Improving soil hydromorphy prediction according to DEM resolution and available pedological data

This study analyses the sensitivity of soil hydromorphy prediction methods with regard to the resolution of topographical information and additional soil data. Seven Digital Elevation Models (DEM) were computed and compared to topographic measurements, with different resolutions (10, 20, 30 and 50 m) and construction mode (inputting actual stream location in addition to contour lines). Prediction models of soil hydromorphy using linear regression and co-kriging were established from detailed descriptions of soil catenas and topographical investigations on a 2 ha site. These models were compared on a validation set. The DEMs with fine resolutions from 10 to 30 m estimated in a unbiased way the elevation (E), the elevation above the stream bank (ES), the downslope gradient (DG), and the upslope contributing area (Amu), whereas prediction errors increased for the lower resolution 50-m DEMs. Apparently, the location of the channel network had no systematic effect on the estimation errors. There was a strong relationship between soil hydromorphy index (HI) and ES (r2=0.80) and the Compound Topographic Index (CTI)=ln(Amu/DG) (r2=0.62). For DEM resolutions of less than 30 m, soil hydromorphy prediction models bound on a regression model with topographic attributes appeared efficient and even better than ordinary kriging (OR) with 10 or 60 point observations. Coarser DEM resolutions (30 and 50 m) highly deteriorated prediction quality. For these resolutions, quality of soil hydromorphy prediction was highly improved by co-kriging of 10 and especially 60 pedological data points with a topographical regression model.