Rothamsted Repository Download Time-intensive geoelectrical monitoring under winter wheat . Near Surface Geophysics.

Several studies have explored the potential of electrical resistivity tomography to monitor changes in soil moisture associated with the root water uptake of different crops. Such studies usually use a set of limited below-ground measurements throughout the growth season but are often unable to get a complete picture of the dynamics of the processes. With the development of high-throughput phenotyping platforms, we now have the capability to collect more frequent above-ground measurements, such as canopy cover, enabling the comparison with below-ground data. In this study hourly direct-current resistivity data were collected under the Field Scanalyzer platform at Rothamsted Research with different winter wheat varieties and nitrogen treatments in 2018 and 2019. Results from both years demonstrate the importance of applying the temperature correction to interpret hourly electrical conductivity data. Crops which received larger amounts of nitrogen showed larger canopy cover and more rapid changes in electrical conductivity, especially during large rainfall events. The varieties showed contrasted heights although this does not appear to have influenced electrical conductivity dynamics. The daily cyclic component of the electrical conductivity signal was extracted by decomposing the time series. A shift in this daily component was observed during the growth season. For crops with appreciable difference in canopy cover, high-frequency direct-current resistivity monitoring was able to distinguish the different below-ground behaviours. The results also highlight how coarse temporal sampling may affect interpretation of resistivity data from crop monitoring studies.


H i g h l i g h t s
-Hourly electrical resistivity tomography data were collected under a high-throughput field phenotyping platform.
-The dynamics of the electrical conductivity (EC) varied mainly with N treatments and canopy cover.
-We identified a shift in the EC diurnal cycle probably due to the root water uptake. -Little EC difference between the wheat varieties was observed. * E-mail: g.blanchy@lancaster.ac.uk

I N T R O D U C T I O N
Field phenotyping Senapati and Semenov (2020) show that European wheat varieties still have genetic potential to be exploited through breeding programs. Traits such as optimal root water uptake are present in the genetic population but still need to be selected and transferred into commercial varieties via crop breeding. To create new varieties with desirable traits (e.g. high yield, short stem and deep rooting), crop breeders cross other varieties which exhibit one or several of the desired traits. This process generates large number of different genotypes (or lines). To select which genotype possesses which 1 C 2020 The Authors. Near Surface Geophysics published by John Wiley & Sons Ltd on behalf of European Association of Geoscientists and Engineers. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. traits, all lines are grown and their respective phenotype (i.e. the combination of all traits) is assessed. The lines which show desired traits are selected and can potentially become new varieties. Although this is a simplistic description of crop breeding techniques, it provides a context for this study.
One of the usual steps to assess crop phenotype is to grow the different lines in large field fields. This step can be labour-intensive due the large number of lines to screen, leading to a`phenotyping bottleneck' (Furbank and Tester, 2011). To relieve it, new tools are being developed (Araus and Cairns, 2014;Atkinson et al., 2019). Among them, automated high-throughput phenotyping platforms permit the collection of many above-ground traits automatically (Prasanna et al., 2013). An example of such infrastructure is the Field Scanalyzer facility at Rothamsted Research . Despite this progress, there has been less advance in the development of below-ground methods (Atkinson et al., 2019). Geophysical methods, such as electrical resistivity tomography, electromagnetic induction and ground penetrating radar, have been identified as promising candidates to fill this gap (Araus and Cairns, 2014;Atkinson et al., 2019).

Geoelectrical monitoring in agriculture
Geophysical methods can image near-surface processes at multiple-scales  and hence have a great potential for agricultural applications, for example, for assessing the spatial and temporal distribution of soil water. Geoelectrical methods, and more specifically electrical resistivity tomography (ERT), has proven useful in imaging variation in soil moisture in several field applications (Michot et al., 2003;Srayeddin and Doussan, 2009;Whalley et al., 2017). ERT data are usually collected at regular time intervals enabling to separate the static and dynamic components of the soil electrical conductivity. The dynamic component is usually dominated by the change in soil moisture caused by various processes, in particular plant water uptake and evaporation. The static component is usually linked to soil textural properties such as clay content. Such time-lapse studies have been used to investigate the root zone moisture interaction for different ecosystems (Jayawickreme et al., 2008). At smaller scales, ERT monitoring has been applied in orchards to investigate, in two dimension (2D) and three dimension (3D), the soil moisture dynamics influenced by the root water uptake and irrigation strategies (Cassiani et al., 2015;Consoli et al., 2017;Vanella et al., 2018). In herbaceous plants, time-lapse ERT was used to determine the spatial pattern of root water uptake of corn and sorghum in irrigated conditions (Srayeddin and Doussan, 2009) as well as corn with cover crops (Michot et al., 2003). More recently, Coussement et al. (2018) used 2D ERT monitoring to measure the effects of a tree border on the soil moisture of a corn field. At the plot scale, Whalley et al. (2017) used time-lapse ERT to differentiate root water uptake of different wheat varieties.
All the studies above used time-lapse monitoring which usually involves collecting a few sets of ERT measurements during the growth season of the crop or around specific irrigation events. As such, they provide a few snapshots of the soil electrical conductivity, showing the effects of the seasonal processes. Hourly monitoring over long periods is rare but it has the potential to offer more insight into the dynamics of plant-soil-water interactions. For example, Vanella et al. (2018) use hourly 3D ERT monitoring to image the effects of full irrigation and partial root zone drying on an orange tree. They highlight that access to time-intensive monitoring provides more information on the soil moisture dynamics than less frequent measurements under specific transient conditions. Mares et al. (2016) linked the diurnal pattern of soil electrical conductivity with the sap flow movement in pine trees. At the laboratory scale, Werban et al. (2008) monitored at hourly intervals the soil moisture beneath a lupin plant using 2D ERT and estimated the root water uptake of the plant. In addition to being able to follow the dynamics of specific events, hourly measurements have the potential to look at daily dynamics. Finally, another advantage of hourly scale sampling is that it is closer to the scale at which physiological processes of the plant take place. Given the wide availability of automated monitoring ERT instrumentation, high-frequency below-ground geophysical measurements may offer more information for crop breeding studies.
To analyse the value of geoelectrical monitoring under high-throughput phenotying platform in a phenotyping context, this paper focuses on the following research questions. (i) What is the potential of geophysical tools for monitoring below-ground dynamics? (ii) How can geophysically derived below-ground information be linked to above-ground traits dynamics? (iii) What are the capabilities and limitations of geoelectrical monitoring for phenotyping applications?

Experimental setup
The experiments were carried out at Rothamsted Research,United Kingdom (51°48 34.56 N,0°21 22.68 W) in Great Field, under the Field Scanalyzer platform area . The platform covers a flat area of 0.12 ha. The soil is described as a Luvisol (WRB) and is composed of a loamy top layer (0.3 m) over a more clayey layer with flints (Batcombe). The soil drainage can be impeded by this second layer especially in the areas around the platform due to heavy traffic during the construction. Two experiments were conducted during the growing season in 2017-2018 (hereafter referred to as 2018) and 2018-2019 (hereafter referred to as 2019) under rainfed conditions.
In 2018, three different varieties of winter wheat (Triticum aestivum L. var. Mercia Rht3, Mercia RhtC and Shamrock) were sown on 2017-10-30 (all dates are expressed in ISO 8601 format) in`sowing plots' of 0.6 m length by 1 m width with a planting density of 350 seeds/m 2 and were grown under normal UK nitrogen rate (ß200 kgN/ha). Each sowing plot', made up of two subplots, 0.6 m by 0.5 m, was sown with the same variety. Two continuous`sowing plots' of the same variety were grouped to form a plot unit for this experimentation. This design was inherited from a larger experiment taking place in the same field. Each plot was equipped with 10 stainless steel electrodes of 0.1 m length with 0.15 m inter-electrode spacing. The electrodes were entirely buried (end of the electrode at 0.1 m below the surface) between the rows of wheat, hence not in contact with the plants. The pins of two nearby plots were attached to an array of 24 pins (four pins were discarded). The two electrical resistivity tomography (ERT) arrays were connected to an ERT monitoring system. The aim of this experiment was to identify any differences in soil electrical conductivity between the varieties.
In 2019, four plots of a nitrogen/variety trial sown on 2018-10-25 were equipped with an ERT array. Two varieties, Crusoe and Istabraq, were grown in plot of 3 m by 1 m under low and high nitrogen fertilization (50 kgN/ha and 350 kgN/ha as dry pellets, respectively). The first application of nitrogen 50 kgN/ha was made on 2019-03-08 and the second application was made on 2019-04-10. Figure 1 shows the four plots being monitored. Each plot was equipped of 12 stainless steel electrodes of 0.1 m length with 0.3 m inter-electrode spacing. As in the 2018 setup, the electrodes were entirely buried between the rows of wheat, avoiding contact with the plants. The pins of two nearby plots were attached to a 24 pins array that was connected to the ERT monitoring system.

Above-ground variables
The above ground data were collected by the Field Scanalyzer platform . The growth parameters were collected from RGB (Red Green Blue) camera (Prosilica GT3300, Allied Vision, 3296 × 2472 pixels) for the canopy cover and from the 3D laser scanner (Fraunhofer Institute, Munich, Germany) for height.
Canopy cover values were derived from the RGB images and expressed as a percentage of the image covered by green pixels belonging to the plot canopy . The height of the crop was obtained from the three-dimensional cloud points using the 98th percentile of the vertical coordinates of the cloud points (adapting from Lyra et al., unpublished). The height and canopy cover of the crops were available for both 2018 and 2019.

Electrical resistivity tomography
Electrical resistivity tomography (ERT) measurements were collected using a remotely controlled Syscal Pro 48 (Iris Instruments, Orléans, France) at hourly intervals. In both years, the measurement sequence used was a dipole-dipole configuration (using one and two electrode spacing between the current/potential dipole and, respectively, eight and six levels between the current and potential dipoles) with electrode spacing of 0.15 m (2018) and 0.3 m (2018). Reciprocal measurements were included in the sequence after each normal set. Additional dummy quadrupoles (40 for the entire sequence) were also added to optimize the sequence (specific to the Syscal instrument); in total, the sequence for both years was composed of 496 quadrupoles (124 per plot).
In 2018, the system was operational between the end of May and July to capture rainfall events when the wheat was fully mature (between flowering and harvest). In 2019, the ERT monitoring system ran successfully from February to the end of August (flowering around 14th June) with a few data gaps. At the end of May, current injection errors were noted and so the instrument was replaced with another Syscal Pro 48 to allow monitoring until September. We noticed that the data from this second device had higher reciprocal errors than the original one, in particular, for larger dipoles. Despite this, the datasets from both instruments show consistency in dynamics by reacting to rainfall events and showing similar daily fluctuations.
The ERT data collected were processed using the ResIPy software (Blanchy et al., 2020) that makes use of the Occam's based R2 inversion code (Binley, 2015). Because of the short electrode spacing compared to the length of the electrode, the nodes of the mesh corresponding to the electrode were positioned at 60% of the electrode length (Rücker and Günther, 2011). Given the relatively small number of quadrupoles per plot, surveys were combined in batches of 24 (a day) and a power-law error model was fitted for each batch using the binned reciprocal errors. This approach ensures a sufficient number of data points to obtain a robust error model, while allowing the error model to vary throughout the season. Each dataset was then inverted independently in a batch mode. The difference inversion method of LaBrecque and Yang (2001) did not work well for our dataset when applied over the entire season either using a single background survey or applied over consecutive surveys. For 2019, it produced satisfactory results until May, before large changes in electrical conductivity occurred. After May 2019, the difference approach was not able to reproduce the small variations in electrical conductivity observed at hourly intervals in the apparent data. This was partly due to the higher reciprocal errors observed after May that forces the inversion towards a smooth solution. Inverting independent surveys and constraining them to the background survey produced better results for the earlier dates. However, after May 2019, this approach produced inverted sections that were too biased towards the background image. For this reason, we decided to invert each survey independently with its own error model. Although this approach does not take advantage of difference or background regularization option that could potentially reduce time-lapse artefacts, it still produces inversions that show clear temporal dynamics. Each inverted section was then averaged into a one-dimensional (1D) profiles per plot used in the rest of the study. The 1D profiles were computed for ease of comparison between plots.

Electrical conductivity temperature correction
It is essential that the temperature correction is applied to be able to distinguish between soil moisture and temperature effects on electrical conductivity. The variation in bulk electrical conductivity with temperature is due primarily to two factors: the change in the ion mobility and the change in the viscosity of the pore water (Hayley et al., 2007). To account for the effect of temperature, different models have been developed. Ma et al. (2011) compared the different corrections found in the literature and concluded that a ratio model performs well in the range 3-47°C. Beyond this range, the empirical model proposed by Sheets and Hendrickx (1995), which appears in the corrected form in Corwin and Lesch (2005), is more appropriate. Hayashi (2004) explored the range of applicability of the ratio model and concluded that this model is applicable within the 0-30°C temperature range, which is similar to the conclusion of Ma et al. (2011).
Given that our soil temperature lies within the 0-30°C range, we applied the ratio model to our data with a 2% increase per degree: where σ 25 is the equivalent electrical conductivity at 25°C, σ T is the bulk electrical conductivity measured at the temperature T in°C. Note that this model makes the correction factor dependent on σ 25 . For our study, we used the hourly soil temperature values measured at five depths (0.1, 0.2, 0.3, 0.5 and 1 m) under grass from the Rothamsted weather station (e-RA Rothamsted electronic archive) located about 100 m from the experimental plots. The temperatures were linearly interpolated with depth to match the depths of the inverted electrical conductivities. All inverted conductivity values presented hereafter have been temperature corrected using this relationship.

Time series analysis
The decomposition of the time series of electrical conductivities was applied to the 2019 dataset because it is longer and allows analysis of seasonal change (not possible with the shorter 2018 dataset). For a selected depth, the series of interest is composed of temperature corrected inverted electrical conductivities from February to September 2019. The signal is broken down into three components using an additive model: where Y(t) represent the raw signal, T(t) represent the trend, S(t) is the daily component, e(t) is the residual. All components are dependent on time t. Note that the daily component is sometimes referred as the seasonality of the time series and represents repeating short-term cycles in the series. This decomposition is simple but enables the identification of different aspects of the signal. To decompose the signal, the algorithm proceeds as follows: 1 The period of the short-term cycles of the signal is identified. In this case, the signal shows a short-term cycle every 24 hours (daily). 2 A moving average is applied on the series with a window size corresponding to this period, this produces the trend.
3 The trend is subtracted from the raw signal and the resulting values are averaged for each period to form the daily component. 4 The residuals are obtained by subtracting the trend and the daily components from the raw data.
The algorithm was implemented using the 'sea-sonal_decompose()' function of the statsmodels Python package (Seabold and Perktold, 2010). Figure 2 shows the impact of the temperature correction by analysing the cross-correlation between the soil temperature at 0.15 m depth and the corresponding averaged inverted conductivity from the plot of Crusoe 50 kgN/ha. The temperature correction has two main effects. First, it increases the overall electrical conductivity to bring it to an equivalent electrical conductivity at 25°C. That allows us to compare different dates throughout the season. Second, it decreases the crosscorrelation between the two variables.    Figure 5 shows the time course of the different variables collected in 2019. Figure 5(a) shows daily precipitation and potential soil moisture deficit (PSMD). The PSMD was obtained from meteorological variables measured at the Harpenden weather station (full methodology at: http://www. era.rothamsted.ac.uk/Met/derived_variables#PSMD). From the end of April, the canopy cover of the two high N plots exceeded the canopy cover of the low N plots and reached a maximum by mid-June, irrespectively of the variety (Fig. 5b). The canopy cover started to decrease in the beginning of July as an effect of the senescence. In contrast, the height of the crops appears to be related to the variety and less influenced by the nitrogen treatments (Fig. 5c). Note, however, that Istabraq 50 kgN/ha is slightly smaller than Istabraq 350 kgN/ha at the end of the season. Figure 5(d,e) shows the temperature corrected inverted conductivity at depths of 0.15 m and 0.45 m, respectively. The shallower depth (Fig. 5d) shows a peak around 2019-03-20 after the first application of fertilizer and then the electrical conductivity of all four plots starts to decrease coinciding with the measured increase in canopy cover. Two other peaks can be observed around 2019-05-10 and 2019-06-25 after significant rainfall events (Fig. 5a). During these two events, Istabraq 350 kgN/ha and Crusoe 350 kgN/ha show larger increases in conductivity but also a more rapid decrease over the following days. A later rainfall event occurred at the end of August but no dramatic decrease in conductivity is seen following this as the crop has been harvested mid-August. The slight decrease observed could be attributed to the usual drying of the soil. The deeper depth presented in Figure 5(e) shows a more attenuated response to that in Figure 5(d): no clear difference between the nitrogen treatments or the varieties can be seen. However, the two major rainfall events of 2019-05-10 and 2019-06-25 appear to drive a slight increase in electrical conductivity at depth, albeit much weaker than that seen at the shallow depth. Note also the increase in electrical conductivity for Crusoe 350 kgN/ha around 2019-03-20 at −0.45 m. Figure 6 shows the decomposition of a selected portion of the temperature-corrected and inverted conductivity curves during the first rainfall event, May 2019. The observed signal (Fig. 6a) comprised a general trend (Fig. 6b), a daily component (Fig. 6c) and a residual component (Fig. 6d) using the additive model described earlier. The diurnal characteristic of the signal is clearly shown by this analysis (Fig. 6c) decreasing during the day and increasing during the night (shaded areas). This cycle is common to all four plots in May 2019.

Time series analysis
The same additive decomposition can be applied to different moving time windows of seven days with two-day offsets between the windows. The daily component extracted is shown for each window in Figure 7 for the 0.15 m depth. The advantage of applying the decomposition on smaller time windows compared to the whole signal is that it allows us to see the evolution of the daily component through the season. In Figure 7, it can be seen that the lower part of the daily component (strong blue), initially around 6h00 in February progressively shifts down to 17h00 by the end of April, when the crops start to grow a mature canopy and extract more water from the soil. This shift is subtle but consistent among consecutive weeks. Note as well that in February and March (Fig. 7b,c), the decrease in electrical conductivity occurs mainly during the night which is the opposite of what is observed later in the season, in May, for instance (Fig. 6c). Figure 8 shows an enlarged graph during a major rainfall event at the end of May 2019. It illustrates how the shallow electrical conductivity of the two crops which, received larger amounts of nitrogen fertilizer, increase immediately after the large rainfall and then decrease at a greater rate over the following days. The average decrease rates in electrical conductivity are computed between 2019-05-11 and 2019-05-29 for each plot. When grouped by N treatments, high N plots decrease faster (−0.47 mS.m −1 .d −1 ) than low N plots (−0.15 mS.m −1 .d −1 ). This behaviour was mainly observed at depths shallower than 0.2 m. The rates of decrease in electrical conductivity of the four plots correlated well (R 2 = 0.57) with their respective maximum canopy covers (Fig. 5b) but not with their heights (R 2 < 0.01). Subsequent (albeit smaller)

Yield
For each year, the grain and straw dry weights were measured and converted to yield in t/ha at 85% dry matter (Table 1). The yield in 2018 was much smaller compared to 2019. This can be explained by the lack of rain in 2018 and several bird damages. In 2018, Mercia RhtC (tall variety) had the largest grain and straw yield while Mercia Rht3 (dwarf variety) had the lowest. In 2019, the two plots which received more nitrogen fertilizer had a higher grain and straw yield compared to those which only received one application of fertilizer. For the same rate of nitrogen fertilizer, Istabraq had higher yield than Cruose. In 2018, there was no clear relationship between the grain yield and the daily rate of decrease in shallow electrical conductivity after the large rainfall event (R 2 = 0.08). In contrast, in 2019, larger grain yield was associated with larger daily rate of decrease in shallow electrical conductivity after the major rainfall event at the end of May (R 2 = 0.52).

Implementation of geoelectrical monitoring
The inversion of long-term time-lapse electrical resistivity data is challenging. In 2019, the procedure was made more difficult because of the higher reciprocal errors of the replacement instrument, used after May. Difference and background-constrained inversion were tested but both could not reproduce the diurnal dynamics observed in the apparent conductivity data during the entire season and most failed to converge at the end of the growing season. Difference inversion performed well when applied on the data collected before the first nitrogen application but failed to reproduce the variations observed in the apparent values afterwards. Difference inversion is usually effective when the surveys shared a high systematic error and a low random error but that might not be the case in this study. As a simpler approach, each survey was inverted individually with a power-law error model based on the binned reciprocal error of the batch of 24 consecutive surveys. We noticed that the inclusion of an error model greatly helps the inversion to converge and would recommend the addition of reciprocal measurements in automated sequence for this purpose. In applications of difference inversion type schemes, a different type of error model that reduces systematic errors can be considered (Lesparre et al., 2019).
One important challenge that we met with the inversion of hourly geoelectrical data was to be able to retain the daynight pattern observed in the apparent resistivity measurements following their inversion. In this study, we successfully retrieved this pattern for shallower depths, but we noted that deeper depths do not show similar daily fluctuations (Fig. 5e).   Figure 9 compares the evolution of the apparent and inverted values for shallow and deeper depths. Apparent values show a daily pattern for shallow and for deep depths while the daily pattern is only visible in the shallow depth for the inverted values. The current study mainly focuses on shallower depths as they exhibit faster responses to meteorological events but also because most of the root system of winter wheat usually lies above 0.3 m depth (see, for example, Hodgkinson et al., 2017). Without detailed root data for our experiments we have to assume this to be the case here. Additionally, another reason for only observing the daily pattern at shallow depths is the structure of the soil texture. Indeed, the higher clay content of the soil below 0.3 m might have substantially slow down water fluxes and hence attenuated the fluctuations. This is a potential limitation of the current study site and the experiment would benefit from a repeat in a well-drained environment to see if these daily fluctuations can be observed deeper.
Finally, an important factor when measuring hourly electrical conductivity is the effect of soil temperature as shown by the cross-correlation plot of Fig. 2(b). The diurnal pattern of temperature strongly influences electrical conductivity, particularly at shallow depths. Applying the usual temperature correction using the ratio model (equation 1) helps to reduce this effect and decreases the cross-correlation (Fig. 2b).

Coupling with other above-ground variables
In 2018, the different wheat varieties did not show large difference in term of canopy cover which can be attributed to the lack of rain during the canopy expansion phase (Fig. 3b). This might explain why no large difference in the dynamics of the inverted conductivities were observed between the varieties (Fig. 3d,e). Figure 4(d) shows that the conductivity at −0.22 m under Mercia RhtC decreased slightly faster after a major rainfall event which might be linked to the larger canopy cover of the variety. In other field trials, Hodgkinson et al. (2017) observed that the dwarf wheat variety (Mercia Rht3) has a deeper root system but that this does not lead to larger root water uptake. No links could be found between the yield and the dynamics of the electrical conductivity in 2018.
In contrast, large differences in canopy cover were observed in 2019 between the plots. The dynamics of the electrical conductivity is clearly related to the development of the canopy cover when no major rainfall events occur (Fig. 5b,c). Figure 8 shows that the plots receiving more nitrogen show a larger increase in electrical conductivity during the rainfall event. One explanation could be that part of the nitrogen from the last application was still in the soil in granular form, and not yet in a form available to the crop. With the rainfall, it was dissolved again in the soil solution and caused a surge in the electrical conductivity. We did observe a small peak after the first application of fertilizer (Fig. 5d). Once dissolved, the nitrogen is quickly taken up the roots resulting in a faster decrease of the soil electrical conductivity. Figure 6 This newly absorbed nitrogen can then be allocated to the growth of the crop, leading to an expansion of the canopy cover (Fig. 5d). The decrease in electrical conductivity could also be due the crop water uptake which depends on the canopy cover. However, the rate of uptake of the different crops is likely to be comparable given their similar canopy cover prior to the event. In this study, separating the two effects is difficult without independent measure of the soil moisture.

Figure 7
Evolution of the daily component of the additive model fitted on a several moving windows of a week (seven days) with a two-day offset between consecutive windows. (a) Observed data (here the temperature corrected inverted conductivity at 0.15 m depth) and two windows. The first window of a week is extracted, and the additive decomposition is applied. The cyclic component is displayed in (b). A second window is chosen two days later, and the same process is repeated (c). The shaded area represents night. (d) Evolution of the daily components for each moving window over the whole growing season during night (19-7 h) and day (7-19 h). Moving windows spanning no data intervals have been removed.
There was no strong correlation between crop height and electrical conductivity. The crop height was more influenced by the variety and less by the nitrogen treatment. In contrast, the yield of the crops which received more nitrogen was much greater compared to those receiving less. However, for a given level of nitrogen (either 50 or 350 kgN/ha), Istabraq shows a slightly higher yield than Cruose. For example, Istabraq 350 kgN/ha has a higher grain yield (13.6 t/ha) than Crusoe 350 kgN/ha (12 t/ha).

Diurnal cycles
As previously stated, no direct measurements of soil moisture content were collected during these two experiments. However, the relationship between the electrical conductivity and the soil moisture content was known for the soil under the Scanalyzer. With it, we can relate the electrical conductivity data from the graphs above to soil moisture content. However, given the suspected contribution of the nitrogen fertilizer  to the electrical conductivity (mainly around large rainfall events), the focus here has been on electrical conductivity variation. Diurnal patterns are present in the apparent conductivities measured (Fig. 9a,b). Once inverted, and temperature corrected, those diurnal cycles are still visible mainly for shallower depths and attenuated for deeper depths (Fig. 5d,e). In order to see if these patterns are related to crop activity, partitioning of the time series was performed. However, we acknowledge that univocally attributing the changes in electrical conductivity to root water uptake is not possible in this study. Figure 6(c) shows that the daily component for all the plots tends to decrease during day and increase during night in May. Note that earlier in the season the opposite trend was observed (Fig. 6) when the crop had probably less effect on the dynamics of the soil moisture. The daily component is arguably noisy, and we explain this partly because of the noise in the original signal ( Fig. 6a) but also because this daily component is extracted as the mean of the periodic difference between the raw signal and the trend. One main limitation of the additive decomposition is that the daily component cannot vary in amplitude from one day to another. We hypothesize that this daily component is mainly influenced by the root water uptake of the crop -which follows a diurnal cycle as seen, for instance, in Verhoef et al. (2006) or Werban et al. (2008). The nightly increase observed from May could be due to soil moisture replenishment or hydraulic lift (Horton and Hart, 1998).
The same decomposition approach was applied on moving windows throughout the whole season (Fig. 7) and revealed a shift from April onward in the daily component of Figure 9 Comparison between two apparent conductivities (a) and (b) and two inverted temperature corrected conductivities (c) and (d) for the two plots of Istabraq in 2019. Both (c) and (d) were smoothed by a moving average (window = 3). Note that the inverted conductivities at deeper depths do not show strong daily fluctuation compared to the apparent resistivity data (compare plot (d) with (b)) but rather an attenuated version of the seasonal dynamics. the signal. This progressive shift appears at a time when the crops start to grow larger canopy cover and show large decrease in electrical conductivity (Fig. 5d). Note also that the diurnal component of the signal was still strong in February when the crops were small and showed a decreasing electrical conductivity during night time. Such a strong daily component in the signal for earlier dates is unexpected. It could be related to the fact that the temperature correction did not completely remove the cross-correlation between temperature and electrical conductivity (Fig. 2). In this case, there may be a residual effect of the temperature cycle that remains in the series. This effect is overcome later in the season by larger effects of the diurnal soil moisture dynamics.

C O N C L U S I O N
This study shows hourly electrical resistivity monitoring applied to small scale agricultural plots with different wheat varieties and nitrogen treatments. A high cross-correlation with the soil temperature and the hourly electrical conductivity makes it essential for the application of a temperature correction. However, diurnal patterns in the electrical conductivity remains and our analysis suggest that this diurnal pattern is mainly influenced by plant activity particularly when the crops are fully grown. Distinguishing differences between varieties remains challenging, and we did not observe any large differences in electrical conductivity either in 2018 or 2019 experiments. However, the effect of nitrogen uptake could be clearly seen in the dynamics of the electrical conductivity during large rainfall events. We acknowledge the limitation of the approach to monitor a few experimental plots, but we believe that higher time resolution has enabled us to gain deeper insight into soil-plant dynamics than the usual coarser timelapse monitoring, in particular during large rainfall and subsequent drying events but also at the daily scale. Specifically, the ERT monitoring system provided non-invasive depth-specific information that can be related to some above-ground measurements. As such, it offers a unique perspective into the soil-water-plant interactions which is essential for breeding more resilient varieties.

G.B. is supported by a Lancaster University -Rothamsted
Research-CEH Graduate School for Environment PhD studentship. M.J.H and W.R.W. at Rothamsted Research are supported by the Designing Future Wheat Program by the UK Biotechnology and Biological Sciences Research Council [BB/P016855/1]. The meteorological data were obtained from the e-Rothamsted Archive (e-RA) of Rothamsted Research. We are grateful to associate editor Jon Chambers, reviewer Edmundo Placencia and an anonymous reviewer for their comments on an earlier version of the manuscript.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.