Reanalysis datasets provide spatially and temporally complete climate fields and are widely used as observational surrogates. Temperature trends from these datasets underpin numerous geoscience studies, yet the alignment of their directional trends with observations at the continental scale remains unclear. This study evaluates the sign of annual mean daily maximum (TMAX) and minimum (TMIN) temperature trends from three reanalysis data, viz. ERA5, MERRA-2, and NLDAS-2 against long-term Global Historical Climatology Network daily (GHCN) observations at 7059 (TMAX) and 6,983 (TMIN) stations across the continental United States (CONUS). We observe substantial trend misalignment between GHCN and reanalysis data, with ∼27%–31% of TMAX and ∼21%–30% of TMIN locations exhibiting discrepancies, mainly driven by false positive trends. Misalignment persists even for the longer records (⩾10–30 years) and is concentrated at stations with strong negative observed trends. Aggregating TMAX and TMIN trend agreement across different regions, including snow and rain-affected regions, western and eastern snow-dominated areas, urban and non-urban locations, and elevation classes, demonstrates that stations with trend misalignment prevail irrespective of the aggregation approach. NLDAS-2 dataset demonstrates a distinct behavior, with a greater percentage of stations exhibiting trend misalignment in snow-affected regions relative to rain-affected areas. Moreover, within the snow-affected domain, NLDAS-2 displays a contrasting spatial pattern, showing elevated misalignment across the western United States compared to other datasets. These findings call for caution while using reanalysis datasets for regional temperature trend analyses and derived applications.

Read original article

Read original article