Scholarlink

<p>This study predicted and characterized meteorological drought in the Sudano–Sahelian region of Nigeria (SSRN). Time series of rainfall, minimum, and maximum temperatures (1972–2022) were obtained from 11 stations from the Nigerian Meteorological Agency (NiMet). The degree of drought was determined via the standardized precipitation index (SPEI) and predicted via temporal convolutional networks (TCNs), linear regression, AdaBoost, and XGBoost, with NINO3.4, rainfall, minimum, and maximum temperatures as predictors. The fast Fourier transform (FFT) was used to extract the cyclical patterns of droughts. The results of the study revealed severe meteorological droughts in the 1970s and 1980s. The results also revealed that the number of droughts in the region ranged from 16 to 28, the average drought duration ranged from 5.39 to 9.19&#xa0;months, and the average drought severity ranged from −7.88 to 13.6&#xa0;months. The results of the FFT revealed that the dominant drought cycles were 24, 72, and 144&#xa0;months, which correspond to quasibiennial oscillation, ENSO, and AMO, respectively. Linear regression had the highest mean R<sup>2</sup> of 98%, whereas the TCNs had a mean R<sup>2</sup> value of 97% and a mean critical severity index (CSI) of 0.765. Specifically, the TCN reduced the MAE by 36–48% compared with AdaBoost and by 28–40% relative to XGBoost across most stations. Although linear regression occasionally recorded marginally lower MAEs and slightly higher R<sup>2</sup> values (e.g., Sokoto and Kano), the TCN exhibited greater generalization consistency across spatial domains, indicating better adaptability to complex climatic variability. This method aims to detect regions facing water stress and enhance water management and mitigation techniques by analysing the geographical and temporal distributions of drought characteristics.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Evaluation of temporal convolutional networks and ensemble machine learning models for meteorological drought prediction in the Nigerian Sudano–Sahelian zone

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月19日

查看全文
  • Muhammad Lawal Abubakar,
  • Auwal F. Abdussalam,
  • Zaharaddeen Isa,
  • Muhammad Sambo Ahmed,
  • Abubakar Sadiq Musa,
  • Jonah Birga,
  • Ismail Mohammed

摘要

This study predicted and characterized meteorological drought in the Sudano–Sahelian region of Nigeria (SSRN). Time series of rainfall, minimum, and maximum temperatures (1972–2022) were obtained from 11 stations from the Nigerian Meteorological Agency (NiMet). The degree of drought was determined via the standardized precipitation index (SPEI) and predicted via temporal convolutional networks (TCNs), linear regression, AdaBoost, and XGBoost, with NINO3.4, rainfall, minimum, and maximum temperatures as predictors. The fast Fourier transform (FFT) was used to extract the cyclical patterns of droughts. The results of the study revealed severe meteorological droughts in the 1970s and 1980s. The results also revealed that the number of droughts in the region ranged from 16 to 28, the average drought duration ranged from 5.39 to 9.19 months, and the average drought severity ranged from −7.88 to 13.6 months. The results of the FFT revealed that the dominant drought cycles were 24, 72, and 144 months, which correspond to quasibiennial oscillation, ENSO, and AMO, respectively. Linear regression had the highest mean R2 of 98%, whereas the TCNs had a mean R2 value of 97% and a mean critical severity index (CSI) of 0.765. Specifically, the TCN reduced the MAE by 36–48% compared with AdaBoost and by 28–40% relative to XGBoost across most stations. Although linear regression occasionally recorded marginally lower MAEs and slightly higher R2 values (e.g., Sokoto and Kano), the TCN exhibited greater generalization consistency across spatial domains, indicating better adaptability to complex climatic variability. This method aims to detect regions facing water stress and enhance water management and mitigation techniques by analysing the geographical and temporal distributions of drought characteristics.