Scholarlink

<p>Stationary vine (S-vine) copula models capture dependencies across both time points and variables for multivariate time series forecasting. This study applies S-vine copulas to model and forecast annual precipitation at 29 locations across Morocco using three configurations: local meteorological variables only, precipitation from climatically similar stations, and a hybrid model. Results show that optimal model choice varies by climate zone—arid regions favor spatial precipitation dependencies (Model 2 best for 73% of desert and semi-arid locations), while temperate coastal areas are better represented by local meteorological factors (Model 1 best for 50% of temperate stations). Critically, forecast quality relates directly to stationarity: stations with clear statistical stationarity (34% of locations, e.g., TANGER_AEROPORT, IFRANE) achieve out-of-sample R<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(^{2}\)</EquationSource> </InlineEquation> values of 0.01, while non-stationary stations (ERRACHIDIA, MIDELT) yield consistently negative R<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(^{2}\)</EquationSource> </InlineEquation> (-0.14 to -0.51). Among locations with inconclusive stationarity (59%), Model 2 achieves MAE reductions of 10-20 mm/year in arid zones. The S-vine structures reveal strong dependencies (Kendall’s tau up to 0.83) between stations in homogeneous climatic regions, captured through diverse bivariate copula families including Gaussian, Clayton, Gumbel, Joe, and BB7 effectively modeling asymmetric and tail dependencies characteristic of precipitation extremes. These findings demonstrate that S-vine forecast quality is systematically linked to precipitation regime stability, recommending preliminary stationarity assessment to guide model selection.</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

Using Stationary Vine-Copulas for Forecasting Precipitation

  • 期刊论文

  • 发表日期：

    2026年4月21日

查看全文
  • Wafaa El Hannoun,
  • Abdelhak Zoglat,
  • ElHadj Ezzahid

摘要

Stationary vine (S-vine) copula models capture dependencies across both time points and variables for multivariate time series forecasting. This study applies S-vine copulas to model and forecast annual precipitation at 29 locations across Morocco using three configurations: local meteorological variables only, precipitation from climatically similar stations, and a hybrid model. Results show that optimal model choice varies by climate zone—arid regions favor spatial precipitation dependencies (Model 2 best for 73% of desert and semi-arid locations), while temperate coastal areas are better represented by local meteorological factors (Model 1 best for 50% of temperate stations). Critically, forecast quality relates directly to stationarity: stations with clear statistical stationarity (34% of locations, e.g., TANGER_AEROPORT, IFRANE) achieve out-of-sample R \(^{2}\) values of 0.01, while non-stationary stations (ERRACHIDIA, MIDELT) yield consistently negative R \(^{2}\) (-0.14 to -0.51). Among locations with inconclusive stationarity (59%), Model 2 achieves MAE reductions of 10-20 mm/year in arid zones. The S-vine structures reveal strong dependencies (Kendall’s tau up to 0.83) between stations in homogeneous climatic regions, captured through diverse bivariate copula families including Gaussian, Clayton, Gumbel, Joe, and BB7 effectively modeling asymmetric and tail dependencies characteristic of precipitation extremes. These findings demonstrate that S-vine forecast quality is systematically linked to precipitation regime stability, recommending preliminary stationarity assessment to guide model selection.