engleski

Use of local vs. universal linear regression models for the prediction of water retention in common soils of eastern Croatia

In this study, 96 samples from 48 soil profiles (Gleysols, Chernozems, Anthrosols) on arable land in eastern Croatia were analyzed. The results were used to build local (specific) pedotransfer functions (PTFs) for predicting soil water retention at field capacity (FC) and wilting point (WP). The built multiple linear regression models are user-friendly and perform very well (RMSEs 2.7% and 1.8%, MAEs 2.1% and 1.5% ; for FC and WP, respectively) in spite having only the contents of organic C, clay and silt as predictors (selected as three basic properties for which the data is widely-accessible). Since in this study the soils on the most common parent materials in continental Croatia (fluvial deposits and loess derivates with mostly silt loam, silt clay loam, and silt clay textures) were considered, we expect our PTFs to be widely applicable. Prediction performance of the universal regression-based PTFs (EU-PTFs), developed by Toth et al. (2015) at the European scale, was also tested. Although Croatian soils are not part of the EU-HYDI database that was used for creating EU-PTFs, the universal model predicts the FC of the here-analyzed soils almost equally well (RMSE=3.3%, MAE=2.7%) as our local model. However, while predicting the WP of the local soils, the universal model was not as accurate – it overestimated WP (RMSE=4.9%, MAE=4.4%) more than the local model did. Because in many soils WP is usually low, such prediction errors may be unacceptable. Nevertheless, in some cases, the EU- PTFs can be used instead of local PTFs, at least in the European regions that have soilscapes similar to these considered herein. However, to increase EU-PTFs reliability, the soil database for their development should incorporate more European countries and regions.

field capacity, fluvial deposits, loess derivates, pedotransfer function, wilting point

nije evidentirano