Scholarlink

<p>Over Northern Algeria, annual precipitation between 1981 and 2024 shows no statistically significant lineartrend when averaged across the domain, a result usually read as climatic stationarity. We probe that reading using 377 CHIRPS v2.0 cells (0.25°; 34.25–37°N; −2 to 8.25°E), checked against ERA5-Land (median KGE = 0.665; 83.3% of cells above KGE = 0.5). A Cancellation Index, CI = 1 − |mean β| / mean |β|, is proposed to measure how much per-cell trend signal disappears when cells are averaged together; bootstrap 95% confidence intervals accompany every estimate. The annual CI reaches 0.682 [0.568–0.807]: positive trends in the southern Tell (mean β = +1.26&#xa0;mm yr⁻¹) coexist with near-flat trends in the High Plateaux interior. CI peaks in DJF (0.774) and collapses in MAM (0.135), where 299 of 377 cells exhibit coherent positive Sen slopes (binomial sign test <i>p</i> &lt; 10⁻³¹, robust to spatial-autocorrelation correction). XGBoost–SHAP identifies longitude as the feature most strongly co-varying with the trend pattern (5-fold R² = 0.847), but 2-D leave-one-block-out validation fails (R² = −0.13); SHAP rankings are reported as within-sample co-variation, not causal attribution. The NAO drives a sign-reversing seasonal signal negative in DJF (<i>r</i> = − 0.385, p_BH = 0.040), suggestive in MAM/SON that cancels annually. An exploratory split at 2002 reveals progressively intensifying drying, consistent with a 2003–2014 wet recovery followed by sharp post-2014 drying. Domain-mean trends alone are insufficient over heterogeneous Mediterranean climates; reporting the Cancellation Index alongside seasonal stratification and dual-product validation provides a more transparent regional picture.</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

Hidden in the average: multi-scale rainfall trends behind apparent stationarity in Northern Algeria (1981–2024)

  • 期刊论文

  • 发表日期：

    2026年6月24日

查看全文
  • Fatima Zohra Tebbi,
  • Ouassila Bahloul,
  • Moufida Belloula,
  • Souhila Khemri,
  • Abdelmoumene Merdassi,
  • Nawel Karoui

摘要

Over Northern Algeria, annual precipitation between 1981 and 2024 shows no statistically significant lineartrend when averaged across the domain, a result usually read as climatic stationarity. We probe that reading using 377 CHIRPS v2.0 cells (0.25°; 34.25–37°N; −2 to 8.25°E), checked against ERA5-Land (median KGE = 0.665; 83.3% of cells above KGE = 0.5). A Cancellation Index, CI = 1 − |mean β| / mean |β|, is proposed to measure how much per-cell trend signal disappears when cells are averaged together; bootstrap 95% confidence intervals accompany every estimate. The annual CI reaches 0.682 [0.568–0.807]: positive trends in the southern Tell (mean β = +1.26 mm yr⁻¹) coexist with near-flat trends in the High Plateaux interior. CI peaks in DJF (0.774) and collapses in MAM (0.135), where 299 of 377 cells exhibit coherent positive Sen slopes (binomial sign test p < 10⁻³¹, robust to spatial-autocorrelation correction). XGBoost–SHAP identifies longitude as the feature most strongly co-varying with the trend pattern (5-fold R² = 0.847), but 2-D leave-one-block-out validation fails (R² = −0.13); SHAP rankings are reported as within-sample co-variation, not causal attribution. The NAO drives a sign-reversing seasonal signal negative in DJF (r = − 0.385, p_BH = 0.040), suggestive in MAM/SON that cancels annually. An exploratory split at 2002 reveals progressively intensifying drying, consistent with a 2003–2014 wet recovery followed by sharp post-2014 drying. Domain-mean trends alone are insufficient over heterogeneous Mediterranean climates; reporting the Cancellation Index alongside seasonal stratification and dual-product validation provides a more transparent regional picture.