Comparison between Global Latent Heat Flux Computed from Multisensor (SSM/I and AVHRR) and from In Situ Data

Abstract

The accurate estimate of the latent heat flux (LHF) is important to understand better the coupling between the atmosphere and the ocean and their respective circulation. In the near future, the availability of satellite-derived datasets over long periods will allow us to perform studies that, so far, have only been possible with historic in situ datasets. Therefore, a natural issue to explore is how the computation derived from both data types agree on LHF estimates. Comprehensive Ocean-Atmosphere Data Set (COADS) on one hand and satellite-derived parameters on the other hand are input to a similarity theory-based model and treated in completely equivalent ways to compute global latent heat flux. In order to compute latent heat flux exclusively from satellite measurements, an empirical relationship (Q-W relationship) is used to compute the air mixing ratio from Special Sensor Microwave/Imager precipitable water W and a new one is derived to compute the air temperature also from retrieved W (T-W relationship). First analyses indicate that in situ and satellite LHF computations compare within 40%, but systematic errors increase the differences up to 100% in some regions. By investigating more closely the origin of the discrepancies, the spatial sampling of ship reports has been found to be an important source of error in the observed differences. When the number of in situ data records increases (more than 20 per month), the agreement is about 50 W m-2 rms (40 W m-2 rms for multilayer averages). Limitations of both empirical relationships and W retrieval errors strongly affect the LHF computation. Systematic LHF overestimation occurs in strong subsidence regions and LHF underestimation occurs within surface convergence zones and over oceanic upwelling areas. The analysis of time series of the different parameters in these regions confirms that systematic LHF discrepancies are negatively correlated with the differences between COADS and satellite-derived values of the air mixing ratio and air temperature. To reduce the systematic differences in satellite-derived LHF, a preliminary ship-satellite blending procedure has been developed for the air mixing ratio and air temperature. The T-W relationship is not used any more and the air temperature is computed by adding the 3-yr-averaged COADS air-sea temperature difference to the satellite SST maps. The method to get the air mixing ratio is based on a weighted combination of COADS and satellite values according to the number of COADS observations available. After the blending process is applied, large improvements are observed in the Northern Hemisphere where both datasets are complementary. At midlatitudes, the blending procedure does not modify LHF values since satellite and COADS air mixing ratio do not differe by more than the expected satellite uncertainty (1 g kg-1). In the eastern and northern part of the basin, where the air mixing ratio difference is large and ship observations are numerous, blended LHF values are efficiently corrected toward in situ estimates compensating for the limitations of Q-W relationship. In the Southern Hemisphere, the number of in situ observations rarely exceeds four values per month, and without `'ground truth,'' more confidence is given to the satellite-derived values. Statistically, the rms difference drops to 28 W m-2 when 20 ship observations are available, which is likely a good approximation of the lowest bound of the blended LHF uncertainty compared with optimal in situ estimates. However, along the eastern boundaries in the southern oceans, local differences are expected to be much larger.