Impact on Sahelian runoff of stochastic and elevation‐induced spatial distributions of soil parameters

Abstract

Abstract:Topography controls surface flows, and thereby exerts a significant action on soil formation. At the hillslope scale, infiltrability of the surface horizon varies gradually along the slope. In semi-arid zones, and especially in the Sahel, runoff is Hortonian and depends mainly on the hydraulic properties of the soil surface horizon (saturated hydraulic conductivity K s and hydraulic roughness of the soil surface n). Using the fully distributed hydrologic model r.water.fea as an experimentation tool, this paper investigates the effects of various spatial distributions of K s (deterministic, stochastic or a combination of both, all with an invariant global mean) and related n (taken as fully correlated to K s ) on the outflow of a small catchment representative of Sahelian conditions. In addition to a uniform distribution used as reference, deterministic distributions here consist of linear variations of K s with elevation. A stochastic component is then added by drawing from a log-normal distribution with different variation coefficients C v . Both hypothetical and real rainstorm events are tested.All K s distributions studied produce hydrographs that are very close to the uniform K s case when rainfall is long and intense. For most other rain events, runoff increases with K s variability. Whatever the rainfall event and C v , outflow is greater when the less infiltrative surfaces are located downhill. The ratio of deterministic to purely stochastic variation is a good indication of the relative importance of the two K s variation sources for catchment runoff. Given the high local-scale (stochastic) soil variability typically encountered, only strong catchment-scale contrasts really deserve to be included in the K s distribution for runoff modelling of all but insignificant events. Spatial K s representation can be further simplified down to a uniform distribution, when only a seasonal water yield is the required result.

Runoff from heterogeneous small bare catchments during soil surface sealing

“…Saghafian et al [1995] have shown that in watersheds, with spatially distributed soil saturated hydraulic conductivity, lumped values of K s underestimate the peak discharges and the runoff volumes. Seguis et al [2002] found that hydrographs from heterogeneous catchments are very close to those obtained assuming uniform ones when rainfall is long and intense. For most other rainfall events, runoff increases with increased variability K s .…”

“…For the higher rainfall intensity, which can be considered as simulating an intense rainfall event, this common conclusion seems to be valid only for the unsealed soil case. The conclusion of Seguis et al [2002] would correspond more to the case where soil surface sealing occurred. Considering that their study was related to the semiarid zone of the Sahel, where soils are highly sensitive to seal formation [Hoogmoed and Stroosnijder, 1984;Perez et al, 1999], it is very likely that their observations reflect the combined effect of soil sealing and catchment heterogeneity on runoff production.…”

“…Woolhiser et al [1996] found that runoff hydrographs are strongly affected by the specific distribution of the hydraulic conductivity for small runoff events, the major impact being obtained for cases when K s decrease downslope. Seguis et al [2002] reported that comparing the effect of stochastic and deterministic spatial distribution of K s on runoff generated by hillslopes, outflow was higher when the less infiltrative surfaces were located downhill, whatever the rainfall and variation coefficient were. These results are in agreement with those resulting from the reanalysis suggested by Woolhiser et al [1996] to the data of Smith and Hebbert [1979].…”

Which spatial discretization for distributed hydrological models? Proposition of a methodology and illustration for medium to large-scale catchments

“…From a hydrological point of view, it may result in a loss of information, as some major hydrological processes can be localised on very small units. Illustrations are re-infiltration of runoff at the bottom of hill slopes in the Sahel (Seguis et al, 2002); runoff decrease due to hedge networks (Viaud et al, 2005).…”

Uncertainty quantification

“…(1) with Monte Carlo simulations (MCS) [11,12] and stochastic finite elements [13][14][15][16][17][18]. For transient nonlinear systems such as (1) these direct approaches are computationally expensive, and often prohibitively so, especially when the parameter fields have small correlation lengths and high variances.…”