How do modeling choices impact the representation of 1 structural connectivity and the dynamics of suspended 2 sediment fluxes in distributed soil erosion models ? 3

Abstract. Soil erosion and suspended sediment transport understanding is an important issue in terms of soil and water resources management in the critical zone. In mesoscale watersheds (> 10 km2) the spatial distribution of potential sediment sources within the catchment associated to the rainfall dynamics are considered as the main factors of the observed suspended sediment ﬂux variability within and between runoff events. Given the high spatial heterogeneity that can exist for such scales of interest, distributed physically based models of soil erosion and sediment transport are powerful tools to distinguish the specific effect of structural and functional connectivity on suspended sediment flux dynamics. As the spatial discretization of a model and its parameterization can crucially influence how structural connectivity of the catchment is represented in the model, this study analyzed the impact of modeling choices in terms of contributing drainage area (CDA) threshold to define the river network and of Manning's roughness parameter (n) on the sediment flux variability at the outlet of two geomorphological distinct watersheds. While the modelled liquid and solid discharges were found to be sensitive to these choices, the patterns of the modeled source contributions remained relatively similar when the CDA threshold was restricted to the range of 15 to 50 ha, n on the hillslopes to the range 0.4–0.8 and to 0.025–0.075 in the river. The comparison of both catchments showed that the actual location of sediment sources was more important than the choices made during discretization and parameterization of the model. Among the various structural connectivity indicators used to describe the geological sources, the mean distance to the stream was the most relevant proxy of the temporal characteristics of the modelled sedigraphs.


and standard deviations are given in Table 1, while the distributions of the distance to the outlet and to the stream 155 https://doi.org/10.5194/esurf-2020-64 Preprint. Discussion started: 31 August 2020 c Author(s) 2020. CC BY 4.0 License. are shown in Figure 2. These characteristics of the catchments indicate that not only erodibility but also structural 156 connectivity differs strongly between the two catchments and between sources. The geometry of the catchments is divided in three main modeling units with different spatial discretizations and 201 roughness coefficients, i.e. the river network, the hillslopes and the badlands. The river bed was delineated by i)

202
identifying the river network using TauDEM (Tarboton, 2010) and ii) creating a polygon by "buffering" the line 203 feature of the river. In order to take into account that the width of the river varies from upstream to downstream,

204
we introduced a distinction between the perennial river network defined using a CDA of 500 ha and the intermittent value of 15 ha was found to create a river network that includes the intermittent streams observed in the catchment.

208
For the former a buffer of 10 m to both sides of the river was applied. For the latter, composed of small tributaries 209 and in good agreement with field observations of the whole extension of the hydrographic network during floods, 210 a buffer of 5 m was applied. The badlands were delineated based on orthophotos and verified during field trips,

211
while the hillslopes cover the rest of the catchments.

212
These principal modeling units were discretized as a finite volume mesh. In our study, we used an unstructured

229
The value of α was estimated separately for every event and every source as:  (Table 1).

3.4.Modeling scenarios
238 In order to test the effects of model discretization and parameterization on the representation of structural 239 connectivity and on the computed suspended sediment fluxes, the modelling scenarios shown in Table 2 were 240 tested.

242
In the basic scenario the threshold to define the river network was set to 15 ha and the sources were classified    source. In Sc 4b and 4d , the geological sources were classified in two groups based on their distance to the stream.

274
The badland sources in both catchments were classified as being directly adjacent to the stream network or not.

275
The diffuse sources in the Claduègne catchment, i.e. soils on basaltic and sedimentary geology, were classified 276 using a threshold of distance to the stream of 150 m. In Sc 4a and 4c, the geological sources were classified in one 277 to four groups depending on their distribution to the outlet (Figures 2a and 2c). Results show that the model was sensitive to the choice of the CDA threshold used to define the river network.
296 Figure 4 shows the modeled hydrographs that were obtained when the CDA threshold was varied from 15 to 500  Table 3). In both differed strongly between the 6 considered catchments.  Table 3). While increasing n also led to less maximum liquid discharge, this was not the case for solid discharge.

357
Peak solid discharge even increased with increasing nriver in the Claduègne catchment and to a lesser degree also 358 in the Galabre catchment (Table 3). Interestingly, in the Claduègne catchment liquid discharge was more sensitive 359 to changes in nhillsl. than to nriver while solid discharge was more sensitive to nriver. This was not the case in the

360
Galabre where both liquid and solid discharges were more sensitive to nhillsl..

377
Our results showed that even though modeled liquid discharges were sensitive to nhillsl., the sedigraphs of the main 378 sources and thus of total suspended solid discharge were much less sensitive to this parameter (Figure 8). This 379 was due to the fact that in both catchments the main sediment sources were located close to the river (Table 1, (shorter Tlag and Tc, Figure 5, Table 3). We assume that this was mainly due to the steeper slopes of the Galabre 398 catchment (Table 1)