Credibility of statistical downscaling under nonstationary climate

摘要: Statistical downscaling (SD) establishes empirical relationships between coarse-resolution climate model simulations with higher-resolution variables of interest to stakeholders. These statistical relations are estimated based on historical observations at the finer resolutions and used for future projections. The implicit assumption is that SD relations, extracted from data stationary or remain unaltered, despite non-stationary change in climate. validity this relates directly credibility SD. Falsifiability projections a challenging proposition. Calibration verification, while necessary SD, unlikely be able reproduce full range behavior could manifest decadal century scale lead times. We propose design-of-experiments (DOE) strategy assess performance under nonstationary evaluate via transfer-function approach. relies selection calibration validation periods such they represent contrasting climatic conditions like hot-versus-cold ENSO-versus-non-ENSO years. underlying as warming predominance El Nino may more prevalent change. In addition, two different time identified, which resemble pre-industrial most severe emissions scenarios. ability generalize these proxy considered an indicator their nonstationarity. Case studies over climatologically disjoint study regions, specifically India Northeast United States, reveal robustness DOE identifying locations where nonstationarity prevails well role effective predictor