EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

Model Agreement and Disagreement in Rainfall Classification: A Region-Dependent Comparative Analysis of ARIMA and Random Forest across Punjab Districts, India

Vanshita Arora, Mohammad Shahfaraz Khan, Imran Azad, Amir Ahmad Dar, Amit Kishore Sinha, Mohammed Wamique Hisam

Research in Ecology · 03.08.2026 · 200-214

DOI: 10.30564/re.v8i4.13458 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Research in Ecology, online erschienen am 03.08.2026. Verfasst von Vanshita Arora und 5 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

For agricultural planning and management in monsoon-dependent regions, accurate classification of annual rainfall into risk categories is crucial. The predictive effectiveness of several models for rainfall forecasting has been compared in numerous studies, but the diagnostic significance of model disagreement has not been examined. This study aimed to analyse whether there is a link between the disagreement between autoregressive integrated moving average (ARIMA) and Random Forest (RF) and the difference in the accuracy of the predictions given by the two models in seven districts of Punjab, India (1970–2024). Prior to modelling, a pre-specified, data-driven district selection protocol was used, which was based on the coefficient of variation (CV) of rainfall, avoiding outcome-driven selection bias. The data were discretized into three categories (bins) with training-set bin edges to avoid data leakage. Fisher's exact test, Wilson score confidence intervals, and block permutation testing were used for statistical inference. In the five reliable districts (cells n ≥ 3), the prediction accuracy was significantly greater when the model disagreed with the data than when it agreed (0.83 vs. 0.52; Fisher's exact p = 0.022; block permutation p = 0.038). However, the Mansa district (dis.: 0.25; agree: 0.33) had a reverse pattern, indicating that the relationship was not pervasive. The findings are empirically consistent with the heuristic intuition of ensemble diversity theory but are not derivable from known theorems. The disagreement between ARIMA and RF can be used as a realistic uncertainty indicator for real-time rainfall monitoring. The findings are exploratory and hypothesis-generating, as there are small test sets per district.

Zitieren

APA

Arora, V., Khan, M. S., Azad, I., Dar, A. A., Sinha, A. K., & Hisam, M. W. (2026). Model Agreement and Disagreement in Rainfall Classification: A Region-Dependent Comparative Analysis of ARIMA and Random Forest across Punjab Districts, India. Research in Ecology, 200-214. https://doi.org/10.30564/re.v8i4.13458

Vancouver

Arora V, Khan MS, Azad I, Dar AA, Sinha AK, Hisam MW. Model Agreement and Disagreement in Rainfall Classification: A Region-Dependent Comparative Analysis of ARIMA and Random Forest across Punjab Districts, India. Research in Ecology. 2026:200-214. doi:10.30564/re.v8i4.13458

BibTeX
@article{Arora2026,
  author = {Arora, Vanshita and Khan, Mohammad Shahfaraz and Azad, Imran and Dar, Amir Ahmad and Sinha, Amit Kishore and Hisam, Mohammed Wamique},
  title = {Model Agreement and Disagreement in Rainfall Classification: A Region-Dependent Comparative Analysis of ARIMA and Random Forest across Punjab Districts, India},
  journal = {Research in Ecology},
  year = {2026},
  pages = {200--214},
  doi = {10.30564/re.v8i4.13458}
}
Kostenlos starten →