STAR-ESDM: A New Bias Correction Method for High-Resolution Station- and Grid-Based Climate Projections

The Seasonal Trends and Analysis of Residuals (STAR) Empirical-Statistical Downscaling Model (ESDM) is a new bias correction and downscaling method that employs a signal processing approach to decompose observed and model-simulated temperature and precipitation into long-term trends, static and dynamic annual climatologies, and day-to-day variability. It then individually bias-corrects each signal, using a nonparametric Kernel Density Estimation function for the daily anomalies, before reassembling into a coherent time series.

Comparing the performance of this method in bias-correcting daily temperature and precipitation relative to 25km high-resolution dynamical global model simulations shows significant improvement over commonly-used ESDMs in North America for high and low quantiles of the distribution and overall minimal bias acceptable for all but the most extreme precipitation amounts (beyond the 99.9^th quantile of wet days) and for temperature at very high elevations during peak historical snowmelt months.

STAR-ESDM is a MATLAB-based code that minimizes computational demand to enable rapid bias-correction and spatial downscaling of multiple datasets. Here, we describe new CMIP5 and CMIP6-based datasets of daily maximum and minimum temperature and daily precipitation for nearly 10,000 weather stations across North and Central America, as well as gridded datasets for the contiguous U.S., Canada, and globally. In 2022, we plan to extend the station-based downscaling globally as well, since point-source projections can be of use in assessment of climate impacts in many fields, from urban health to water supply.

The projections have furthermore been translated into a series of impact-relevant indicators at the seasonal,  monthly, and daily scale including multi-day heat waves, extreme precipitation events, threshold exceedences, and cumulative degree-days for individual RCP/ssp scenarios as well as by global mean temperature thresholds as described in Hayhoe et al. (2018; U.S. Fourth National Climate Assessment Volume 1 Chapter 4).

In this presentation we describe the methodology, briefly highlight results from the evaluation and comparison analysis, and summarize available and forthcoming projections using this computational framework.