5:15 PM - 6:30 PM
[ACG30-P01] Evaluation of externally forced component on interannual-scale SST anomalies over the tropical Indo-Pacific Ocean
Keywords:global warming, externally forced component, sea surface temperature, ENSO/IOD/IOBW, tropical Indo-Paicifc Ocean
In order to investigate internal variability phenomena in the atmosphere-ocean climate system, it is necessary to remove the components generated by the external forcing such as anthropogenic global warming and volcanic eruptions (hereinafter referred to as externally forced component) from analyzed time series. In past studies, a method removing a linear trend at each grid point of the data has been widely used (hereinafter, LT method). On the other hand, in recent studies (e.g., Steinman et al. 2015; Dai et al. 2015), the new method of evaluating the externally forced component using the multi-model ensemble mean (MME) of the global mean surface temperature (GMST) anomaly using a large number of simulations in the Coupled Model Intercomparison Project Phase 5 (CMIP5) (hereinafter, GMST method) was applied to surface air temperature anomalies and sea surface temperature anomalies at each grid point by using the MME of GMST (GMSTmme). The purpose of this study is an inter-comparison between the LT and GMST methods when applying to interannual SST anomalies in the tropical Pacific and Indian Oceans.
In this study, we use the historical simulations with three ensemble members on a basis of the 12 CMIP5 models including MIROC5 (i.e., totally 36 historical experiments) to calculate the GMSTmme. After the linear regression coefficient (α) of the SST anomalies at each grid point onto the GMSTmme is calculated, the product of α and GMSTmme at each grid point is removed from the original SST anomalies. The remaining SST anomalies will be treated as the component of the natural internal variability.
We examined the different characteristics in the component of the natural internal variability in the Nino-3 index between the LT and the GMST methods on a basis of the observed SST of the COBE SST 2 (analysis period: 1850-2017). The monthly values of the Nino-3 index for the period 1960-1999 (2000-2017) tend to be 0.2℃ lower (higher) by the GMST method than by the LT method.
To quantify the different characteristics in the Nino-3 index between GMST and LT methods, the ratio of the root mean square deviation of externally forced components between the GMST and LT methods (RMSD) and the standard deviation of the internal variability component estimated by the GMST method (SD) was defined as R (R = RMSD/SD x 100). R of the Nino-3 index for a period of 1960-2017 is approximately 10%.
For the Indian Ocean Basin-wide warming (IOBW) index, the difference in amplitude between the GMST and LT methods was approximately 0.3°C, i.e., approximately 50% lager than that of Nino-3 index. The amplitude of the internal variability of the IOBW index was a fraction of that of the Nino-3 index. Therefore, R of the IOBW (approximately 50%) was much larger than that of Nino-3 index (10%). On the other hand, Indian Ocean dipole mode index (DMI) shows little difference between the two methods. Because the externally forced component are mostly equal in the western and eastern tropical Indian Ocean, it is mostly canceled out by definition of the DMI.
To investigate the different characteristics in natural internal SST variability between the two methods, R is calculated in the whole ocean basin in two analysis periods such as a long term during 1850-2017 and a short term during 1958-2017, respectively. R is 10 to 20% larger in the long-term analysis than in the short-term analysis for the whole ocean basin. When the GMSTmme by MIROC5 is employed for a longer period using the historical simulation and the RCP4.5 scenario simulation (1850-2100), R becomes much larger in all ocean basins. The results of this study indicate that the conventional LT method induces large estimation errors of the externally forced component by changing the calculation periods, suggesting usefulness of the GMST method.
In this study, we use the historical simulations with three ensemble members on a basis of the 12 CMIP5 models including MIROC5 (i.e., totally 36 historical experiments) to calculate the GMSTmme. After the linear regression coefficient (α) of the SST anomalies at each grid point onto the GMSTmme is calculated, the product of α and GMSTmme at each grid point is removed from the original SST anomalies. The remaining SST anomalies will be treated as the component of the natural internal variability.
We examined the different characteristics in the component of the natural internal variability in the Nino-3 index between the LT and the GMST methods on a basis of the observed SST of the COBE SST 2 (analysis period: 1850-2017). The monthly values of the Nino-3 index for the period 1960-1999 (2000-2017) tend to be 0.2℃ lower (higher) by the GMST method than by the LT method.
To quantify the different characteristics in the Nino-3 index between GMST and LT methods, the ratio of the root mean square deviation of externally forced components between the GMST and LT methods (RMSD) and the standard deviation of the internal variability component estimated by the GMST method (SD) was defined as R (R = RMSD/SD x 100). R of the Nino-3 index for a period of 1960-2017 is approximately 10%.
For the Indian Ocean Basin-wide warming (IOBW) index, the difference in amplitude between the GMST and LT methods was approximately 0.3°C, i.e., approximately 50% lager than that of Nino-3 index. The amplitude of the internal variability of the IOBW index was a fraction of that of the Nino-3 index. Therefore, R of the IOBW (approximately 50%) was much larger than that of Nino-3 index (10%). On the other hand, Indian Ocean dipole mode index (DMI) shows little difference between the two methods. Because the externally forced component are mostly equal in the western and eastern tropical Indian Ocean, it is mostly canceled out by definition of the DMI.
To investigate the different characteristics in natural internal SST variability between the two methods, R is calculated in the whole ocean basin in two analysis periods such as a long term during 1850-2017 and a short term during 1958-2017, respectively. R is 10 to 20% larger in the long-term analysis than in the short-term analysis for the whole ocean basin. When the GMSTmme by MIROC5 is employed for a longer period using the historical simulation and the RCP4.5 scenario simulation (1850-2100), R becomes much larger in all ocean basins. The results of this study indicate that the conventional LT method induces large estimation errors of the externally forced component by changing the calculation periods, suggesting usefulness of the GMST method.