Date of Award


Document Type


Degree Name

Doctor of Philosophy (PhD)


Ocean Engineering and Marine Sciences

First Advisor

Pallav Ray

Second Advisor

Charles Bostater

Third Advisor

Efthymios Nikolopoulos

Fourth Advisor

Gary Zarillo


An important component of the earth’s surface energy budget is the surface heat flux that allows the exchange of mass and energy between the ocean and the atmosphere and thereby influences oceanic and atmospheric circulations. For better prediction of weather and climate, numerical models must be able to capture the mean and variability of surface heat flux since surface heat flux directly feeds into model simulation of convection and precipitation. In this study, we explore the spatial and temporal variations of different surface heat flux components over the tropical oceans using observations and atmospheric global climate models (AGCMs). We confine our assessment to the tropical oceans (30°S-30°N) since tropics receives most solar radiation and precipitation, and the surface area of this latitudinal band is about 50% of the earth’s surface. We consider only AGMCs for evaluation since research related to the AGCMs is far fewer than coupled models. Yet, a better understanding of the reason behind the bias in surface heat flux in AGCMs can help to improve the coupled models. There are three main results from this study. First, the performance of 20 models participating in the Atmospheric Model Intercomparison Project Phase 5 (AMIP5) is evaluated concerning surface latent (QLH), sensible (QSH), net longwave (QLW) and net shortwave (QSW) over the tropical oceans (30°S-30°N) from 1979 to 2000. The comparison was conducted between AMIP surface heat flux and that from the Objectively Analyzed Air-Sea Fluxes (OAFlux) and in situ buoys. All 20 AMIP models overestimate QLH with an ensemble mean bias of 20 W m-2 , and 18 of the 20 models overestimate QSH with an ensemble mean bias of 5 W m-2 when compared to OAFlux, implying a systematic positive bias over the tropical oceans. The model ensemble mean of QSW was underestimated compared to the International Satellite Cloud Climatology Project (ISCCP) data by 4 W m-2 . On the other hand, QLW was overestimated by 4 W m-2 , leading to an underestimation of net surface radiation (Qrad) by 8 W m-2 . A comparison with buoy observations also showed similar biases. To obtain insights into the causes behind model bias, we quantified the contribution from near-surface winds, specific humidity, and temperatures on QLH and QSH biases. It is found that near-surface humidity contributes more to the bias in QLH than wind speed, while air temperature contributes more to bias in QSH than wind speed. On the other hand, the root mean squared error (RMSE) in QLH has contributions from both near-surface humidity and wind. The contribution from humidity to the mean bias in QLH is 13 W m-2 , with RMSE of 15 W m-2 , suggesting a systematic overestimation of sea-air humidity difference in models. The model ensemble, in general, simulates QLH and QSH better than individual models. Models with higher horizontal and vertical resolutions perform better than coarse resolution models. Second, for Qrad, the most prominent bias appears to be over the regions of low-level clouds in the off-equatorial eastern Pacific, eastern Atlantic, and south-eastern Indian Ocean. The RMSE in QLW was larger than that in QSW in 17 out of 20 AMIP5 models. Overestimation of total cloud cover and atmospheric humidity contributed to the underestimation of Qrad. In general, models with higher horizontal resolutions performed slightly better than those with coarser horizontal resolutions, although some systematic bias persists in all models and in all seasons, particularly in the regions of low-level clouds for QLW, and high-level clouds for QSW Third, the large bias in surface heat flux components in AMIP5 led us to explore whether the simulation of surface heat flux was improved in AMIP6 models that were released recently. The modelling groups worldwide have devoted much effort to further developing and improving model simulations. Therefore, validations of the newly released climate models will further help to detect weaknesses and the causes behind it that would lead to further improvement in the models and their predictions. To accomplish this, we use the OAFlux dataset and nine AMIP5 and nine AMIP6 models integrated over 30 years (1979-2008) for model-data comparisons. We choose to use AMIP5/AMIP6 models with same horizontal resolutions. This allows us to explore and quantify any improvement that would be independent of model horizontal resolutions. The AMIP6 models captures the annual mean and seasonal cycle better than AMIP5 models, yet, there remains large uncertainties among different AMIP models when comparing with OAFlux. For example, the RMSE and mean bias in QLH in AMIP5 are 30 and 21 W m-2 compared to 28 and 19 W m-2 in AMIP6. This is an improvement of 2 W m-2 for both RMSE and bias in QLH, yielding a reduction in QLH bias by 10% and RMSE by 7%. For QSH, an improvement of 1 W m-2 is seen in both bias and RMSE, yielding a reduction in QSH bias by 25% and RMSE by 13%. The primary reason for the improvement in QLH in AMIP6 is the better representation of the 10m wind speed and air-sea humidity difference than those in AMIP5 when compared against OAFlux. The improvement in QSH is due to an improvement in air-sea temperature difference. These results are expected to offer guidance to the modeling community in their efforts to improve model simulation of near-surface meteorological variables in the tropical oceans.