Quantifying Differences between 2-m Temperature Observations and Reanalysis Pressure-Level Temperatures in Northwestern North America

Abstract

AbstractIn northwestern North America, which is a large area with complex physiography, Climatic Research Unit (CRU) Time Series, version 2.1, (TS 2.1) gridded monthly mean 2-m temperatures are systematically lower than interpolated monthly averaged North American Regional Reanalysis (NARR) pressure-level temperatures––in particular, in the winter. Quantification of these differences based on CRU gridded observations can be used to estimate pressure-level temperatures from CRU 2-m temperatures (1901–2002) that predate the NARR period (since 1979). Such twentieth-century pressure-level temperature fields can be used in glacier mass-balance modeling and as an alternative to calibrating general circulation model control runs, avoiding the need for accurate boundary layer parameterization. In this paper, an approach is presented that is transferable to moisture, wind, and other 3D fields with potential applications in wind power generation, ecology, and air quality. At each CRU grid point, the difference between CRU and NARR is regressed against seven predictors in CRU (mean temperature, daily temperature range, precipitation, vapor pressure, cloud cover, and number of wet and frost days) for the period of overlap between CRU and NARR (1979–2002). Bayesian model averaging (BMA) is used to avoid overfitting the CRU–NARR differences and underestimating uncertainties. In cross validations, BMA provides reliable posterior predictions of the CRU–NARR differences and outperforms predictions from three alternative models: the constant model (24-yr mean), the regression model of highest Bayesian model probability, and the full model retaining all seven predictors in CRU.