The obtained inverted responses were compared with the responses of the true models

Because of their central positions, the electrodes are closer to the expected sources of current and thus in the region of maximum potential gradient. Hence, this electrode configuration maximizes the changes in both magnitude and sign of the measured ΔV associated with a change in the CSD distribution. The electrode diameter was 1.5 mm. The penetration of the electrodes into the rhizotron was 4 mm ± 1 mm. To evaluate the possible distorting effects of the densely populated electrodes on the potential field distribution, a test was performed with low conductivity water . The test showed no resistivity anomalies, which may be caused by the presence of the electrodes . Therefore, while rhizotron setups with electrodes only on the sides were successfully adopted , we found that the current setup represents a better solution for iCSD experiments . Data were acquired with a MTP DAS-1 resistivity meter with 8 potential channels. For the ERT acquisition over the 2D grid of electrodes, we chose a dipole-dipole skip 2 configuration. For each skip 2-couple of injection electrodes the remaining skip-2 couples of electrodes were used as potential dipoles . The associated complete set of reciprocals was also acquired,vertical grow table the resulting acquisition sequence contained 3904 data points . Following the ERT data acquisition, the MALM data acquisition required little setup adjustments and time.

As the two current electrodes are fixed, the use of a multichannel resistivity meter significantly reduced the acquisition time and, consequently, supported the acquisition of more robust data sets. Electrode 1 was used to inject the current into the plant stem, while electrode 64 was used as a return electrode in the growing medium . The remaining 62 electrodes were used to map the resulting potential field. A sequence with 204 ΔVs was used. Considering the grid in Fig. 2a, the sequence included the vertical, horizontal, and diagonal ΔVs between adjacent electrodes. While 61 ΔVs would provide all the independent differences, the 204 ΔV sequence was preferred because of its redundancy and consequent lower sensitivity to acquisition errors. The acquisition time remained relatively short as the multichannel instrument was optimized with fixed current electrodes that allowed 8 ΔVs to be measured at once.ERT and MALM data were filtered considering: 1) reciprocal error , 2) stacking error , 3) minimum measured ΔV , 4) apparent resistivity , and 5) maximum electrode contact resistance . The filtering was implemented to enhance the control on the ERT inversion and obtain a reliable ρmed for the successive iCSD inversion. After the processing, the data were inverted with the BERT inversion software to obtain ρmed . Data error was set to 5% in line with the stacking and reciprocal thresholds used for the data filtering. The regularization was adjusted using the lambda optimization algorithm provided by the BERT software.

Generally speaking, a rhizotron is treated as a 2D geometry. However, this bounded and thin geometry leads to some complications into the ERT inversion. Specifically, nocurrent-flow boundary conditions were set for all the rhizotron surface and the inversion had to be adapted for the resulting pure Neumann problem . For a higher quality forward calculation, the rhizotron volume was discretized in 3D by extruding a 2D mesh with 5 layers. The discretization allowed the refinement of the mesh near the electrodes while maintaining high mesh quality. At the same time, in order to limit the inversion time and force the inversion to be two-dimensional , the elements in z direction were grouped together and inverted as a unique variable. This way, a 3D forward calculation was implemented within a 2D inversion .The iCSD inversion that we developed was based on the physical principles of a bounded system in which linearity and charge conservation were applied to decompose the investigated CSD distribution into the sum of point current sources. This provided a discrete representation of the root system portions where the current leaks from the roots into the surrounding medium. Because of the linearity of the problem, the collective potential field from multiple current sources is the linear combination of their individual potential fields. As such, the measured ΔV can be viewed as and decomposed into the sum of multiple ΔVs from a set of possible current sources. These possible current sources are namedViRTual electrodes . As purely numerical electrodes, they are simulated by mesh nodes representing possible current sources, but with no direct correlation with the real electrodes used during data acquisition. Basically, the VRTe were distributed to represent a grid over which the true CSD distribution is discretized. In order to account for any possible CSD, a 2D grid of 306 VRTe was arranged to cover the entire rhizotron . The charge conservation law implies that the sum of the current fractions associated with the VRTe has to be equal to the overall injected current, which is provided by the resistivity meter.

If we normalize the injected current to be equal to 1, the sum of the VRTe weights has to be 1 as well. Briefly, for Ohm’s law, normalizing the current to 1 is equivalent to calculating the resistance, R, from ΔV. Then, the use of R simplifies the presentation of the numerical problem. Once the VRTe nodes are added to the ERT-based ρmed structure, the potential field associated with each of the VRTes is simulated with BERT. From these simulated potential fields, the same sequence of 204 R is extracted, each corresponding to a single VRTe. Each extracted sequence contains the resistances that would be measured in the laboratory if all the current sources were concentrated at the VRTe point .Synthetic numerical and laboratory experimental tests were performed in order to evaluate the capabilities of the setup and inversion routine to couple the ERT and MALM approaches for the iCSD. In the numerical tests both the true source response and VRTe responses were calculated with BERT. Figure 3 shows an explanatory numerical test with inversion of a point source, and the associated Pareto front that was used to select the optimum regularization strength. As this first experiment was performed to specifically test the inversion routine, a homogeneous ρmed was used in order to avoid influence from the baseline resistivity distribution complexity. For the second experiment, the laboratory tests were conducted. Because of the ρmed heterogeneity of any experimental system, these laboratory tests need to include the ERT inversion,nft hydroponic and the use of the obtained ρmed as input in the iCSD. The true current sources were obtained using insulated metallic wires inserted into the rhizotron . The insulating plastic cover was removed at the tips of the metallic wires to obtain the desired current sources. Six experimental tests were performed using different numbers and positions of these current sources. The rhizotron was filled with tap water and left to equilibrate to achieve steady state conditions of water temperature and salinity, thus minimizing ρmed heterogeneity and changes during the experiment. Changes in ρmed during the ERT and MALM acquisition periods would make the ERT-based ρmed less accurate and compromise the iCSD. To make sure ρmed was stable, a second ERT was performed after the MALM acquisition and compared with the initial measurement. The conductivity of the solution was also measured in several locations of the rhizotron with a conductivity meter to validate theρmed obtained from the ERT inversion. This setup allowed the acquisition of good quality data sets since less than 5% of the data were discharged during the data processing. Because of the controlled laboratory conditions, the ρmed obtained with the ERT was stable and consistent with the direct conductivity measurements. The quality of the ERT inversion was also confirmed by comparing the model responses with the acquired data . Similarly, the acquired iCSD data were plotted against the resistances calculated with the CSD distribution obtained from the iCSD.

The tests also allowed a more informed definition of the VRTe grid. For our setup, a spacing of 3 cm provided a good compromise between resolution, stability, and duration of the iCSD routine. The 3-cm spacing also agrees with the ERT resolution, which would not support a higher iCSD resolution. Successive numerical tests were based on the 8- source laboratory tests shown in Fig. 4. These tests aimed to 1) link laboratory and numerical tests to evaluate the influence of the numerical iCSD routine and laboratory setup on the overall iCSD stability and resolution; 2) account for a more complex CSD, given by the 8 wire-tip sources that were used to simulate distal current pathways; and 3) account for possible ρmed heterogeneity. To address goals 1 and 2, the position of the 8 sources was replicated in the numerical tests and a test with homogeneous ρmed was included to simulate the water resistivity of the laboratory tests. To address goal 3, heterogeneous ρmed were tested.In order to account for the heterogeneous ρmed the following modeling steps were carried out. First, a true ρmed was assigned to the mesh cells of the rhizotron ERT model. We included homogeneous, linear, and quadratic resistivity profiles in the y direction, see Fig. 5. Second, the ERT acquisition was simulated with the ERT laboratory sequence and 3% of Gaussian error, in line with reciprocal and stacking errors observed in the laboratory data sets. Third, the forwarded ERT data sets were inverted following the exact laboratory procedure. A refined and different mesh was used for forward and inverse problems to, respectively, increase the simulation accuracy and avoid the inverse crime . The ERT forward calculation was then repeated over the inverted ρmed.As for ERT, we compared true and inverted MALM responses. First, the true response was simulated with the 8 current sources overt the true ρmed. Second, a MALM response was calculated over the inverted ρmed and inverted to obtained the inverted CSD. Third, the obtained inverted CSD was used to forward calculate the inverted MALM response over the inverted ρmed. True and inverted MALM responses were then compared.We performed hydroponic and soil experiments using maize and cotton plants. In all the plant experiments, the injection electrode was positioned in the plant stem at a height of 1 cm from the surface of the growth media. For the hydroponic experiments, the plants were first grown in columns with aerated nutrient solution . They were then moved to the rhizotron for the experiments. As in the metallic roots test, the rhizotron was filled 1 day before the experiment to reach stable and homogeneous temperature and salinity conditions. The plant was positioned at the center of the rhizotron with soft rubber supports. The plants were submerged at the same level as in the growing column to avoid discrepancies caused by the plant tissues adaptation to the submerged and aerated conditions, as discussed above with regard to the growing conditions. Consequently, the root crown was approximately 3 cm below the water surface. For the soil experiments, seedlings were grown directly in the rhizotron to avoid damaging the roots and altering the root-soil interface. The soil was prepared by mixing equal volumes of sandy and clay natural soils acquired from an agricultural study site run by U.C. Davis, CA . The plants were irrigated with double strength Hoagland solution . Two soil experiments were performed. In the first experiment, four cotton plants were grown for four months. For these experiments, the plants were positioned with the root crown approximately 8 cm deep . In the second experiment, a pregerminated maize seed was planted 3 cm deep and then grown for four months .Figure 3 shows the result of a synthetic numerical test performed to evaluate the iCSD resolution, inversion stability, and influence of imposed constraints. The obtained CSD matches the true position of the simulated current source. The sum of the current sources equals 1 as expected and required by the continuity constraint. The resolution of the CSD is in line with the electrode interspace . The first-order regularization does not hinder the reconstruction of simulated point source. Figure 4 shows the results of a laboratory experiment where the iCSD method was tested with a known distribution of current sources obtained with metallic wires. The use of metallic wires offered a comprehensive solution to test the overall correct functioning of the laboratory setup and inversion routine. The iCSD correctly characterized both position and intensity of the test sources with no need for prior information to constrain the solution.