Generating 3D soil surfaces from 2D height measurements to determine depression storage

Abstract

Surface depression storage is an important factor in preventing runoff and erosion in agricultural fields. Furthermore it is a highly dynamic property throughout the growing season. Maximum depression storage (MDS) is generally estimated from roughness indices derived from 2D transects, but these MDS estimations have a high uncertainty. The most reliable way of estimating MDS is by simulating the filling of a digital elevation model (DEM) of the soil surface using an algorithm. Yet, measurements of DEMs are often not available because they result from demanding equipment (laser roughness meters or photogrammetry). The aim of this study was to produce 3D DEMs with correct storage characteristics, using a simulation model that uses common 2D field elevation measurements as input. Two different types of models generating random surfaces, Gaussian and Boolean, were tested for their ability to simulate the soil surface of seedbeds. The models were tested on two experimental soil surfaces representing freshly tilled seedbeds. The DEM of these surfaces was constructed from laser measurements on a regular grid. Transects of these DEMs served as basis to derive the model parameters. Finally the models generated artificial DEMs. MDS was estimated on all DEMs using a filling algorithm included in the PCRaster software. The quality of the simulated surfaces was not only evaluated based on MDS, but also on the roughness index RR, the height distributions, the semivariograms, and their form. Results showed that only one Boolean model generated surfaces with the correct mean MDS values for the smooth soil surface. Almost all models overestimated MDS, although most had correct RR values. None of the models is capable of generating surfaces with realistic storage characteristics for both experimental surfaces. However, the method is promising because it is reproducible; simulated surfaces derived from a same set of parameters are very similar in MDS. When improved, this method is potentially more accurate than the statistical two-dimensional methods based on roughness indices.