Evaluation of bulk surface flux algorithms for light wind conditions using data from the coupled ocean‐atmosphere response experiment (COARE)

Five bulk surface flux formulae algorithms, currently used in large‐and small‐scale atmospheric and coupled ocean‐atmosphere models, are tested with the same observations. Over the past decade many improvements have been made to the well‐known bulk formulae used for estimating surface fluxes. This paper concentrates upon those improvements that have centred upon light wind conditions. This condition is especially important because light winds are often found in the equatorial zone of tropical oceans, particularly in the western Pacific Ocean area known as the Pacific warm pool. the Pacific warm pool is important to the maintenance of the general circulation of the atmosphere and to the initial phases of the El Niño Southern Oscillation (ENSO) phenomenon. the input data to the bulk algorithms are mean quantities of 15 m air temperature, moisture, wind velocity, and 2 cm depth sea surface temperature obtained by the RV Moana Wave during the Coupled Ocean‐Atmosphere Response Experiment (COARE; 1 November 1992‐28 February 1993). the output from the bulk formulae are tested against 15 m eddy correlation flux observations which were also part of the RV Moana Wave near‐surface observational package. the RV Moana Wave was stationed in the Pacific warm pool for three periods during the COARE. Four formulae used exactly the same conditioning of the sea‐surface‐temperature input variable that was based upon upper ocean observations during the COARE. A surface renewal formulation had its own sea‐surface‐temperature adjustment scheme. the tests show that four of the five estimates of latent‐heat flux magnitude are within 8% of observed values. For sensible heat, about 10% of the latent‐heat flux magnitude, the range was 8‐23%, and for momentum the range was 8‐31% of observed values. Momentum flux was, as in the past, the most difficult to estimate. All approaches also used extensions to higher wind speeds. the test of those conditions was limited by a small dataset. It is shown that, for most of the algorithms tested, the bulk flux either overestimates or underestimates the covariance flux, depending upon the magnitude of the flux. In the COARE region this overestimation or underestimation is effectively a wind‐speed dependent bias in the model due to the surface flux parametrization. an error analysis indicates that there may be substantial errors in the bulk flux due to instrument uncertainties in the inputs to the formula. These two conclusions have an important implication concerning deterministic atmospheric modelling. Chaos theory and other approaches to the accuracy‐of‐prediction problem show that small differences in the initial value (or nudging) of the deterministic predictive integration can lead to widely varying (but coherent) results. Since surface fluxes are a major input to atmospheric, ocean, and coupled models, the consequence of this effective bias error in surface flux, a bias‐magnitude uncertainty due to natural variability, and errors associated with the inputs to surface flux parametrizations must be considered in determining the accuracy of deterministic models of the atmosphere and ocean. Otherwise the results of the integration, especially if long‐term, when compared with the real atmosphere may have wide differences. From these considerations, it is suggested that current general‐circulation models and climate models undergo surface flux sensitivity studies to test the effect of bias errors as discussed, and errors associated with inputs to the surface flux algorithms as well as natural variability.