Comparative Assessment of LOCA2 and STAR-ESDM Downscaled Surface Temperature Over the Conterminous United States

For the conterminous United States, we compare two statistically downscaled climate datasets derived from CMIP6 multi-model ensembles: the Localized Constructed Analogs version 2 (LOCA2) and the Seasonal Trends and Analysis of Residuals Empirical-Statistical Downscaling Model (STAR-ESDM). Evaluating daily maximum (Tmax) and minimum temperature (Tmin), diurnal temperature range (DTR), temperature variability, annual extremes, and projection changes, we find that: 1) Both datasets demonstrate broad consistency with observations for the historical period and yield similar magnitudes and spatial patterns of projected climate change; 2) Each enhances local-scale features relative to raw CMIP6 outputs and effectively corrects large-scale biases, such as the historically underestimated DTR common in CMIP6 simulations; and 3) Both reduce inter-model spread in future climate projections compared to the original CMIP6 ensemble. Despite these common strengths, notable differences are observed: 1) Both datasets reflect uncertainties stemming from the observational products used in their training, particularly for Tmin and DTR; 2) LOCA2 produces a stronger fine-scale scale signal than STAR-ESDM; and 3) The difference between LOCA2 and STAR-ESDM is more pronounced for Tmin than for Tmax. These results underscore that training data and methodology influence downscaled outcomes, with LOCA2 generally better suited for fine-scale impact studies and STAR-ESDM for applications prioritizing regional coherence.