Seasonal variations and interannual trends could be observed considering temporal resistivity changes at both sites. Half-yearly alternating drying, and wetting periods were detected using ERT, which was verified by soil moisture data from both study sites (Figs. 5 and 8). This pattern is consistent with the seasonally fluctuating evapotranspiration of temperate forests with a summer maximum and a winter minimum (Oishi et al. 2010). However, the spatial resistivity variations were opposed with respect to the vertical gradient, i.e., the lowest resistivities in the clayey soil at WUE were in the effective root zone (top half of the probing depth), while they were found below the effective root zone at UHW. This contrast is most likely due to the sand substrate which has lower water retention and higher drainage compared to clay. Plant available water during the humid periods is directly retained by the clayey soil (WUE site) during subsequent growing seasons with potential water deficits. Due to the exceedingly high water storage capacity of the topsoil, deeper layers do not receive any percolating water. Over several years, the water stored in the upper half of the exploration depth at WUE is successively used by the trees during the growing season. Therefore, no seepage water reaches the discharge horizons from the effective root zone into the deeper layers due to the scarcity of plant available water. Consequently, the deeper layers are not affected by seasonal or semi-annual fluctuations, but rather by the water balance in the topsoil. At the sandy UHW site, the situation is reversed, meaning that the topsoil stores a minimum amount of water and most of it seeps directly through the effective rooting zone in a very short time. The interpretations above are also supported by the temporal ratios of the 2D and 3D tomograms (Figs. 5 and 7). Accordingly, they show seasonal fluctuations of the resistivities in the near surface region (changing ratios between > 1 and < 1), which are accompanied by seasonal fluctuating soil moisture values (Figs. 4 and 6). In contrast, the deeper half of the ERT-probing does not show changing ratios between the respective half-years but an interannual trend, revealing a different tendency. The continuous temporal increase in resistivity per season is due to more severe and longer dry periods, resulting in a cumulative effect of desiccation in the deeper half of the subsurface at WUE and should have careful attention in the future.
Spatial anomalies due to trees
The high-resistivity anomalies in the WUE main rooting zone could be attributed to a higher tree density. In contrast, Norway spruce and Scots pine in the 3D-grid of the UHW site promoted low-resistivity model blocks compared to the surrounding sections. This contradiction is due to the higher resistivity ranges of sands compared to clay. Resistivity values of wood (300–1200 Ωm) lie between those measured at the WUE site (15–300 Ωm) and the resistivity values from the main rooting zone at UHW (500–60,000 Ωm), which justifies the different relations. However, there are further reasons for different resistivity values related to the presence of trees, like stem flow, promoting water supply close to the roots. This phenomenon can be very high and varies considerably, depending on species, crown geometry and size (Taniguchi et al. 1996; Návar 2011), resulting in high moisture values and consequently low resistivity values in the soil close to the trunk. Owing to its dependence on precipitation, this part of the water supply occurs for a short time and differs widely, depending on the season (with or without leaves). However, this parameter can be ignored in this study since only coarse barked trees (Scots pine, red oak, Norway spruce), whose stemflow are low or non-existent, formed the main stand. In contrast, seasonally varying water uptake by trees reduces moisture close to the trunk. This is shown by the high-resistivity sections in the ERT-profile at the end of the growing season for the WUE site (Fig. 4) characterized by a higher stand density. Since mature deciduous forests can contribute to groundwater fluctuations up to a depth of several meters due to their transpiration (Mitscherlich 1971), such resistivity anomalies can be attributed to different stand densities. Moreover, different degrees of water uptake depending on species are also possible. When comparing Norway spruce (Figs. 1 and, y: 4 m, x: 2 m) with Scots pine (Figs. 1 and 7, y: 7 m, x: 3 m) located in the 3D grid of UHW site, there is always a stronger low-resistivity anomaly around the pine than with the spruce, especially during the growing season. This matches the generally lower water use of a mature Scots pine compared to Norway spruce during the growing season (Schmidt-Vogt 1986). However, the situation is complicated by trees segregating root exudates seasonally for nutrient mobilization (Gerke 1992; Grayston et al.1997), thereby increasing the electrolyte content of the soil and consequently decreasing resistivity. Therefore, a quantitative assignment of certain tree parameters to the measured resistivity anomalies was not possible in this study. To investigate these anomalies in more detail, models with a higher resolution would have to be developed with the help of, for example, a smaller electrode spacing. However, temporal changes in resistivity values at the same location are still mainly driven by water content, even if the site is covered by trees (see Fig. 6).
Resistivity vs. soil moisture
The strongest resistivity anomalies as well as temporal variations reach down to approximately 1.0 m and 1.5 m depths, respectively, corresponding to the main rooting zones of both forest sites. The temporal variation of the resistivity values correlates well with the soil moisture changes down to 1 m depth (Fig. 6). Since mean or median values from single point measurements are also used in classical soil moisture estimations or for hydrological modelling in forest ecology, calibration of the spatial resistivity model with the help of the functional correlation obtained from mean soil moisture values is considered permissible, at least for the depth range considered in this process. However, this should be verified or corrected using pedotransfer functions (e.g., Archie 1942), incorporating soil physical parameters. Since the sole value of volumetric water content is not sufficient, the estimation of the corresponding soil matric potential would be another step towards a more accurate, two- or three-dimensional high-resolution determination of plant-available water in the effective rooting zone of trees. This could finally allow an estimation of tree-available water for determining critical time spans and even small-spatial soil sections with a limited water supply. Such progress would also be important in times of climate change, not least because the effects of drought on forest ecosystems are not fully understood due to small-scale heterogeneities (Etzold et al. 2014; Schuldt et al. 2020).
Due to the influence of soil temperature in the subsurface (Hayley et al. 2010), a preliminary correction of the resistivity values would be necessary prior to quantitative derivations from resistivity data. Even though there is, according to Ma et al. (2014), no significant correlation between soil temperature and changing resistivity values in forest soils, the moisture values estimated by the apparent resistivity without a temperature correction would be overestimated in summer and underestimated in winter, resulting in an insufficiently stringent evaluation of the trees’ water supply. The fact that the results presented here were not corrected for temperature should be viewed critically, but this had no influence on our relational-qualitative observations and statements. Since dry seasons are characterized by high temperatures and wet periods are shaped by low temperatures in this region, the contrasts between these half-years (Figs. 5 and 8) would even be intensified with a temperature correction of the resistivity values. This is justified by the fact that rising temperatures in the subsurface affect a lowering of its resistivity (Besson et al. 2008; Reynolds 2011).
The electrode arrays selected were based on the respective site characteristics and the measurement set-up. At the WUE site, the more stable, robust, and therefore better suited for monitoring Wenner-Schlumberger array (Furmann et al. 2003; Loke 2004) was used. The data sets show high quality and only low error values over the measurement period (cf. Fig. 4). Individual data sets with larger errors were caused exclusively by minor defects of the cable, which could be limited to short periods by repair or replacement. The corresponding data sets were sorted out before processing. At the UHW site, the dipole–dipole array was used because of better lateral resolution within the comparatively small grid to better resolve the small-scale, three-dimensional variability (cf. Loke 2004). Due to the poorer signal-to-noise ratio and the higher susceptibility to poor ground coupling, the sandy substrate with strongly fluctuating moisture values does not seem to deliver the most suitable conditions for this type of array at this site. This is also reflected in the higher errors compared to the WUE site which, however, are still within an acceptable range. Figure 7 further illustrates horizontal imbalances. Despite generally visible chronological trends given by the 3D-models, there are neighboring model cells in blue and red delivering opposed ratios. Nevertheless, this contrast should not be overrated since the fluffy sand takes a wide span of resistivity values (Reynolds 2011), resulting in big differences even for neighboring model cells. This problem was also enhanced by the warm and very dry summer months when the sandy soil only allowed poor coupling of the electrodes. Nevertheless, in comparative measurements with the Wenner-Schlumberger array at this site, the dipole–dipole array resolved the small-scale variabilities significantly better and was therefore preferred at this site, despite the slightly higher error values.
ERT in forest ecosystem monitoring and outlook
Based on our results, ERT offers great potential in monitoring soil moisture changes at forest sites and at a small scale. As a minimally invasive geophysical method, ERT-measurements do not disturb the soil system and are quickly established. Sampling for estimating gravimetric water content or for installing TDR-probes is considerably more complicated, time-consuming, and costly. Furthermore, they provide limited spatial and temporal resolution, which are both offered by ERT monitoring. Figures 4, 5, 7 and 8 illustrate the importance of each additional dimension in resistivity changes, like the interannual increase of the degree of influence from dry periods and their duration as well as rising resistivity values as a result of decreasing soil moisture driven by the forest stand itself. A higher spatial and temporal resolution in soil moisture distribution particularly offers a better understanding of the water supply for trees in times of dry growing seasons, which is becoming more important.
Nevertheless, for producing quantitative soil moisture values from measured resistivity values (cf. Ma et al. 2014), a site- related calibration with installed in-situ data loggers must be done, and pedotransfer functions (e.g., Archie 1942) should be applied. Soil conditions are to be assumed homogenous, which are rare in forests. The presence of stems and roots is the main driver for this pedological inhomogeneity, owing to their different specific resistivities compared to the surrounding substrate. For this reason, the use of both methods (pedotransfer functions and data logger calibration) is suggested for verifying the site-specific relationship between absolute soil moisture and resistivity values.
To provide evidence for plant-available soil water using resistivity values, it is important to regard both absolute water content and soil matric potential. Concerning forest sites, different sensor distances to the nearest tree should also be realized to ensure an additional view of the impact of roots on the substrate, water content and soil temperature. Such a combined consideration has not yet been addressed in respective studies, which would be a minimum requirement for spatially estimating plant-available water on forest sites. The processing of resistivity data by pedotransfer functions like Archie’s law (Archie 1942), should also be supplied by wide-ranging physical soil data with probing depths extended at least to the investigation depth of the ERT-measurements. Therefore, drilling is unavoidable for this purpose due to the deep probing depth of ERT-measurements. Finally, a link to aboveground parameters, like topography and stand parameters (e.g., leaf area index or crown geometry), would deliver further information for verifying and interpreting the subsurface data. A complete tree mapping or even airborne or terrestrial laser scans would provide a comprehensive basis for this.