Climate Model Data for Local Resilience: A New Downscaled and Bias-Corrected CMIP6 Synthesis to Better Characterize Evolving Climate Risks at Local and Regional Scales

We present a new downscaled and bias-corrected CMIP6 surface temperature synthesis designed to equip climate stakeholders in local and regional jurisdictions with tools to better predict extreme events and forecast risk at the community scale. This synthesis will form the backbone of an open-source interactive platform designed to smoothly connect historical climate and weather observations with projections of future climate states at regional scales, allowing for a better understanding of changes. The outputs of this project will effectively translate the full suite of the CMIP6 global climate models into a single easier-to-use and more reliable set of tools that convey the projected long-term changes, the evolving risk of extreme weather events, and the underlying uncertainties. This synthesis will enhance the ability of local communities to assess and respond to the future risks associated with climate change.

The synthesis integrates global surface temperature data from 40 GCMs and 374 runs, extending to the year 2100 across five SSP emission scenarios. Bias correction is performed by comparing reanalysis data from ERA5 (1940-present) and observations from Berkeley Earth's high-resolution dataset (1850-present) against historical model runs (1850-2015). The surface temperature curve from each GCM grid cell is decomposed into three components: (1) long-term climatology, (2) annual seasonality, and (3) daily weather variability. The climatology and seasonality are modeled using an effective 30-year moving window and rescaled based on historical observations. The daily weather variability is modeled using an evolving empirically-estimated probability distribution that explicitly models the low probability tails using generalized Pareto formalism. This allows the probability of temperature extremes to be immediately estimated. Probability distributions for weather events are also adjusted to match historical patterns.

This approach leverages the value of historical observations to minimize biases in future projections. Climatology, trends, and seasonality are adjusted to reflect historical data, with probability distributions recalibrated accordingly. The corrected data are then downscaled using statistical methods and machine learning techniques to produce estimates of past and future climate evolution that are useful for local and regional applications. This culminates in a synthesis, guided by historical accuracy, that summarizes the full suite of CMIP6 GCMs to accurately characterize future climate changes at local scales. This synthesis will provide a more robust, reliable, and easier-to-use set of tools for understanding past and future climate change.

The initial iteration of this synthesis focuses on the changes in minimum, maximum, and average temperatures for both long-term averages and short-term extremes. This will be followed shortly thereafter by a synthesis focusing on precipitation changes. The resulting data products will be made available in an open-source format that integrates the downscaled and bias-corrected synthesis data product with high-resolution historical observations and rapid attribution tools allowing for real-time analysis of current weather events. This analysis and the associated tools will help inform and empower local governments, policymakers, and other key stakeholders with the critical insights needed for effective climate resilience and adaptation planning.