How much confidence should we place in downscaled CMIP6 products for local extremes?
In late January 2026, my most recent paper was published in Climate Services, an international peer-reviewed journal. The paper explains the approach taken by my colleagues and I at UConn CIRCA to answering a very pressing question: how much confidence should we place in high-resolution, downscaled CMIP6 products when we use them for local extreme-weather analysis in Connecticut?
It is a vast topic, and this paper presents our first insights based on a limited subset of the available projections. We statistically compared multiple downscaled CMIP6 datasets from three independent research groups against observations between 1980 and 2024.
At first order, our findings suggest that reliability depends far more strongly on the climate variable in question than on the specific downscaling approach or underlying model. In other words, some variables are simply harder to project than others, across methods.
We found projections of daily minimum temperature (Tmin) to be reliable enough, in both summer and winter, to inform local decision-making. That matters because Tmin is important for summertime cooling and protection against heat stress. Daily maximum temperature and daily mean wind speed were projected with mixed reliability at local scale, while daily mean precipitation showed low reliability under our evaluation framework.
We then applied extreme value analysis to the reliable projections for the period 2015-2100. Among other findings, the analysis suggests that by late century, the 1-in-10-year hot day increases by roughly 3 °C under SSP2-4.5 and 4 °C under SSP5-8.5, with comparable shifts in the 1-in-100-year event. Put another way, temperature thresholds that were once relatively rare are projected to be crossed much more frequently, with obvious implications for heat stress, infrastructure and public health.
There is, however, an important caveat. Our research question was deliberately narrow, stemming from applied, town-scale use-cases. We therefore adopted a strict station-centric evaluation framework, rather than, for example, area averaging, which can be a valuable way of dealing with natural small-scale spatial variability such as the hit-or-miss nature of rainclouds. As a result, our findings do not preclude meaningful climate changes at broader regional scales, for which there is already strong evidence, including signals such as fewer but heavier rainfall events across New England.