Agriculture, Vol. 29 No. 2, 2023.
Original scientific paper
https://doi.org/10.18047/poljo.29.2.7
Evaluation of Ensemble Machine Learning for Geospatial Prediction of Soil Iron in Croatia
Dorijan Radočaj
; Josip Juraj Strossmayer University of Osijek, Faculty of Agrobiotechnical Sciences Osijek, Vladimira Preloga 1, 31000 Osijek, Croatia
*
Nedim Tuno
; University of Sarajevo, Faculty of Civil Engineering, Patriotske lige 30, 71000 Sarajevo, Bosnia and Herzegovina
Admir Mulahusić
; University of Sarajevo, Faculty of Civil Engineering, Patriotske lige 30, 71000 Sarajevo, Bosnia and Herzegovina
Mladen Jurišić
; Josip Juraj Strossmayer University of Osijek, Faculty of Agrobiotechnical Sciences Osijek, Vladimira Preloga 1, 31000 Osijek, Croatia
* Corresponding author.
Abstract
Soil fertility is pivotal for agricultural productivity, and iron (Fe) is a critical micro- nutrient essential for a successful crop development. This study investigates a potential of ensemble machine-learning methods in geospatial prediction of soil Fe in Croatia. Using a dataset of 686 soil samples, three individual machine-learning methods, including the extreme gradient boosting (XGB), support vector machine (SVM), and Cubist, as well as their ensemble, were evaluated for the soil Fe predi- ction. The ensemble method outperformed the individual models, exhibiting a higher prediction accuracy expressed by the coefficient of determination (R2 = 0.578), with a lower root-mean-square error (RMSE = 0.837) and the mean absolute error (MAE = 0.550). The soil clay content emerged as the most influential predictor, followed by the sand content, pH values, and select bioclimatic variables. This study’s results demonstrate the effectiveness of ensemble machine learning in an accurate predicti- on of soil Fe content and contribute to an informed decision-making in sustainable agricultural land-use planning and management. By including the complementary machine-learning methods into an ensemble with the representative environmental covariates, a geospatial prediction aids to a reliable comprehension of soil proper- ties and their spatial variability.
Keywords
soil samples; extreme gradient boosting; support vector machine; cubist; land-use planning
Hrčak ID:
311686
URI
Publication date:
19.12.2023.
Visits: 990 *