HTML
--> --> -->Previous studies have analyzed the forecast skill of IOD events in different models and shown that the lead time for skillful predictions of IOD events is about one season, or sometimes two seasons for stronger ones (Wajsowicz, 2004, Wajsowicz, 2005; Luo et al., 2005, Luo et al., 2007; Shi et al., 2012). On the one hand, models are imperfect, and some cannot even simulate the basic characteristics of IOD events (Gualdi et al., 2003; Yamagata et al., 2004; Cai et al., 2005). That is, there are model errors in IOD predictions. On the other hand, errors in the initial field due to sparse observations is also a major limitation to the forecast skill (Luo et al., 2007; Feng and Duan, 2018). (Feng et al., 2014a) demonstrated that a winter predictability barrier (WPB) exists in IOD predictions, which is caused by the initial errors in the tropical Indian Ocean. The WPB indicates that, whatever the start month, the forecast skill of IOD events decreases rapidly across winter (Luo et al., 2007). As IOD events usually reverse the sign of the DMI and occur in winter (Wajsowicz, 2004; Fig. 1), the existence of the WPB probably causes the low forecast skill of the occurrence of an IOD event. Furthermore, (Liu et al., 2018) showed that, in addition to the WPB, there is also a significant summer predictability barrier (SPB) for IOD events, which is closely related with the initial errors in the tropical Pacific Ocean. The SPB indicates that, whatever the start month, the forecast skill decreases rapidly across the summer. As the DMI usually develops rapidly in summer (Wajsowicz, 2004; Fig. 1), the SPB greatly limits the forecast skill for the development of IOD events, and probably causes the low forecast skill of the IOD intensity
Figure1. Evolution of DMIs for 10 reference IOD events. E1-E10 signify the positive IOD events, with the model years 1, 3, 11, 20, 24, 59, 81, 88, 90, and 95, respectively. "-1" signifies the year preceding the IOD year; "0" signifies the IOD year; and "+" signifies the year following the IOD year.It has been demonstrated that only one third of IOD events are independent of El Ni?o-Southern Oscillation (ENSO) (Saji et al., 1999; Loschnigg et al., 2003; Stuecker et al., 2017), emphasizing the close relationship between the IOD and ENSO. When El Ni?o happens, the Walker circulation is weakened. Subsequently, the anomalous divergence in the Indo-Pacific warm pool further induces anomalous easterly winds near Sumatra. These easterly wind anomalies are favorable for upwelling in that region, and thus the appearance of a positive IOD event (Yu and Lau, 2005; Zhang et al., 2015). Therefore, ENSO is one important forcing that induces the appearance of IOD events (Li et al., 2003). Furthermore, ENSO also favors the appearance of the basin-wide wintertime surface warming in the tropical Indian Ocean under the combined effects of the anomalous latent heat flux and solar radiation, resulting in the decay of IOD events (Chowdary and Gnanaseelan, 2007). In short, ENSO affects not only the occurrence, but also the development, of IOD events. The role of ENSO should not be ignored when exploring the predictability of IOD events.
It has been previously shown that a WPB and an SPB exist in IOD predictions, which are respectively related to the initial errors in the tropical Indian and Pacific oceans (Feng et al., 2014a; Liu et al., 2018). In this study, we found that, in addition to the WPB, the initial errors in the tropical Indian Ocean also yield a significant SPB, which greatly affects the summertime predictions of IOD events. That is, not only the initial errors in the tropical Pacific Ocean, but also those in the tropical Indian Ocean could yield a significant SPB. Based on the above context, in this paper, we mainly discuss the role of initial errors in the tropical Indian Ocean, as well as the role of ENSO, on IOD predictions. In particular, the discussion will primarily focus on the SPB phenomenon. Due to the increasing frequency of occurrence under global warming and the larger climatic effects for positive IOD events compared with negative ones (Vinayachandran et al., 2002; Black et al., 2003; Annamalai and Murtugudde, 2004; Behera et al., 2005; Cai et al., 2009; Weller and Cai, 2013), the focus is restricted to positive events.
The remainder of the paper is organized as follows: The model and the experimental strategy are described in section 2. The predictability barriers caused by initial errors, as well as the influence upon them from ENSO, are analyzed in section 3. The spatial pattern of predictability barrier-related initial errors and their developmental physical mechanisms are explored in section 4. Finally, a summary and discussion is provided in section 5.
The GFDL CM2p1 coupled model is run for 150 years under the forcings of land cover, insolation, aerosols and tracer gases in 1990. Only the last 100 years are studied, to exclude the effects of the initial adjustment process in the first 50 years. Ten positive IOD events, i.e., where the DMI exceeds 0.5 for at least three consecutive months (Song et al., 2007), are randomly selected from the 100-year integration as the "true states", i.e., reference states, to be predicted. The DMIs of these IOD events generally reverse the sign in winter, develop in summer, peak in autumn, and decay rapidly in the following winter (Fig. 1), which is consistent with observational results (Wajsowicz, 2004).
Perfect model predictability experiments are conducted in this study. That is, the model is assumed to be perfect, and the prediction errors are only caused by the initial errors. As the ocean actively forces the atmosphere in the tropical oceans in most cases, the sea temperature only is perturbed to explore the effects of initial errors on IOD predictions. However, perturbing all levels of see temperature in the tropical Indian Ocean will cause incompatibility between the initial fields and the model, which further results in robust initial shock and conceals the effects of initial errors on IOD predictions, especially in the first few months. Therefore, we step back and only superimpose the initial errors on several important levels of sea temperature. The thermocline depth in the tropical Indian Ocean, which is closely related to the development of IOD events, is about 100-130 m in GFDL CM2p1 (Song et al., 2007). Therefore, the perturbations on the sea temperature at the depth of 95 m, which is close to the thermocline depth, would quickly affect the surrounding sea temperature and may reflect the variation of thermocline depth to some extent. Meanwhile, the sea surface temperature (SST) connects the atmosphere and the ocean, and is also closely related to the evolution of IOD events. Therefore, initial sea temperature errors are only superimposed on two levels of sea temperatures, i.e., the temperatures at the sea surface and at the depth of 95 m.
It is demonstrated that initial errors that cause large prediction uncertainties for ocean-atmosphere coupled phenomena, such as ENSO and the Kuroshio Large Meander, usually have a dynamic developmental behavior similar to these events themselves (Yu et al., 2009; Wang et al., 2012; Duan et al., 2013). Therefore, to explore the predictability barriers caused by the initial errors in the tropical Indian Ocean, we generate the initial errors from IOD-related sea temperature anomalies. Furthermore, the period of IOD events in GFDL CM2p1 is four years (Feng et al., 2014a). That is, there is generally a positive and a negative IOD event within four years. Hence, the sea temperature anomalies in the four years preceding each positive IOD event are plentiful and sampled every other month to generate lentiful initial errors. That is, there are 24 pairs of initial errors for each reference IOD event.
To impartially compare the effects of different initial errors on IOD predictions, all the initial errors are scaled to the same magnitude and then superimposed on the initial field of the reference-state IOD events. The original SSTAs and temperatures at 95 m deep are signified as T1 and T2, which are scaled by T'1=T1/δ1 and T'2=T2/δ2, respectively, where δ1 and δ2 signify positive values. T'1 and T'2 denote the surface and subsurface components of the scaled initial errors superimposed on the initial fields, respectively. The norms of the scaled initial errors, $\|T'_1\|=\sqrt{\sum_{i,j}(T'_{1,i,j})^2}$ and $\|T'_2\|=\sqrt{\sum_{i,j}(T'_{2,i,j})^2}$, are set as 2.4°C, where the grid point (i,j) covers the tropical Indian Ocean, i.e., (10°S-10°N, 45°-115°E). It is shown in (Kaplan et al., 1998) that the standard deviation of analysis errors for sea temperature near the equator is about 0.2°C, which is larger than the initial errors generated in this study in each grid. Therefore, these initial errors probably exist in analysis errors and the magnitude of our initial errors is reasonable. The spatial patterns of two samples of initial errors are shown in Fig. 2, as well as the spatial pattern of the initial field of one reference-state IOD event.
Figure2. Spatial pattern of the initial field of a given reference-state IOD event at (a) the sea surface and (b) 95 m deep, in the tropical Indian Ocean (left-hand column; units: °C). The middle and right-hand columns denote the spatial patterns of two samples of initial errors at (c, e) the sea surface and (d, f) 95 m deep, respectively.After superimposing the initial errors on the initial field of the reference-state IOD events, the predictions are started from July and October in the year preceding the IOD year with a lead time of 12 months. For simplicity, in the following discussions, the start months are signified as July(-1) and October(-1), respectively, where (-1) signifies the year preceding the IOD year. Therefore, we obtain 240 predictions for each start month. The prediction errors are then calculated as the absolute values of the difference in the DMI between the predicted IOD events and the "true state" of IOD events. As perfect model predictability experiments are carried out in this study, the prediction errors here are only caused by the initial errors. The growth rates of prediction errors η are calculated as \begin{equation*} \label{eq1} \eta=\frac{\partial E(t)}{\partial t}\approx \frac{E(t_2)-E(t_1)}{t_2-t_1} ,\nonumber \end{equation*} where E(t1) and E(t2) denote the prediction errors at times t1 and t2, respectively. The difference between t2 and t1 is one month in this study, which indicates that η signifies the monthly growth rates of prediction errors. A positive (negative) value of η indicates that the prediction errors increase (decrease). The larger the positive value of η, the faster the increase in the prediction errors.
Table 1 shows the existence of ENSO in the different developmental phases of 10 positive IOD events. In the winter of the growing phase, four IOD events are accompanied by a decaying El Ni?o or La Ni?a, with the remaining IOD events being independent of ENSO. In spring and summer, seven IOD events are accompanied by a developing El Ni?o or La Ni?a, with the remaining IOD events unassociated with ENSO. Therefore, the relationship between the IOD and ENSO varies in different developmental phases of IOD events. It is thus necessary to separately explore the effects of ENSO on IOD predictions according to the different developmental phases of IOD events.
Figure 3 shows the monthly growth rates of prediction errors for each reference IOD event based on 24 predictions. It is evident that the prediction errors develop rapidly in two main periods, i.e., the winter and the late spring/early summer in the growing phase of IOD events. Specifically, when the monthly growth rate is the largest or the second largest in winter (in late spring/early summer, i.e., in May), and the value of the monthly growth rate is larger than 0.2, a significant WPB (SPB) phenomenon occurs in IOD predictions. It is found that a significant WPB phenomenon exists in predictions of six reference IOD events. These reference IOD events, coincidentally, are independent of ENSO in the winter of the growing phase. In contrast, no WPB occurs in the predictions of the other four reference IOD events, which are accompanied by ENSO in winter. That is, when ENSO exists in the winter of the growing phase, it is not favorable for the occurrence of the WPB. A previous study demonstrated that ENSO favors the appearance of a basin-wide surface warming in winter under the combined effects of ocean dynamics, latent heat flux and solar radiation (Chowdary and Gnanaseelan, 2007). Here, it is found that ENSO also affects the development of temperature errors and favors a basin-wide mode of temperature errors in the tropical Indian Ocean in the winter of the growing phase (figure omitted), which might be based on a similar physical mechanism. This results in small prediction errors in winter, and thus no appearance of the WPB. In addition to the WPB, a significant SPB phenomenon exists in six reference IOD events, regardless of whether these IOD events are accompanied by ENSO in spring and summer. That is, there is no inherent connection between the occurrence of the SPB and ENSO. Based on the above discussions, the existence of ENSO is only closely connected with the occurrence of the WPB in IOD predictions.
Figure3. Monthly growth rates of prediction errors for 10 reference IOD events (units: month-1). Panels (a-j) denote 10 reference IOD events with the model years 1, 3, 11, 20, 24, 59, 81, 88, 90, and 95, respectively. Each letter on the horizontal axes denotes the first letter of different months. The vertical axes denote the growth rates of prediction errors. The four colors indicate the four seasons. The blue and brown bars with diagonal stripes correspond to the WPB and SPB in IOD predictions, respectively.The largest monthly growth rate of prediction errors in each prediction is labeled as ηmax and the second largest as ηs-max. If η max occurs in winter and the difference between ηmax and ηs-max is larger than 0.3, the corresponding initial errors are selected as those that are most likely to yield a significant WPB and defined as WPB-related initial errors. SPB-related initial errors are similarly selected. In total, there are 39 WPB-related initial errors and 31 SPB-related initial errors. The DMI errors (i.e., the difference in the DMI between the predictions and the corresponding reference IOD event) for these initial errors are shown in Fig. 4. The DMI errors for WPB-related initial errors are large (positive or negative) in winter, which results in large prediction uncertainties and causes the low forecast skill of IOD occurrence. For SPB-related initial errors, there are large negative DMI errors in summer, indicating that positive IOD events are generally predicted as weaker positive events, or even as negative events, causing the low forecast skill of IOD intensity.
Figure4. Time-dependent evolutions of DMI errors for initial errors that yield a significant (a) WPB and (b) SPB. Red lines denote the ensemble mean of prediction errors for WPB- and SPB- related initial errors, respectively. Each of the other colored lines denotes an individual prediction.Besides, the mean absolute value of the DMI errors (i.e., mean prediction error) for SPB-related initial errors in summer is larger than that for WPB-related initial errors in winter, indicating larger prediction uncertainties in summer.
2
4.1. Spatial patterns of initial errors that induce a significant SPB
As stated above, 31 SPB-related initial errors were selected in section 3. We put these initial errors together and assume them as time-varying fields corresponding to 31 different time points. Then, empirical orthogonal function (EOF) analysis is applied to these initial errors and the leading EOF mode (i.e., EOF1) is obtained, which describes the dominant spatial patterns of SPB-related initial errors. Corresponding to the EOF1 mode, the time series (i.e., PC1) has positive and negative values, indicating that some SPB-related initial errors have similar spatial patterns to the EOF1 mode and others have opposite ones. Therefore, the SPB-related initial errors are classified into two categories. The SPB-related initial errors that correspond to positive values of PC1 comprise the first category, and their composite is defined as type-1 initial error. Similarly, the composite of the SPB-related initial errors corresponding to the negative values of PC1 is defined as type-2 initial error.Type-1 initial error shows a significant west-east dipole pattern, both in the surface and subsurface components, with negative values in the western Indian Ocean and positive values in the eastern Indian Ocean (Fig. 5). The largest absolute values of type-1 initial error are concentrated in the subsurface component of the eastern tropical Indian Ocean. Type-2 initial error is almost opposite to type-1 initial error. Interestingly, the dominant spatial patterns of SPB-related initial errors in the tropical Indian Ocean are similar to those of the WPB-related initial errors analyzed in (Feng et al., 2017). The absolute values of spatial correlation coefficients between them are larger than 0.89. Therefore, initial errors with a west-east dipole pattern in the tropical Indian Ocean are inclined to yield a significant WPB and SPB, which results in large prediction uncertainties in winter and summer, and greatly limits the forecasting skill of the occurrence and the intensity of IOD events. (Feng et al., 2017) demonstrated that the subsurface eastern tropical Indian Ocean is the potential observing location (i.e., sensitive area) for advancing beyond the WPB for positive IOD events. That is, if intensive observations are carried out over this area, it will reduce the prediction errors in winter and weaken the WPB, improving the skill of wintertime IOD-event forecasts. Notably, the large values of SPB-related initial errors are also concentrated in the subsurface eastern Indian Ocean. That is, the initial errors in this area contribute greatly to the prediction uncertainties of positive IOD events. Therefore, if intensive observations are carried out over this area, it will not only weaken the WPB, but also reduce the prediction uncertainties in summer, and further weaken the SPB, ultimately greatly improving the forecasting skill with respect to the occurrence and intensity of positive IOD events.
Figure5. Spatial patterns of two types of SPB-related initial errors (units: °C). Panels (a, b) denote the surface and subsurface components for type-1 initial error; panels (c, d) denote the surface and subsurface components for type-2 initial error. Dotted areas indicate that the composites of temperature errors exceed the 95% confidence level, as determined by the t-test.Based on the above discussions, in addition to the initial errors in the tropical Pacific Ocean (Liu et al., 2018), the initial errors in the tropical Indian Ocean also play an important role in yielding a significant SPB. Therefore, to reduce the prediction uncertainties in summer and weaken the SPB, initial errors in the Indian Ocean should also be considered.
2
4.2. Developmental physical mechanisms of initial errors that induce a significant SPB
In this section, we calculate the sea temperature anomalies, which are defined as the difference in the sea temperature between the predictions and the "true state" of IOD events. Similarly, surface wind anomalies are also defined and calculated. Then, by analyzing the evolution of these temperature and wind anomalies, we identify how the SPB-related initial errors develop and cause large prediction uncertainties in summer, ultimately resulting in a significant SPB. Although type-1 and type-2 initial errors have opposite spatial patterns, the dominant characteristics of their evolution are only different in the first month and almost the same in the rest of the prediction year. Therefore, the evolution of the temperature and wind anomalies for two types of SPB-related initial errors is shown together in Fig. 6.
Figure6. Evolutions of the SSTA (units: °C) and sea surface wind anomaly (units: m s-1) over the tropical Indian and Pacific oceans (left-hand column), and the equatorial (5°S-5°N) subsurface temperature anomaly (units: °C; right-hand column) for SPB-related initial errors. Dotted areas indicate that the composites of temperature anomalies exceed the 95% confidence level, as determined by the t-test.When SPB-related initial errors are superimposed on the initial field of the reference-state IOD events, significant SST anomalies appear in the tropical Pacific Ocean in August, with positive values in the western Pacific Ocean and negative values in the central-eastern Pacific Ocean (Fig. 6). In the meantime, significant temperature anomalies appear in the subsurface Indian Ocean, with negative values in the western Indian Ocean and positive values in the eastern Indian Ocean. It is noted that the SST anomalies in the tropical Indian Ocean are relatively small in August and September. Therefore, the westerly wind anomalies at the equator in the tropical Indian Ocean may not be due to the local SST anomalies, but induced by the SST anomalies in the tropical Pacific Ocean. Specifically, the zonal SST gradient in the tropical Pacific Ocean causes easterly wind anomalies at the equator, which further modulate the Walker circulation in tropical oceans and induce anomalous westerly wind in the tropical Indian Ocean (Chen, 2011; Lian et al., 2014). Then, the anomalous westerly wind piles up warm water in the eastern Indian Ocean, and strengthens the subsurface warming there.
In the first half of the prediction year, the positive temperature anomalies in the subsurface eastern Indian Ocean indicate a deepening of the thermocline depth and an increase in the sea surface height (SSH). This will reduce the translation of the warm water from the western Pacific Ocean to the eastern Indian Ocean, resulting in the deepening of the thermocline depth and the warming of the subsurface ocean in the western Pacific Ocean. The positive subsurface temperature anomalies further propagate eastward to the eastern Pacific Ocean, and cause the warming of the ocean and the weakening of the negative SST anomalies there. In January-March, in response to the zonal gradient of the surface temperature anomalies in the tropical Pacific Ocean, westerly wind anomalies appear at the equator in the central-eastern Pacific Ocean, which further induce easterly wind anomalies in the tropical Indian Ocean by modulating the Walker circulation in tropical oceans. On the one hand, the anomalous easterly wind favors upwelling to the coast of Sumatra and Java, resulting in an elevation of the thermocline depth and a decrease in the SSH, which advances the translation of the warm water from the western Pacific Ocean into the eastern Indian Ocean. This further results in an elevation of the thermocline depth and negative subsurface temperature anomalies in the western Pacific Ocean, which propagate eastward to the eastern Pacific Ocean to cool the ocean there. On the other hand, the southeast wind anomalies in the eastern Indian Ocean are opposite in direction to the climatological wind, which decreases the total wind speed, and thus the release of the latent heat flux from ocean to atmosphere, warming the sea surface water there. With the eastward propagation of the negative temperature anomalies in the Pacific Ocean and the warming of the sea surface water in the eastern Indian Ocean, northwest wind anomalies appear in the Indian Ocean in April-June. In summer, the northwest wind anomalies are opposite in direction to the climatological wind in the eastern Indian Ocean, which decreases the total wind speed and thus the release of latent heat flux from ocean to atmosphere, finally warming the sea surface water. Therefore, a significant west-east dipole pattern of SST anomalies appears, which is further strengthened under the effect of Bjerknes feedback, ultimately resulting in a significant SPB.
Furthermore, we also analyze the evolution of sea temperature and surface wind anomalies for one sample of initial error that does not yield a significant SPB (Fig. 7). It is found that the SST anomalies grow fast and present a significant west-east dipole pattern in winter, which indicates large prediction errors in winter. In contrast, the SST anomalies show a basin-wide warming in summer, indicating small prediction errors in summer and thus no occurrence of an SPB.
Based on the above discussions, although initial errors are superimposed on the tropical Indian Ocean only, they further cause temperature and wind anomalies in the tropical Pacific Ocean. The interaction between the Indian and Pacific oceans plays an important role in yielding a significant SPB. That is, if there is no Pacific Ocean, and only the Indian Ocean exists in models, no SPB is likely to occur in IOD predictions, even though dipole-pattern initial errors exist in the tropical Indian Ocean. Therefore, the Pacific Ocean is indispensable in yielding a significant SPB.
Figure7. As in Fig. 6 but for one sample of initial error that does not yield a SPB.The WPB exists in predictions of six reference IOD events, which are independent of ENSO in the winter of the growing phase. In contrast, no WPB occurs in the predictions of the remaining reference IOD events accompanied by ENSO in winter. That is, ENSO in the winter of the growing phase is not favorable for the occurrence of the WPB. Therefore, the existence of ENSO in winter generally causes small prediction errors in winter, and is thus favorable for skillful forecasting of IOD occurrence. In addition to the WPB, an SPB also exists in IOD predictions, which has no direct interaction with ENSO.
There are two types of SPB-related initial errors. Type-1 initial error presents a west-east dipole pattern in both the surface and subsurface components, with the largest absolute values of errors concentrated in the subsurface eastern Indian Ocean. Type-2 initial error shows an opposite spatial pattern to type-1 error. The dominant spatial patterns of SPB-related initial errors are similar to those of WPB-related initial errors, with the largest absolute values concentrated within the same area. It was demonstrated in (Feng et al., 2017) that the subsurface eastern Indian Ocean is a sensitive area for advancing beyond the WPB. Therefore, if intensive observations are carried out in this area, not only will the WPB be weakened, but the prediction uncertainties in summer will also be reduced and the SPB weakened, greatly improving the forecasting skill of the occurrence and intensity of positive IOD events.
When SPB-related initial errors are superimposed on the initial field of the reference IOD events, significant SST anomalies appear in the tropical Pacific Ocean. By modulating the Walker circulation, anomalous westerly wind is induced in the tropical Indian Ocean, which piles up warm water in the eastern Indian Ocean, and thus causes a deepening of the thermocline depth and an elevation of the SSH in the eastern Indian Ocean. This further reduces the translation of the warm water from the western Pacific Ocean to the eastern Indian Ocean, resulting in the deepening of the thermocline depth and positive subsurface temperature anomalies in the western Pacific Ocean. The positive temperature anomalies propagate eastward to the eastern Pacific Ocean and warm the ocean there. Similarly, in the second half of the prediction year, the anomalous easterly wind appears in the tropical Indian Ocean, which causes upwelling and elevation of the thermocline depth in the eastern Indian Ocean and induces negative temperature anomalies in the western Pacific Ocean. The negative temperature anomalies propagate eastward and cool the sea water in the eastern Pacific Ocean. Furthermore, the anomalous easterly wind features wind of opposite direction to the climatology in the eastern Indian Ocean, which reduces the total wind speed and thus the release of latent heat flux from ocean to atmosphere, ultimately warming the sea water in the eastern Indian Ocean. In response to these SST anomalies, westerly wind anomalies appear in the tropical Indian Ocean and are opposite in direction to the climatological wind in summer, which warms the sea surface water in the eastern Indian Ocean by decreasing the release of latent heat flux from ocean to atmosphere, finally forming a significant west-east dipole pattern and thus a significant SPB.
(Liu et al., 2018) demonstrated that initial errors in the tropical Pacific Ocean are favorable for the occurrence of the SPB and identified its important role in yielding a significant SPB. Here, we find that, in addition to the initial errors in the tropical Pacific Ocean, those in the tropical Indian Ocean are also non-negligible. Of note is that, although the initial errors are superimposed on the tropical Indian Ocean only in this study, these initial errors further advance the occurrence of the SPB by modulating the temperature and wind anomalies in the tropical Pacific Ocean. Therefore, the role of the Pacific Ocean is indispensable in yielding an SPB.
Based on the above discussions, both the initial errors in the tropical Indian and Pacific oceans could induce a significant SPB. Therefore, to reduce the prediction uncertainties in summer and weaken the SPB, the initial errors in the Pacific and Indian oceans need to be reduced by carrying out intensive observations. Furthermore, due to the high degree of similarity in the spatial patterns between WPB-related initial errors and SPB-related initial errors in the tropical Indian Ocean, if intensive observations were to be carried out over the eastern subsurface Indian Ocean, this will not only reduce the prediction uncertainties in winter, but also in summer, thus weakening the WPB and SPB. Therefore, by increasing the intensity of observations in the tropical Indian and Pacific oceans, it will considerably reduce the prediction uncertainties and greatly improve the forecasting skill with respect to the occurrence and intensity of IOD events. Furthermore, (Sayantani et al., 2014) demonstrated that the Arabian Sea plays an important role in the evolution of IOD. Therefore, in addition to the initial errors in the tropical oceans, those in the Arabian Sea might also affect the IOD predictability, which needs to be further investigated and discussed in a future study.
