Identifying Sensitive Areas for Targeted Observations to Improve Indian Ocean Dipole Predictions Using a Coupled CNOP Approach

The Indian Ocean Dipole (IOD) is an interannual air-sea coupled phenomenon in the tropical Indian Ocean that significantly influences global weather and climate. Accurate prediction of the IOD is therefore of great importance. However, the current skill of IOD forecasting is limited, partly due to inaccuracies in initial conditions.  This study employs the coupled conditional nonlinear optimal perturbation (C-CNOP) method, which incorporates initial coupling uncertainties, to identify sensitive areas of targeted observations for positive Indian Ocean Dipole (IOD) events. Results show that the initial errors most likely to yield large prediction uncertainties of IOD events are mainly concentrated in sea temperatures near the thermocline in the eastern Indian Ocean (IO_Temp, 70-110m depth, 5°S-5°N, 85°E-105°E) and western Pacific (PO_Temp, 120-160m depth, 5°S-5°N, 130°E-150°E), as well as zonal winds (UWind), exhibiting an east–west dipole pattern over the tropical Indo-western Pacific. Through sensitivity experiments—designed to assess the impact of initial uncertainties in different areas on IOD predictions while bypassing the assimilation process and avoiding initial shock effects—we find that prediction uncertainties are more sensitive to initial errors in the UWind area than in the IO_Temp and PO_Temp areas, demonstrating a stronger impact on forecast skill, particularly in winter and summer. Further analysis demonstrated that the IO_Temp & PO_UWind (70-110m depth, 5°S-5°N, 85°E-105°E & zonal wind in the western Pacific) coupled area exhibits greater sensitivity than the UWind area alone, emerging as the most sensitive area of positive IOD events. This key area highlights both the Pacific's remote influence and the crucial role of local ocean on IOD development. These results underscore the critical role of coupled initialization in IOD predictability, offering a theoretical basis for advancing coupled data assimilation.