TESTING QUANTITATIVE SOIL-LANDSCAPE MODELS FOR PREDICTING THE SOIL HYDROMORPHIC INDEX AT A REGIONAL SCALE

Abstract

Quantitative soil-landscape models, based on topographic attributes, make possible the characterization of large areas because of the widespread availability of digital elevation models (DEMs). However, these soil-landscapes models, which are usually generated and validated on the same detailed, single research site, such as a hillslope or an elementary catchment, may show high prediction errors when applied to other areas of a region. The effect of the regional variations of topography, climate, parent material, land-use, and/or soil has seldom been analyzed. The objective of this study was to test multiple-regression relationships between the soil hydromorphic index (HI) and topographic attributes on different catchments of the Armorican Massif (30,000 km2) in western France. Regression models were validated using 565 data points collected from four sites along hillslopes. These four sites, located throughout this region, were included in three catchments with surface areas of 78, 82, and 120 ha, respectively, and differing in topography (mean elevation from 39 to 202 m and slope gradient from 3.4 to 7.9%), parent material (granite and schist), and precipitation (700 to 900 mm y−1). The existing models were multiple regression equations between the HI and the elevation above the stream bank, the compound topographic index (regression 1, r2 = 0.84), or the upslope drained area (regression 2, r2 = 0.86). At each validation point, systematic soil observation for HI estimation was compared with estimations from terrain attributes derived from DEMs at a 30-m resolution. Results showed small prediction errors for all study sites, with mean absolute errors between 5 and 15% of the HI range. Errors were not spatially correlated. Minimum prediction errors were encountered in the catchment for which the models were generated and also in one other that differed only in the parent material. In the other validation site, the models systematically overestimated HI. At the site of model generation, both regressions were accurate. However on the other sites, prediction errors using regression 2 were systematically higher than for regression 1, which uses a topographic index with a physical basis. These results revealed that soil-landscape models may be useful for predicting soil hydromorphy over a region but only when validated under several environmental conditions.