Subsurface topography to enhance the prediction of the spatial distribution of soil wetness

Publisher Website
Google Scholar
Abstract
The estimation of the spatial distribution of soil wetness within a catchment is one of the most important issues in hydrological and erosion modelling. So far, such models have been based on soil surface topographic information only. However, soil hydrology is also controlled by subsurface flow pathways that may not be explained only by surface terrain features. This study examined how the topography of the lower limit of the soil cover could improve quantitative modelling for the spatial prediction of soil wetness. The study was conducted in an agricultural catchment of the Armorican Massif (western France) characterized by impermeable granitic saprolites. Two digital elevation models (DEMs) with a 10‐m grid mesh and with a 0·3 m vertical resolution were generated from field investigations throughout the catchment. One DEM was a numerical representation of the soil surface and the other described the topography of the boundary between the soil cover and the underlying impermeable saprolite. Soil wetness (θ) was surveyed systematically from 1996 to 1997 along a hillslope. The sampling scheme consisted of 149 nodes of a 10‐m grid where θ at 0–10 cm was estimated using time‐domain reflectometry. The value of θ at depths of 20–30, 50–60 and 110–120 cm was estimated for a subset of 112 data points using a gravimetric method. For both surface and subsurface DEMs, soil wetness at all depths significantly correlated with the topographic attributes, namely the distance to the stream bank, the elevation above the stream bank (E), the downslope gradient, the revised compound topographic index (CTI), and the specific monodirectional and multidirectional catchment areas. The best correlations were observed between θ10 of winter 1996 and the physically based attributes E and CTI estimated by using the subsurface DEM (r = 0·83 and 0·86, respectively). Two multiple non‐linear regression models for θ10 spatial prediction were generated using non‐autocorrelated topographic attributes estimated from both surface and subsurface topography. Model validation using a new set of 41 data points showed root mean square errors (RMSE) lower than 10% of the θ10 range. The model based on subsurface topography decreased RMSE by 43%. Prediction errors were not spatially distributed. Finally, theses results are discussed in respect of processes involved in hillslope hydrology. Copyright © 2003 John Wiley & Sons, Ltd.