View in EDS

A Novel Statistical Weighting Approach for Multiple GCMs Ensembles to Enhance Drought Projections.

Saved in:
Bibliographic Details
Title: A Novel Statistical Weighting Approach for Multiple GCMs Ensembles to Enhance Drought Projections.
Authors: Ali, Farman1,2,3 (AUTHOR), Han, Jing-Cheng1,2,3 (AUTHOR) hanjc.2014@tsinghua.org.cn, Su, Dong2,3 (AUTHOR), Yousaf, Mahrukh4 (AUTHOR), Ali, Zulfiqar4 (AUTHOR), Rahman, Shafeeq Ur1 (AUTHOR), Huang, Yuefei5,6 (AUTHOR)
Source: Water Resources Management. Jul2026, Vol. 40 Issue 9, p1-21. 21p.
Subject Terms: *Statistical weighting, *Drought forecasting, *Climate change models, *Droughts, *Precipitation forecasting
Geographic Terms: Tibet (China)
Abstract: Ensembling multiple Global Climate Models (GCMs) reduces biases of individual models, leading to more accurate drought forecasts. However, traditional Multi-Modeling Ensembling (MME) can't account for volatility and non-linear relationships in GCMs, resulting in inaccurate precipitation estimates and unreliable drought assessments. This study introduces the hybrid GARCH-SVR weighting scheme for MME, which enables the simultaneous treatment of volatility and nonlinear predictive errors. This novel MME approach combines two models: Generalized Autoregressive Conditional Heteroscedasticity (GARCH) and Support Vector Regression (SVR). The GARCH captures variance and irregularity in time series through assigning weights to models based on their GARCH variances. Meanwhile, SVR models the complex nonlinear relationship and assigns weights to models based on their prediction error. As a result, a novel drought index, the Hybrid Weighted GARCH-SVR Drought Index (HWGSDI) is derived. HWGSDI improves drought assessment by addressing variance fluctuations and non-linearity in the precipitation time series from GCMs. The projection of this index is based on the relationship between observed and ensembled data using linear regression. Different comparative measures were used to evaluate the performance of the proposed weighted scheme, with results demonstrating that the Hybrid GARCH-SVR weighting scheme outperforms the Simple Model Average (SMA), Weighted Ensemble (WE) Scheme, GARCH, and SVR. Further, the Steady-State Probabilities of drought categories were spatially computed to understand the long-term behavior of drought in the Tibetan Plateau. The results showed that the probability of experiencing 'No Drought' remained relatively high, while the persistent probabilities of 'Extreme Drought' highlight a warning for the impacts of climate change. [ABSTRACT FROM AUTHOR]
Database: Energy & Power Source
Description
Abstract:Ensembling multiple Global Climate Models (GCMs) reduces biases of individual models, leading to more accurate drought forecasts. However, traditional Multi-Modeling Ensembling (MME) can't account for volatility and non-linear relationships in GCMs, resulting in inaccurate precipitation estimates and unreliable drought assessments. This study introduces the hybrid GARCH-SVR weighting scheme for MME, which enables the simultaneous treatment of volatility and nonlinear predictive errors. This novel MME approach combines two models: Generalized Autoregressive Conditional Heteroscedasticity (GARCH) and Support Vector Regression (SVR). The GARCH captures variance and irregularity in time series through assigning weights to models based on their GARCH variances. Meanwhile, SVR models the complex nonlinear relationship and assigns weights to models based on their prediction error. As a result, a novel drought index, the Hybrid Weighted GARCH-SVR Drought Index (HWGSDI) is derived. HWGSDI improves drought assessment by addressing variance fluctuations and non-linearity in the precipitation time series from GCMs. The projection of this index is based on the relationship between observed and ensembled data using linear regression. Different comparative measures were used to evaluate the performance of the proposed weighted scheme, with results demonstrating that the Hybrid GARCH-SVR weighting scheme outperforms the Simple Model Average (SMA), Weighted Ensemble (WE) Scheme, GARCH, and SVR. Further, the Steady-State Probabilities of drought categories were spatially computed to understand the long-term behavior of drought in the Tibetan Plateau. The results showed that the probability of experiencing 'No Drought' remained relatively high, while the persistent probabilities of 'Extreme Drought' highlight a warning for the impacts of climate change. [ABSTRACT FROM AUTHOR]
ISSN:09204741
DOI:10.1007/s11269-026-04516-3