Analysis of different atmospheric physical parameterizations in COAWST modeling system for the Tropical Storm Nock-ten application

Abstract

A coupled ocean–atmosphere–wave–sediment transport modeling system was applied to study the atmosphere and ocean dynamics during Tropical Storm Nock-ten. Different atmospheric physical parameterizations in WRF model were investigated through ten groups of numerical experiments. Results of atmosphere, ocean wave and current features were compared with storm observations, ERA-Interim data, NOAA sea surface temperature data, AVISO current data and HYCOM data, respectively. It was found that the storm track and intensity are sensitive to the cumulus and radiation schemes in WRF, especially around the storm center area. As a result, using Kain–Fritsch cumulus scheme, Goddard shortwave radiation scheme and RRTM longwave radiation scheme in WRF may lead to much larger wind intensity, significant wave height, current intensity, as well as lower SST and sea surface pressure. Thus, they are not recommended for this study. Ocean parameters such as significant wave height, SST and current speed are more sensitive to Single-Moment 6-class microphysics scheme than to Eta microphysics scheme at the storm center. By analyzing modeled data with JASON-2 altimeter data, ERA-Interim data and HYCOM data in terms of fitting coefficient, root-mean-square error, correlation coefficient and model performance, the recommended atmospheric physical parameterization in this coupled system, have been obtained.

Appendix

Statistical analysis methods are applied to compare simulated data (X _n ) with observations (M _n ) quantitatively. Root-mean-square error (RMSE) is defined by

$${\text{RMSE}} = \left[ {\frac{1}{N}\mathop \sum \limits_{n = 1}^{N} \left( {X_{n} - M_{n} } \right)^{2} } \right]$$

(1)

The correlation coefficient (R) is given by

$$R = \frac{{\sum\limits_{n = 1}^{N} {\left( {X_{n} - \overline{{X_{n} }} } \right)\left( {M_{n} - \overline{{M_{n} }} } \right)} }}{{\sqrt {\sum\limits_{n = 1}^{N} {\left( {X_{n} - \overline{{X_{n} }} } \right)^{2} } \cdot \sum\limits_{n = 1}^{N} {\left( {M_{n} - \overline{{M_{n} }} } \right)^{2} } } }}$$

(2)

The formulation explained for the model performance (Willmott 1981) is as follows

$$S = 1 - \frac{{\sum\limits_{n = 1}^{N} {\left( {X_{n} - M_{n} } \right)^{2} } }}{{\sum\limits_{n = 1}^{N} {\left[ {\left( {X_{n} - \overline{{M_{n} }} } \right)^{2} + \left( {M_{n} - \overline{{M_{n} }} } \right)^{2} } \right]} }}$$

(3)

where $\overline{{X_{n} }}$ and $\overline{{M_{n} }}$ are the mean value of X _n and M _n . Both R and S range from 0 (bad) to 1 (good).