Univariate and Multivariate Downscaling for Climate Change Impact Assessment: A Comparative Study

Código

I-EBHE0056

Autores

Mohd Khairul Idlan Muhammad, Mansour Almazroui, Zulfiqar Ali, Shamsuddin Shahid

Tema

WG 1.14: Droughts in Mountain Regions

Resumo

Statistical downscaling methods bridge the gap between global climate models (GCMs) and local impact studies. Climate impact studies, such as reliable projections of drought hazards, necessitate the removal of biases in climate models. The drought impacts in the Himalayan mountainous regions of Pakistan need to be evaluated due to their vulnerability to frequent droughts. One of the bias correction approaches is univariate approaches, typically applied to individual target variables, which performs the bias correction while disregarding the model?s biases in inter-variable relationships. Multivariate bias correction addresses, on the other hand, address multiple climate variables simultaneously, either by considering their full dependence or assuming strong stationarity in the temporal order of model variables. This study comprehensively evaluates three multivariate bias correction methods (Pearson Correlation Dependence Structure (MBCp), N-dimensional Probability Density Function Transform (MBCn), and rank resampling for distributions and dependencies (R2D2)) against a univariate approach, quantile mapping (QM), for their effectiveness in correcting temperature and precipitation, preserving trends and assessing impacts on droughts and aridity in Pakistan including Himalayan regions. The results show that MBCn outperforms QM in capturing spatial and statistical characteristics of temperature and precipitation. MBCn exhibits superior performance in bias reduction, variance matching, and correlation between bias-corrected and observed datasets. However, QM excels in capturing the spatial patterns of drought frequency at different time scales (3-, 6-, and 12-month). QM also performed very well in preserving the significant trend of bias-corrected variables. This study highlights that the choice between univariate and multivariate bias correction methods depends on the specific research question and desired level of detail. MBCn bias correction is ideal for studies focusing on indices like the aridity index, which depends on long-term mean climate, while univariate bias correction excels in representing drought frequencies across temporal scales. Therefore, MBCn may be suitable for climate impact assessments of time-independent indices due to its strong performance in mean climate variables. Conversely, QM excels in climate change impact assessments on frequency-dependent indices, crucial for short and long-term adaptation planning. This study provides valuable guidance for climate research scientists to select bias correction methods based on the nature of climate indices in their impact studies.