EUVIMEDEUVIMED entdecken
Crossref

Prediction of soil potassium forms using physicochemical properties and exchangeable potassium: I. Artificial neural network modeling

Manoocher Gholipoor

Dry Land Soil Research · 01.03.2025 · 1 (2)

DOI: 10.47176/dlsr.01.02.1042 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Dry Land Soil Research, online erschienen am 01.03.2025. Verfasst von Manoocher Gholipoor.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate prediction of soil potassium (K) fractions is critical for advancing precision nutrient management and promoting agricultural sustainability. This study aimed to develop artificial neural networks (ANNs) to predict multiple K fractions, including water-soluble, available (water-soluble + exchangeable), non-exchangeable, and total K prior to K amendment, and water-soluble, exchangeable, and fixed K following K amendment. Predictions were made using soil physicochemical properties such as clay, silt, sand, pH, organic carbon, cation exchange capacity, and electrical conductivity, along with measured exchangeable K and fertilizer-derived K. A comprehensive dataset, curated from peer-reviewed studies encompassing diverse global soils, served as the foundation for ANN development, ensuring broad applicability across different soil types and agroecological conditions. Multilayer perceptron ANNs with varying architectures were systematically optimized in MATLAB software, yielding high-fidelity models with robust predictive performance (correlation coefficients: 0.91–0.99). The complexity of the ANNs was tailored to the chemical dynamics of each K fraction. Simple architectures (8-8-1) effectively captured the distributions of water-soluble and fixed K, while more intricate configurations (8-10-10-10-1) were requisite for accurately modeling total K due to its complex interactions with soil matrices. Model validation confirmed high accuracy and reliability, with minimal mean squared error across all predicted fractions. To bridge the gap between research and practical application, these ANNs were embedded into an open-source, Excel-based tool, enabling seamless prediction of K fractions through user-friendly inputs of soil properties and measured exchangeable K data. This tool empowers farmers, agronomists, and researchers to optimize K fertilization strategies, reduce nutrient waste, and enhance crop productivity. The tool is accessible for download at: https://drive.shahroodut.ac.ir/index.php/s/fayE0zUH16TQe2M

Zitieren

APA

Gholipoor, M. (2025). Prediction of soil potassium forms using physicochemical properties and exchangeable potassium: I. Artificial neural network modeling. Dry Land Soil Research, 1(2). https://doi.org/10.47176/dlsr.01.02.1042

Vancouver

Gholipoor M. Prediction of soil potassium forms using physicochemical properties and exchangeable potassium: I. Artificial neural network modeling. Dry Land Soil Research. 2025;1(2). doi:10.47176/dlsr.01.02.1042

BibTeX
@article{Gholipoor2025,
  author = {Gholipoor, Manoocher},
  title = {Prediction of soil potassium forms using physicochemical properties and exchangeable potassium: I. Artificial neural network modeling},
  journal = {Dry Land Soil Research},
  year = {2025},
  volume = {1},
  number = {2},
  doi = {10.47176/dlsr.01.02.1042}
}
Kostenlos starten →