EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

From static grids to dynamic response models: predicting wheat phosphorus needs with Random Forest

Amine Kassam, Lotfi Khiari, Ismail Kouera, Hamza Jouichat

Frontiers in Soil Science · 21.04.2026 · 6

DOI: 10.3389/fsoil.2026.1736046 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Frontiers in Soil Science, online erschienen am 21.04.2026. Verfasst von Amine Kassam und 3 weiteren Autor:innen. Die Arbeit wurde bisher 3-mal zitiert (Crossref) und nennt 79 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: OCP Group.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Introduction

Phosphorus (P) fertilizer recommendations for wheat are still largely based on soil test P (STP) classes and fixed fertilization grids. While simple to apply, these approaches often fail to reflect field-level variability and the combined influence of soil and climate factors on crop response. As a result, recommendations can be inconsistent and poorly aligned with actual yield potential. This study explores whether a data-driven, multifactorial approach can better explain and predict wheat yield response to P across diverse agroecological conditions.

Methods

We compiled a global meta-dataset including 294 field trials and 927 observations from 41 peer-reviewed studies. Yield response to P (ΔY = Y fertilized − Y unfertilized ) was modeled using a Random Forest (RF) algorithm. The model incorporated applied P rate, soil organic matter, soil pH (pH water ), annual precipitation, soil texture, P fertility class, and application method. Model performance was evaluated using separate training and testing datasets. RF-derived response curves were then used to estimate economically optimal P rates and compared with reported field optima.

Results

The conventional STP-based approach explained only a small portion of the observed yield response variability (R 2 = 5–9%) and produced highly variable recommendations across fertility classes. In contrast, the RF model showed strong predictive performance (R 2 = 89% for training and 78% for testing data) and reproduced site-specific response curves in 80% of cases. Economically optimal P rates derived from the RF model closely matched observed optima (R 2 = 93%, slope = 1.02).

Discussion

The findings indicate that wheat's response to P depends on several interacting factors that are not adequately represented in static STP-based recommendation systems. By jointly considering soil characteristics, climatic conditions, and management practices, the RF model generates more consistent and flexible response curves. This approach provides a practical basis for site-specific, economically optimized P management, advancing precision agriculture and more sustainable nutrient use.

Förderung

  • OCP Group

Zitieren

APA

Kassam, A., Khiari, L., Kouera, I., & Jouichat, H. (2026). From static grids to dynamic response models: predicting wheat phosphorus needs with Random Forest. Frontiers in Soil Science, 6. https://doi.org/10.3389/fsoil.2026.1736046

Vancouver

Kassam A, Khiari L, Kouera I, Jouichat H. From static grids to dynamic response models: predicting wheat phosphorus needs with Random Forest. Frontiers in Soil Science. 2026;6. doi:10.3389/fsoil.2026.1736046

BibTeX
@article{Kassam2026,
  author = {Kassam, Amine and Khiari, Lotfi and Kouera, Ismail and Jouichat, Hamza},
  title = {From static grids to dynamic response models: predicting wheat phosphorus needs with Random Forest},
  journal = {Frontiers in Soil Science},
  year = {2026},
  volume = {6},
  doi = {10.3389/fsoil.2026.1736046}
}
Kostenlos starten →