Soil Management Effects on Soil Water Erosion and Runoff in Central Syria—A Comparative Evaluation of General Linear Model and Random Forest Regression
Safwan Mohammed,
Ali Al-Ebraheem,
Imre J. Holb,
Karam Alsafadi,
Mohammad Dikkeh,
Quoc Bao Pham,
Nguyen Thi Thuy Linh,
Szilard Szabo
Affiliations
Safwan Mohammed
Institution of Land Utilization, Technology and Regional Planning, University of Debrecen, Böszörményi út 138, 4032 Debrecen, Hungary
Ali Al-Ebraheem
Department of Soil Science, Faculty of Agriculture, Al-Bath University, Homs, Syria
Imre J. Holb
Institute of Horticulture, University of Debrecen, Böszörményi út 138, 4032 Debrecen, Hungary
Karam Alsafadi
Department of Geography and GIS, Faculty of Arts, Alexandria University, Alexandria 25435, Egypt
Mohammad Dikkeh
Department of Soil and water Science, Faculty of Agriculture, Tishreen University, Latakia, Syria
Quoc Bao Pham
Environmental Quality, Atmospheric Science and Climate Change Research Group, Ton Duc Thang University, Ho Chi Minh 700000, Vietnam
Nguyen Thi Thuy Linh
Institute of Research and Development, Duy Tan University, Danang 550000, Vietnam
Szilard Szabo
Department of Physical Geography and Geoinformatics, Faculty of Science and Technology, University of Debrecen, 4032 Debrecen, Hungary
The Mediterranean part of Syria is affected by soil water erosion due to poor land management. Within this context, the main aim of this research was to track soil erosion and runoff after each rainy storm between September 2013 and April 2014 (rainy season), on two slopes with different gradients (4.7%; 10.3%), under three soil cover types (SCTs): bare soil (BS), metal sieve cover (MC), and strip cropping (SC), in Central Syria. Two statistical multivariate models, the general linear model (GLM), and the random forest regression (RFR) were applied to reveal the importance of SCTs. Our results reveal that higher erosion rate, as well as runoff, were recorded in BS followed by MC, and SC. Accordingly, soil cover had a significant effect (p < 0.001) on soil erosion, and no significant difference was detected between MC and SC. Different combinations of slopes and soil cover had no effect on erosion, at least in this experiment. RFR performed better than GLM in predictions. GLM’s median of mean absolute error was 21% worse than RFR. Nonetheless, 25 repetitions of 2-fold cross-validation ensured the highest available prediction accuracy for RFR. In conclusion, we revealed that runoff, rain intensity and soil cover were the most important factors in erosion.
