MCDM-Based Ranking and Trend-Preserving Bias Correction of CMIP6 Models for Regional Downscaling over Northeast India

A reliable assessment of Impact, Adaptation, and Vulnerability of a region to climate change requires a high resolution climate information, particularly for regions with complex terrain. The North East Region (NER) of India, bounded by the eastern Himalayas in the north and the Bangladesh floodplains and Bay of Bengal in the south, is one such region. Several studies have identified NER as highly vulnerable to global warming, with six of its eight Indian states classified under high to moderate vulnerability. Dynamical downscaling for such regions necessitates the selection of a reliable Global Climate Model (GCM) and effective correction of its inherent biases. This study evaluates precipitation and 2 m air temperature from 16 GCMs available in the Coupled Model Intercomparison Project Phase 6 (CMIP6) over NER using eight performance metrics: Jensen–Shannon distance, mean absolute error, percentage bias, mutual information, correlation, Nash–Sutcliffe efficiency, Kling–Gupta efficiency, and root mean square error. These metrics collectively capture different aspects of model skill. The GCMs are subsequently ranked using two multi-criteria decision-making (MCDM) approaches: VIKOR and TOPSIS, both based on distance from an ideal solution but differing in their optimization philosophies. VIKOR and TOPSIS both use distance from ideal solution to rank with TOPSIS preferring the alternative closest to the ideal solution and VIKOR finding the alternative with the maximum group utility of the criterion and minimum individual regret. The ranking from both VIKOR and TOPSIS indicates MPI-ESM-1-2-HR and EC-Earth-Veg as the most reliable models annually and during monsoons over NER. In addition, the study also assessed a trend-preserving bias-correction framework for generating reliable initial and boundary conditions for regional dynamical downscaling, using MPI-ESM-1-2-HR as the primary driver. The method decomposes the climate time series into a non-linear trend and a perturbation component, with variance bias correction applied to the perturbations assuming the variance bias to be same for the future scenarios. To address uncertainty in long-term mean from single GCM simulation, the bias in trend is corrected for the multi-model ensemble (MME) mean trend,  derived from all 16 GCMs. Singular Spectrum Analysis (SSA) is employed to extract the non-linear trend due to its strong mathematical foundation and orthogonality properties. Preliminary results demonstrate the method’s effectiveness in correcting both long-term trend and variance biases, supporting its suitability for regional climate downscaling over NER.