EUVIMEDEUVIMED entdecken
Crossref

Projected rice yield responses to climate change scenarios in Ugandan smallholder systems using hybrid statistical–machine learning models

Juma Ndhokero, Lekia Nkpordee, Anumolu Goparaju

Agronomy Journal · 2026-07 · 118 (4)

DOI: 10.1002/agj2.70508 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Agronomy Journal, online erschienen am 04.08.2026. Verfasst von Juma Ndhokero und 2 weiteren Autor:innen. Die Arbeit nennt 55 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Rice ( Oryza sativa L.) production in Uganda is increasingly exposed to climatic variability, while yields remain constrained by interacting climatic and structural limitations. This study analyzed climate–yield relationships and projected rice yield responses under alternative climate scenarios using a hybrid statistical–machine‐learning modeling framework. A 30‐year district‐level dataset (1995–2024) of climate variables and rice yields from major rice‐growing regions of Uganda was analyzed using Autoregressive Integrated Moving Average Models With Exogenous Variables (ARIMAX), stochastic frontier analysis, Random Forest, and Support Vector Machine models. Climate perturbation scenarios analogous to Representative Concentration Pathways (RCPs) 4.5 and 8.5 were used to evaluate model robustness and simulate yield trajectories. Results showed that rice yield variability was strongly influenced by temperature‐related conditions, rainfall variability, and temporal persistence effects. Partial dependence analysis identified a nonlinear thermal threshold at approximately 21.8°C, beyond which predicted yields declined substantially. The hybrid ARIMAX–Random Forest model achieved the strongest predictive performance (root mean square error [RMSE] = 0.87 t ha − 1 ; mean absolute error [MAE] = 0.64 t ha − 1 ; R 2 = 0.45). Predictive performance deteriorated across all models under intensified climate perturbation, although the hybrid framework consistently maintained lower prediction errors than standalone approaches. Projected yield trajectories indicated greater productivity instability under conditions analogous to RCP 8.5 than under RCP 4.5. These findings demonstrate the value of hybrid modeling for climate‐sensitive crop forecasting and agricultural risk assessment in heterogeneous smallholder production systems. Strengthening climate‐smart agronomic adaptation, climate information services, and dynamic monitoring systems will be important for sustaining rice productivity under increasing climatic uncertainty in Uganda.

Zitieren

APA

Ndhokero, J., Nkpordee, L., & Goparaju, A. (2026). Projected rice yield responses to climate change scenarios in Ugandan smallholder systems using hybrid statistical–machine learning models. Agronomy Journal, 118(4). https://doi.org/10.1002/agj2.70508

Vancouver

Ndhokero J, Nkpordee L, Goparaju A. Projected rice yield responses to climate change scenarios in Ugandan smallholder systems using hybrid statistical–machine learning models. Agronomy Journal. 2026;118(4). doi:10.1002/agj2.70508

BibTeX
@article{Ndhokero2026,
  author = {Ndhokero, Juma and Nkpordee, Lekia and Goparaju, Anumolu},
  title = {Projected rice yield responses to climate change scenarios in Ugandan smallholder systems using hybrid statistical–machine learning models},
  journal = {Agronomy Journal},
  year = {2026},
  volume = {118},
  number = {4},
  doi = {10.1002/agj2.70508}
}
Kostenlos starten →