**4. Discussion**

#### *4.1. Implications for Noninvasive (Remote Sensing and Ground-Based Electromagnetic) Investigations*

This study shows how remote sensing and ground-based geophysical data represent an ideal combination of noninvasive techniques that can guide direct investigations and, at the same time, can be integrated to provide a 3D reconstruction of the shallow subsoil once supported by the local direct evidence for verification and calibration.

Our results show that in the study area the use of aerial images is very effective in supporting the rapid recognition of geomorphological and sedimentological features with a resolution approaching 0.5 m. However, the aerial pictures dating back to before the 1980s do not provide useful

paleohydrographic evidence, probably because (a) conventional zenithal pictures were originally taken for cadastral purposes and, thus, were shot during the winter season when vegetation cover was limited; and (b) widespread leveling of the fields and use of strong plowing machines was common until that time, thus causing severe erosion of the topsoil, especially in the zones where convex landforms were formerly present, with slight accumulation in the depressions. Thus, in correspondence with natural levees and sand ridges related to scroll-bar sequences, the sandy and silty sediments were exhumed, showing a lighter signal in the soilmarks.

Cropmarks appear to be, in general, the most e ffective indicators of shallow geomorphological features in the considered environment. This is linked to the much higher permeability of the coarser sediments, leading to greater drainage and, in absence of irrigation, water stress to crops especially during the hot season (July and August). As documented in other areas of the Venetian Plain, e.g., [79], besides the weather condition of the period before the image acquisition, the maximum detail shown by the cropmarks strongly depends on the type of cultivated plants and, in particular, it decreases with the size of the leaves and the plant spacing. The best data are generally found in zones with soya bean or hay meadow crops. Wheat, barley and corn display variable visibility, as the first two are seeded in tight rows and have small leaves, but are harvested before mid July; in contrast, corn has a larger plant size and larger distance between rows, but is harvested in September or October and thus can (usefully) experience water stress. Note that this study was carried out analyzing freely available high-resolution images, reaching a resolution of about 0.5 m. This suggests that the use of specific images, acquired through latest commercial satellite or drone-borne multispectral scanners, could easily support the recognition of features between 0.1 and 0.5 m with superior results.

Geophysical data play a critical role in our analyses. They bridge the gap between surface-extensive information provided by remote sensing (that primarily guided the identification of the study area and the location of the surveys) with the information at depth carried out through traditional drilling and sampling operations. The latter in turn were positioned on the basis of remote sensing and geophysical evidence, thus minimizing the sampling to the locations where this information was needed for verification and calibration. The geophysical data constitute the backbone of the study in that they provide ultimately the 3D information to fully reconstruct the sedimentary structure of the site. This result, however, is not trivial to achieve.

First, grea<sup>t</sup> care must be posed in the acquisition phase. This entails not only the choice of the instrument and the strategy developed for covering a large area—in this case a (nonconductive) sledge towed by a tractor at su fficient distance not to induce current in the metal frame of the tractor itself. In addition, care must be paid when setting the instrument to have a good control of the measurement geometry, which is in turn essential to obtaining reliable inversion results. In this case, the stability of the sledge and the positioning care allowed us to obtain, for all data points, an RMS error of less than 10% between measured and simulated apparent conductivity data at the end of the inversion process. Note that carrying the same sonde by hand, particularly on rugged terrain, can easily unbalance the instrument, with measurements thus taken with some coils much closer to the ground than others. This might induce a very large measurement error that is impossible to correct a posteriori, as the true acquisition geometry is then completely unknown.

Second, obvious outliers must be removed from the dataset. These may include negative or extremely high apparent conductivity values (that are physically implausible for the given acquisition geometry), which may be due to local metallic or magnetic features.

Third, an accurate reconstruction of positioning must be made. The availability of reliable colocated data from HMD and VMD is essential to have the six independent pieces of information necessary for depth inversion. This requires both a guided pace in the field to reoccupy roughly the same locations and proper postprocessing to assemble the data that pertain to the same reasonable surrounding of each measurement point, in this case with a radius of 1 m. Noting that the sonde size itself is of a few meters (Figure 2a), this accuracy is perfectly acceptable for the purpose at hand.

Fourth, geophysical inversion is an inherently ill-posed problem. In this case, it means that while the general pattern of the electrical conductivity variation with depth is constrained by the data, the subtler details may not be retrieved uniquely. In other words, at each measurement location, a number of di fferent 1D models, which all, however, maintain certain key patterns, can be equivalent in terms of goodness of fit to the measured data within the data uncertainty range. In particular, for example, if conductivity increases with depth, all inverted profiles will display the same general pattern. However, the transition from lower to higher conductivity may happen continuously, or in steps, and steps may occur at slightly di fferent depths. While this can be viewed as a weakness of geophysical methods (and EMI in this particular case), it is not without remedy. Indeed, geophysics should never be applied without some "ground-truth" coming (as in this case) from drilling investigations (local and not without their uncertainties, but still necessary). Thus, an iterative revision of the inverted vertical profiles was conducted to select, among the plausible electrical conductivity profiles those that also fit reasonably to the direct evidence where this evidence is available. The procedure was simply performed manually, particularly selecting suitable layer interfaces compatible with drilling evidence. This type of procedure should always be applied, and it is in the energy exploration where 3D seismic data are blended, e.g., with well logs coming from deep borings. For the geophysical data used in this study, the "ground-truth" is given by the local comparison between lithology from sedimentary cores (AV\_1, AV\_a, AV\_b and AV\_c) and electrical conductivity derived from EMI inversion at the same locations. A plot showing the resulting correlation, also taking into account the variability of electrical conductivity for the same lithology (shown as standard deviation error bars) is shown in Figure 9.

**Figure 9.** Plot of the interpretation of the EMI inverted conductivity intervals after the correlation with the sedimentary cores AV\_1, AV\_a, AV\_b and AV\_c.

#### *4.2. Implications for Investigation of Meandering River Deposits*

Integrated remote-sensing, geophysical and sedimentological approaches provide insights to discuss key features of investigated point-bar deposits, with specific emphasis on their genesis and internal grain-size variability.

The external boundary of this body is consistent with the morphology of a point bar that originated through lateral migration of a meandering channel [80–82]. Nevertheless, the curved profile of the inner boundary suggests that the bar accretion started from the inner bank of an arcuate channel that showed a sinuosity of ca. 2.3 (Figure 6b). This evidence contrasts the common assumption that point bars originate from a progressive increase in channel sinuosity of a relatively straight channel, which gradually migrates laterally until reaching a sinuous configuration [83–86]. The onset of accretion from a sinuous channel allows for storing a reduced volume of bar sediment in comparison with that produced by inception from a straight channel. In the study case, the 3D reconstruction of the bar body from EMI data shows that about 1.8 million m<sup>3</sup> of sand are stored in the arcuate point bar body (Figure 10a). If one assumes that the accretion of this bar started from a straight channel, as would be suggested by the application of classical sedimentological models to remote-sensing data, the estimated volume of the bar would have been of ca. 3.1 million m<sup>3</sup> of sand (Figure 10b), leading to a remarkable overestimation of the accumulated sand. The onset of point-bar bar accretion from a sinuous channel was probably driven by pre-existing floodplain morphologies, which, at the early stage of channel development, forced water to drain through the paths defined by adjacent depressed areas [87,88]. The establishment of a curved planiform morphology could have been forced by floodplain lithological and morphological heterogeneities [89,90], which are associated with the occurrence of numerous overbank subenvironments, including crevasse-splays, levees, floodplain lakes and floodbasins (cf., [91]). Different deposits and morphologies associated with these subenvironments possibly forced the newly formed watercourse to connect adjacent depressed areas and assume a sinuous geometry.

**Figure 10. old9**. Volume of the point-bar body (**a**) reconstructed through the present study and (**b**) inferred assuming bar growth from a straight channel, as suggested by classical sedimentological models.

The location of the sedimentary cores within the point bar provides information concerning both vertical and lateral grain-size variability within this sedimentary body, with a particular focus on the comparison between upstream and downstream bar deposits. Although a fining-upward grain size distribution has been broadly considered to be typical of point-bar deposits [85,91,92], a certain variability of vertical grain-size distribution has also been documented [93,94]. Core data show that muddy layers occur in the upper part of the bar, although mud is visibly subordinate to sand. The grain size of sand varies significantly neither vertically nor downstream, and the bar is, therefore, characterized by widespread weak vertical grain-size trends. Although the lack of a clear vertical fining in the upstream bar zone is consistent with the occurrence of high bed shear stress [95,96], the paucity of muddy deposits in the downstream part of the bar is a peculiar feature, which cannot be ascribed to the overall lack of mud in the system, being that the overbank deposits are entirely made of mud. The open morphology of the bend [97] associated with the study bar could have hindered the formation of a dead zone, which commonly forms in sharp bends [18], preventing the accumulation of mud in the downstream bar zone.

Figure 11 shows a summary of the workflow we used to extract the relevant information from each data source and blend the pieces of information towards the final conceptual 3D distributed model of the studied site.

**Figure 11.** Workflow of the overall methodology adopted in this study.
