Toward Anchoring Reanalysis with Absolute Data

We investigate the potential of assimilating spectral infrared data without bias correction into a four-dimensional variational assimilation system. Our investigation has two motivations: (1) the need to “anchor” data assimilation—and especially reanalysis—with data of trusted accuracy without bias correction, and (2) the possibility of developing a seamless weather-climate prediction system by tuning model physics to reduce data assimilation diagnostics. Modern atmospheric reanalysis systems have incorporated variational bias correction in order to deliver timeseries of atmospheric analyses that are unaffected by heterogeneity and bias in remote sensing instrumentation, yet the climate trends they produce disagree at the level of ~0.2 K per decade over the satellite era. Furthermore, improving a core model in numerical weather prediction may well be hindered by the presence of variational bias correction, resulting in the possible confusion of inaccuracy in data for error in model physics. To address both, we performed a numerical experiment using the four-dimensional variational assimilation system of ECMWF.

Anticipating the assimilation of satellite spectral infrared data without bias correction in the future, we compare first guess departures of a system run with perturbed radiative transfer in the forward model for HIRS channel 12 and another with perturbed vertical diffusion to a control run of the same system. The system absorbs the perturbation in radiative transfer for HIRS channel 12 into the bias correction for that channel, as it is designed to do, but it also absorbs the perturbation of vertical diffusion into the bias correction of all bias-corrected data types. The first guess departures of GPS radio occultation, though, do show a weak signature of the perturbation of vertical diffusion, but the forward model for GPS radio occultation is only marginally accurate enough to detect it. We conclude that, in order to better anchor reanalysis and diagnose errors in model physics, it is necessary to assimilate a complementary and redundant suite of trusted accurate data without bias correction and improve the accuracy of the radiative transfer forward models for the data.