The choice of historical data product dominates climate uncertainty in projections of climate impacts in a 2-degree world

Abstract:

Estimates of the risk of climate change on society rely on historical estimates of true weather conditions and future projections from global climate models (GCMs), which are typically bias-corrected and downscaled before use. Future projections of climate impacts are affected by uncertainty in the underlying climate data through multiple pathways, only some of which are regularly accounted for in the literature. We investigate the importance of the choice of gridded historical data product used to fit impact models and bias-correct and downscale GCMs on the spread in projections of climate impacts. This decision is often either ad hoc in econometric climate impact studies or made for reasons orthogonal to a given product's performance for metrics and regions of interest, despite known limitations of any particular gridded product and difficulties in product evaluation in regions most vulnerable to climate damages.

We re-estimate three climate impact models from the literature, relating exposure to daily mean or max temperature to annual GDP per capita growth, mortality, and payroll, using four different reanalysis products. We then project damages for each dose-response function using a novel ensemble of GCM projections that accounts for all sources of climate uncertainty, bias-corrected and downscaled to the same four reanalyses to estimate this “observational” uncertainty, and incorporating all runs from multiple Large Ensembles of GCMs to estimate model uncertainty and internal variability. This Bias-Corrected and Downscaled Massive Ensemble (BCD-ME) allows us to partition uncertainty in damage projections between model, internal, and reanalysis sources.

We find that the choice of gridded historical data product dominates the spread in future projections of GDP per capita growth, mortality, and payroll at a given Global Warming Level for most parts of the globe, particularly in the mid-latitudes. Since in common practice this source of uncertainty is not considered, existing climate risk assessments likely underestimate uncertainty in future damages, underestimating the Social Cost of Carbon and possibly undercounting the possibility of plausible but extreme damages. We thus recommend that users of climate data test the sensitivity of their results to the choice of historical data product and use products that have been evaluated for the metrics and regions of interest whenever possible, and call for more research into constraining uncertainties about past estimates of the climate.

Draft available upon request.

Presented at EGU (Vienna, Austria, 7 May 2026).

Data: BCD-ME.