC in the stand. Other studies such as Brunet et al. (2010) also showed that measured resistivities were significantly related to soil water saturation (R 2 = 0.99). However, how well soil moisture can be inferred from resistivities is highly dependent on the soil substrate, such as clay content or fractures or other irregularities (Alamry et al., 2017). Therefore, it was important to separately investigate the data by depth level because the relationship between water content and the electrical resistivity varies due to different soil characteristics (cf. Friedman, 2005;Michot et al., 2003). In our study, we related the 40-and 80cm depths to the corresponding ERT depth ranges of 35-45 and 75-85 cm. These ranges are small enough to not include significant changes in grain size or soil density because the boundaries of the soil horizons do not lie within these depth ranges. As shown in Figure 5, the correlation between the VWC and the averaged resistivity is strong at higher VWC values but weaker at low VWC and high resistivity values. This is perhaps due to the decreasing influence of soil moisture on the resistivity. The high-residual points are almost all in the late summer to fall of 2021, and especially noticeable at 40 cm depth, so maybe the roots of the trees also have a greater influence here. There was a longer dry phase, which can also RIEDER AND KNEISEL Vadose Zone Journal F I G U R E 5 Correlation of the measured volumetric soil water content (VWC) (m 3 m −3 ) and pF and the median inverted resistivity (Ωm) for the sensor depths 0.4 and 0.8 m; r: Spearman rank correlation coefficient.

F I G U R E 6
Linear correlation of the measured and the ERT-derived volumetric soil water content (VWC) (m 3 m −3 ) using Archie's law (left: 40 cm depth, right: 80 cm depth and respective ERT depths), gray zones indicate 95% confidence interval. be observed in the higher pF values in Figure 3b. This could also have led to changes in the roots, for example, shrinkage, which might cause small gaps, filled with air and this could lead to higher resistivities. The fact that the values do not always correlate very well could also be because it is difficult to compare 2D subsurface data with a single point value of the sensor. If the course of the pF value and the averaged resistivities is compared (Figure 3), it seems very similar.
In the presentation of our results, the inclusion of the topography could be omitted, since the investigated plot is very flat. More complex sites with different slope inclinations and slope orientations are of interest for future investigations, since Seyfried et al. (2021), for example, have already found that the soil climate, that is, soil temperature and especially water content, react sensitively to changes in local slope and aspect.

Topsoil anomalies linked to root zones
Exploiting the benefits of ERT as a noninvasive method with great spatial coverage and time efficiency, Ehosioke et al. (2020) have demonstrated that ERT can serve as a suitable method for the detection of root zones. In detail, root zones cause anomalies in ERT data and can therefore be detected by geophysical methods. Near the surface, in the upper meter, the distribution of resistivities at our site is rather heterogeneous and apparently dependent on the arrangement of trees and their roots. Generally lower resistivities are found in the topsoil, which can be spatially linked to the position of the beech and pine trees close to the monitoring profile. The anomalies are in most times of the year in the range of approximately 2000-3000 Ωm, only higher in fall where they show resistivities of about 4000 Ωm. The typical resistivities of wood are usually lower (about 1200 Ωm; cf. Bieker & Rust, 2010) but the resistivity of the anomalies is a mixture of roots and sandy substrate. These anomalies could also be magnified, at least temporarily, by moisture differences associated with precipitation events, where stem flow around trees can result in higher water concentrations and wetter zones compared with canopy gaps where rain falls directly to the ground. Lower resistivities correlated to the position of roots like at this plot have also been observed in other studies (al Hagrey, 2007;Bass et al., 2017;Rodríguez-Robles et al., 2017). But some studies also show high-resistivity anomalies in comparison with the surrounding substrate (Barker & Moore, 1998;Jayawickreme et al., 2010;Zenone et al., 2008), which is sometimes explained by soil moisture reduction due to the absorption of soil water by the roots (Mulyono et al., 2019). But it should always be considered in context with the surrounding substrate, where clay-rich soils can show low resistivities just like water-saturated soils and where roots may lead to higher resistivities (Fäth et al., 2022). These opposing observations could be demonstrated in more detail in a study by Giambastiani et al. (2022) where ERT was used for the investigation of root distribution. Here, in a sandy soil, similar to our study, lower resistivities were detected in the vicinity of roots or almost identical resistivities compared to the substrate in wet conditions. In clay-rich substrate, the behavior was opposite, which is why there were also different results in the aforementioned studies. In order to show the root areas with higher resolution, which was not the aim of our study, it might be advisable to choose a smaller electrode spacing (cf. Giambastiani et al., 2022). The trees near the profile vary not only in terms of distance from the profile but also in tree species, height, diameter, or CPA. Therefore, these parameters were determined using the application 3D Forest (cf. Trochta et al., 2017) because these factors also have an influence for instance on root dimensions, and thus could influence the resistivity values. Comparing the shape of anomalies and the properties of the trees (Table 3) visually, it seems that the greater the height and the tree diameter, the more prominent is the anomaly. We assumed that the resistivity below the beeches (close to the surface in 0-0.5 m) would be lower as for pine trees, due to the smooth bark and the resulting higher stem flow of beech trees which might lead to higher soil moisture close to the trunks (Návar, 2011) but within this depth no significant difference in resistivity was found between tree species. The anomalies seem to have greater lateral extent close to beeches, while the anomalies linked to pines show a greater vertical extent, corresponding to the respective root shapes since the tree species differ in pile-shaped and heart-shaped root systems (Pretzsch, 2019) and they also differ in age, so the mature pines have probably developed a larger root system. Regarding the correlations of tree parameters and resistivity related to tree position and different depths, the only significant relationship could be found between tree CPA and resistivity in 1-2 m depth. Some of the parameters could correlate with the resistivity below the trees if more trees could be included for this purpose, since n = 12 is very small as in our study.

Soil water content (VWC) derived via ERT data
Overall, the attempt to estimate VWC via ERT data and a petrophysical function has proven to be very successful at this site. Also, in other studies, the estimation of soil water content via soil resistivities worked well and is advantageous because it reflects the water content in a better spatial resolution than only point measurements (Fan et al., 2015). The estimated VWC at our study site fits relatively well with the course of the measured VWC at the respective depth, only in marginal areas, that is, when the soil was particularly dry or very wet, the VWC is under-or overestimated in each case. This phenomenon could already be recognized in experiments by Brunet et al. (2010). There are several factors that can additionally influence resistivity, such as tree roots, so the location of roots should also be considered as far as possible. Since the root distribution can lead to variations, this can complicate the estimation of the water content of forest soils (Sun et al., 2020). Rao et al. (2019) investigated the impact of roots on the conductivity of the soil and showed that the impact is higher in sand than in loam. Therefore, the effects of roots should be considered as far as possible in the future, particularly for the upper depth layers and depending on the tree species also for deeper soil layers.
It is also important that the estimation of VWC via ERT is calibrated to the site beforehand, for example by in situ VWC measurements in the soil or in the laboratory (Brunet et al., 2010;Chambers et al., 2014;Rings et al., 2008). Since the soil resistivity is influenced by many factors, the measurement of further parameters is necessary. We did not perform a calibration in the laboratory, which may have resulted in a better fit of the parameters but using field measurements for such relationships has the advantage that we can get a spatial impression of the variability in this relationship.
However, the relationship between resistivity and water content is not always strong, for example, de Jong et al. (2020) observe less strong correlations between resistivity and measured water content at the same depth below a meadow on aeolian deposits. This can be due to several reasons, such as the lack of a temperature correction or, as shown by Alamry et al. (2017), that the rock or substrate and its characteristics strongly dominate the relationship. In their study, they demonstrate that for basalt there was a very strong relationship between resistivity and soil moisture, while river terrace sediments or marine sandstone showed much weaker correlations.
In addition, different equations can be used to estimate VWC, some of them including surface conductivity, although this is rather controversial (Shah & Singh, 2005). For soils with high porosity and low cementation exponent, for example, well-rounded grains and poor cementation, the surface conductivity is usually neglected, because it has less effect on the overall conductivity (Brovelli & Cassiani, 2011). We have therefore neglected surface conductivity. The clay content of the soil at our study site (cf. Table 2) is quite low but even this amount of clay could affect conductivity. This effect of clay content on conductivity is rather complex and depends, for example, on clay mineralogy, base fluid conductivity, particle size distribution, and volumetric concentration of clay (Hashemi et al., 2021). It might be, that including a surface conductivity term would explain the relationship between bulk electrical resistivity and VWC better. Especially, when the soil is dry, the relationship between resistivity and VWC is weaker, and maybe bulk electrical resistivity is in that case more influenced by other factors as surface conductivity or pore water conductivity. But to include a surface conductivity term, additional information on soil characteristics needs to be known which are not available within our study. Archie's law seems to work quite well for our study area, owing to the high sand content of the soil. For silty or clayey soils, another model such as the one according to Waxman and Smits (1968) should be tested, since the models work differently depending on the soil composition (Wunderlich et al., 2013). Some other studies therefore see major problems in deriving soil water content via Archie's equation. According to Bass et al. (2017), soil water content calculations from ERT data are inadequate for "normal" soils, even when clay content is considered in the formula, because the variation in clay content varies along the 2D area, which can lead to incorrect results. However, the stand area at our study site has been studied and compared at several locations by ERT and GPR and can be considered as quite homogeneous, at least horizontally. Only at greater Vadose Zone Journal depths larger variations are present, but these are accounted by the division of the data into depth layers and the associated laboratory analyses. But especially in the upper 1 m of the soil, there are a lot of anomalies in the tomograms corresponding to the root zones.
As for our findings, VWC is sometimes underestimated or overestimated. Since soil conditions can be considered relatively constant, both the correlation between resistivity and VWC and the comparison of change in resistivity or soil moisture (cf. Figure 7) show that changes in the tomograms (cf. Figure 4) can also be interpreted as changes in soil moisture. But especially when the soil is very dry, the correlation between resistivity and VWC is less strong and therefore the interpretation is a bit more complex and quantitative estimates should be used with caution in the low VWC range.

Spatiotemporal dynamics
In general, 2D time-lapse ERT studies are well suited to distinguish the water-saturated and unsaturated zones and thus detect waterlogging or the influence of groundwater (cf. Barker & Moore, 1998). Likewise, other studies using ERT measurements show that soil moisture variability can be very heterogeneous and not necessarily directly related to precipitation. A study by Ma et al. (2014) shows the investigation of soil moisture variability in a deciduous forest based on ERT measurements. The results showed that throughout the growing season, soil moisture constantly decreased despite large variations in precipitation. The decrease was only occasionally broken by weeks of high precipitation, which did not change the trend. The spatial variability of electrical resistivity and soil moisture did not match soil temperature differences but showed a strong correlation with tree crown variables. A recent study by Carrière et al. (2020) demonstrates the use of ERT monitoring to detect spatial changes in soil conditions and clarify the relationship of these soil changes with variability in tree response to drought. For this purpose, in addition to ERT monitoring, measurements of leaf water status and different leaf traits were collected. Their study was able to demonstrate a significant correlation between the percentage change in resistivity (between dry and wet conditions) and biological variables as a function of drought intensity. During our measurement period, the water supply of the trees was apparently sufficient, as no clear signs of drought stress, such as premature leaf shedding or discoloration of the leaves, could be observed in the period August 2020-November 2021. Therefore, no conclusions can be drawn at this point from tree vitality to resistivity distribution or changes. That the soil was not too dry within this period and that water was still available is also shown by the course of the pF values ( Figure 3b) since their maxima never reach the permanent wilting point here. At our studied site, the suffi-cient water supply may also be related to waterlogging, which we were able to prove by various methods. The waterlogging caused by the loamy layer in the subsoil could also provide a slightly better water supply in future drought periods, at least for trees with roots reaching deep enough, which is why the trees in our study area may be more resilient to drought. However, this concerns only the older trees with extensive and deeper root systems, the natural rejuvenation and new plantations suffer from drought stress, which we already suspected, and which is also evident in our data since last summer.
When considering the change in resistivity compared with the change in VWC (cf. Figure 7), it is clear that the change in resistivity is highly dominated by the change of the water content, which is why the percentage changes in the tomograms (cf. Figure 4) can be mostly considered as changes in water content. The relationship would be almost linear if dry soil was not considered, which was also observed by Whalley et al. (2017). This could have different reasons, for example, the surface conductance as already discussed or perhaps it is because the soil sensors for measuring the VWC are poorer connected to the soil if the soil is very dry. An important reason could also be the fact that we relate a point measurement value to a value calculated as a median over a 2D subsurface. Here, spatial differences, for example, in material, are summarized within a depth layer, even if the soil conditions can be seen as relatively homogeneous within the used depth levels. But also drying processes run with a spatial variability, as also shown in Figure 4, which is why the direct comparison of resistivity within a depth (extracted from 2D data) with a punctual measured value cannot be perfect.
While the spatiotemporal dynamics of soil water content can be mapped well by our ERT study, it is still unknown how much water is actually used by plants or what the exact soil-plant interaction looks like. Brillante et al. (2015) have combined ERT measurements with ecophysiological methods such as the measurement of plant water status to have a closer look at the plant and soil water relationship. This joint consideration could also become of greater interest for future studies.

CONCLUSIONS AND OUTLOOK
Both seasonal fluctuations and the development of water content at different depths can be assessed much better by ERT monitoring than by point measurements. Apart from the fact that sufficient information about the soil must be known, ERT monitoring simply provides insights into the spatiotemporal soil moisture variability and dynamics. Hence, the combination of petrophysical functions such as Archie's law and 2D ERT time-lapse analyses shows new possibilities in the investigation of the water balance in forests.
ERT, as a noninvasive method, offers a good possibility to detect material differences and temporal changes in forest soils. Our study area showed relatively homogeneous soil conditions, which makes it easier to derive the soil water content than with heterogeneous soils. For future studies, it would nevertheless be of great interest to investigate forest areas with a greater soil heterogeneity in more detail, since 2D ERT or even 3D investigations can provide significantly better insights into the subsurface than just punctual sensor measurements. The use of ERT in forests with small-scale changing soil conditions would also allow to study the effects of soil conditions on the vitality of single trees in more detail. When using ERT data to estimate the VWC at such sites, maybe the data need to be subdivided not only in terms of depth, but also along the profile into spatial groups, in case the soil properties differ too much. Since we have seen that the roots of the trees have a greater influence on the resistivity values it might be important to know, how deep the roots reach into the waterlogged zone. The anomalies caused by the roots vary depending on the surrounding substrate and we also found a significant correlation between the CPA of the trees and the average resistivity in 1-2 m depth. Since the dataset is very small (n = 12), it would be interesting to perform the correlation with a higher number of trees, which might allow further conclusions to be drawn. Hence, we plan to use 2D ERT with a smaller electrode spacing as well as 3D ERT in conjunction with EMI and GPR to better map the root zones.

A C K N O W L E D G M E N T S
The research is funded by the German Federal Ministry of Food and Agriculture and the German Federal Ministry for the Environment, Nature Conservation, Nuclear Safety and Consumer Protection (funding code 2218WK22X1). Thanks go to Julian Fäth, Julius Kunz, and Tim Wiegand who helped during field work. The authors gratefully acknowledge two anonymous reviewers and Sebastian Uhlemann for valuable input that helped to improve our manuscript. This publication was supported by the Open Access Publication Fund of the University of Wuerzburg.

AU T H O R C O N T R I B U T I O N
Julia S. Rieder: Conceptualization; formal analysis; investigation; methodology; validation; visualization; writingoriginal draft. Christof Kneisel: Conceptualization; funding acquisition; investigation; methodology; project administration; supervision; writing-review and editing.

D A T A AVA I L A B I L I T Y S T A T E M E N T
The data and codes used in this study are available on request.

C O N F L I C T O F I N T E R E S T S T A T E M E N T
The authors declare no conflicts of interest.