Impact of Bias in the Marine Air Temperature Observation Set on Atmospheric Reanalyses

1. Introduction

Marine air temperature observations are collected by volunteer observing ships carrying automatic meteorological sensor packages. Close proximity of the package mount to a dark colored ship’s deck is known to introduce a warm bias in these observations on sunny afternoons, in summer under low wind conditions (Berry and Kent, 2005, 2011). A version of this observation set is an influential constraint for the MERRA-2 reanalysis, as well as for several other reanalyses, which may influence the otherwise poorly sampled surface marine boundary layer. In this paper we explore the consequences of a possible diurnal bias in the marine air temperature observation set on estimates of surface air temperature as well as surface fluxes during the 36 year period 1980-2015.

2. Observations

Marine air temperature observations have mainly been collected along major shipping routes, with the densest coverage in the North Atlantic and North Pacific. In the first part of this presentation we review the historical observation coverage and compare air temperature and contemporaneous SST measurements to identify systematic differences (Fig. 1). The comparison suggests the presence of diurnal bias in the air temperature observations peaking in subtropics and midlatitudes in mid-afternoon local time.

3. MERRA-2 reanalysis increments and the tendency equation

In the second part of the presentation we explore the impact of the marine air temperature observations on MERRA-2 through analysis of the terms in the temperature forecast tendency equation, the equation which describes the contributions to changes in temperature during each analysis cycle.

The six-hour tendency of MERRA-2 atmospheric temperature (expressed in pressure coordinates) is balanced by five terms: a dynamical component representing advective processes, a thermodynamic component representing phase change and radiative absorption (Rad), a contribution due to turbulent exchange across constant pressure surfaces , and the analysis temperature increments discussed above:

dT/dt = -advection - thermodynamics - turbulent exchange + Rad + temp. incr. (1

(Bosilovich et al., 2016). Examination of (1) allows us to separate systematic errors due to such processes and radiation, mixing, and advection and to identify the major source of systematic forecast tendencies. For example, a nonzero time average of (1) indicates mean bias while a diurnal cycle indicates diurnal bias. This examination shows that the marine air temperature observations introduce an error that can reach 0.4K/dy, which peaks in local afternoon at subtropical latitudes (Fig. 2). This bias varies with season and gradually reduces in size after the mid-1990s.

4. Observing System Experiments

In the final part of the paper we present results from identical twin numerical experiments carried out with a version of the MERRA-2 analysis system. In the first twin we reproduce a few months of MERRA-2 during different decades with the complete set of in situ and satellite observations. In the second twin we repeat the experiment with the same initial conditions and data, but remove the marine air temperature data. The comparison of experiment twins, which are ongoing, confirm the impact of the error in marine air temperature. Further analysis should allow us to quantify the error introduced into the analysis of turbulent fluxes through errors in relative humidity and air/sea temperature difference.

5. Conclusions

This study uses information from the temperature tendency equation to clarify the presence of a systematic error in the MERRA-2 marine surface air temperature, which was introduced by ingestion of a somewhat biased version of the marine air temperature observation set. In an ongoing set of Observing System Experiments with the MERRA-2 analysis system we are exploring the impact on air temperature turbulent heat flux analyses of removing the biased observations. If time permits we will explore the impact of replacing the older somewhat biased marine air temperature observation set with the latest bias-corrected version of this data set.

References

Berry D.I., and Kent, E.C. (2005), The effect of instrument exposure on marine air temperatures: An assessment using VOSClim data. International Journal of Climatology 25: 1007– 1022, DOI: 10.1002/joc.1178.

Berry, D. I. and Kent, E. C. (2011), Air–Sea fluxes from ICOADS: the construction of a new gridded dataset with uncertainty estimates. Int. J. Climatol., 31: 987-1001. doi:10.1002/joc.2059

Bosilovich, M. G., R. Lucchesi, and M. Suarez (2016), MERRA-2: File Specification. GMAO Office Note No. 9 (Version 1.1), 73 pp, available from http://gmao.gsfc.nasa.gov/pubs/office_notes.

Figure Legends

Fig. 1 1980-2007 average of the monthly standard deviation of surface air temperature minus SST based on the latest release of the COADS data set. In the tropics and southern latitudes SST and surface air temperature are highly correlated and the variance is lower and so the standard deviation is low. However in the latitude band 30N-60N the standard deviation is quite high. It is in this latitude band that marine surface air temperature measurements will have a large impact on MERRA2.

Fig. 2 Comparison of the amplitude of the Fourier diurnal harmonic (computed using (1.2)) of MERRA-2 total temperature tendency at 950hPa (upper left) to the diurnal harmonic of temperature increments (upper right) and the diurnal harmonic of the other terms in (1.1). The same units and scales are used for all panels. The increments dominate the total diurnal tendency over the ocean between 30-60N. Computed over the full 36 year record.