Dynamical Extended-Range Prediction of Early Monsoon Rainfall over India

Frédéric Vitart European Centre for Medium-Range Weather Forecasts, Reading, United Kingdom

Search for other papers by Frédéric Vitart in
Current site
Google Scholar
PubMed
Close
and
Franco Molteni European Centre for Medium-Range Weather Forecasts, Reading, United Kingdom

Search for other papers by Franco Molteni in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

The 15-member ensembles of 46-day dynamical forecasts starting on each 15 May from 1991 to 2007 have been produced, using the ECMWF Variable Resolution Ensemble Prediction System monthly forecasting system (VarEPS-monthy). The dynamical model simulates a realistic interannual variability of Indian precipitation averaged over the month of June. It also displays some skill to predict Indian precipitation averaged over pentads up to a lead time of about 30 days. This skill exceeds the skill of the ECMWF seasonal forecasting System 3 starting on 1 June. Sensitivity experiments indicate that this is likely due to the higher horizontal resolution of VarEPS-monthly. Another series of sensitivity experiments suggests that the ocean–atmosphere coupling has an important impact on the skill of the monthly forecasting system to predict June rainfall over India.

Corresponding author address: Frédéric Vitart, European Centre for Medium-Range Weather Forecasts, Shinfield Park, Reading RG2 9AX, United Kingdom. Email: nec@ecmwf.int

Abstract

The 15-member ensembles of 46-day dynamical forecasts starting on each 15 May from 1991 to 2007 have been produced, using the ECMWF Variable Resolution Ensemble Prediction System monthly forecasting system (VarEPS-monthy). The dynamical model simulates a realistic interannual variability of Indian precipitation averaged over the month of June. It also displays some skill to predict Indian precipitation averaged over pentads up to a lead time of about 30 days. This skill exceeds the skill of the ECMWF seasonal forecasting System 3 starting on 1 June. Sensitivity experiments indicate that this is likely due to the higher horizontal resolution of VarEPS-monthly. Another series of sensitivity experiments suggests that the ocean–atmosphere coupling has an important impact on the skill of the monthly forecasting system to predict June rainfall over India.

Corresponding author address: Frédéric Vitart, European Centre for Medium-Range Weather Forecasts, Shinfield Park, Reading RG2 9AX, United Kingdom. Email: nec@ecmwf.int

1. Introduction

The prediction of Indian rainfall in June represents a particularly difficult challenge. During the month of June, the level of Indian rainfall is strongly linked to the onset of the monsoon, which is usually difficult to predict more than a couple of weeks in advance. Although, there have been clear improvements over the last decades in the short- and medium-range predictions thanks to better, higher-resolution numerical models, improved data assimilation systems, and the extension of ensemble techniques, progress of extended-range dynamical forecasts has been difficult to ascertain. Indeed, Gadgil et al. (2005) expressed concerns about the ability of currently available dynamical systems to produce effective predictions of the monsoon rainfall. Xavier and Goswami (2007) claimed that statistical models of rainfall evolution over a 2–4-week time scale may provide more useful information than seasonal-mean estimates from coupled GCMs. For example, Webster and Hoyos (2004) show significant skill in predicting monsoon rainfall on the 20–35-day time scale using a statistical model.

The skill of the European Centre for Medium-Range Weather Forecasts (ECMWF) seasonal forecasting System 3 (hereafter referred to as System 3; Anderson et al. 2007) to predict monsoon rainfall has been evaluated by Molteni et al. (2008). They found that this dynamical seasonal forecasting system displays some skill in predicting the monthly-mean precipitation over India after July, but has surprisingly low skill to predict the June precipitation over India. For instance the linear correlation between the June precipitation predicted by the seasonal forecast starting in May (forecast range: days 32–61) and the Global Precipitation Climatology Project dataset (GPCP; Adler et al., 2003) is only 0.29 for the period 1991–2005, and only 0.5 for the seasonal forecasts starting on 1 June (forecast range: days 1–30), which are available to the public at 1200 UTC 15 June. The 15.5-day delay allows the acquisition of sea surface temperature (SST) fields from the previous month, time to run the forecast, and a margin to ensure a reliable operational schedule. Since System 3 is not successful to predict the June precipitation over India, a series of experiments has been set up to see if the ECMWF Variable Resolution Ensemble Prediction System monthly forecast system (VarEPS-monthly; Vitart et al. 2008; available with a delay of 22 h), which has a higher horizontal resolution than System 3 and more updated physics, was more successful.

So far, most of the evaluations of the skills of dynamical models to predict monsoon rainfall involved either medium-range forecasting systems or seasonal forecasting systems (e.g., Someshwar Das et al. 2002; Sperber et al. 2001; Kang and Shukla 2006). This is mostly due to the fact that monthly forecasting is often considered a difficult time range for weather forecasting and few operational centers produce forecasts for this time range. Therefore, the first goal of the present paper is to document the skill of the current operational ECMWF monthly forecasting system, extended to 45 days (instead of 32 days in operations) and show that dynamical models can be useful for the prediction of monsoon rainfall 2–6 weeks in advance. A second objective of this paper is to display the results obtained with a few sensitivity experiments, which were designed to evaluate the impact of horizontal resolution and ocean–atmosphere coupling on the skill of the dynamical model to predict monsoon rainfall.

After this introduction, section 2 will describe the setting of the experiments in more detail. Section 3 will discuss the data we used to verify the forecasts. Section 4 will display the skill of the dynamical model to predict all India rainfall averaged over June, along with the forecasts obtained with the various sensitivity experiments. The skill of the monthly forecasting system to predict pentad-mean, area-averaged Indian rainfall will be addressed in section 5. Section 6 will discuss the impact of the ocean mixed layer model on the forecasts of Indian precipitation rainfall. Finally section 7 will summarize the results of this paper.

2. Experiment setting

This paper displays the results obtained with four different experiments that will be referred to as VEPS [high-resolution atmospheric model coupled to an ocean general circulation model (OGCM)], PERS (high-resolution atmospheric model forced by persisted SST anomalies), MOFC (low-resolution atmospheric model coupled to an OGCM), and ML (low-resolution atmospheric model coupled to an ocean mixed layer model). The comparison of VEPS with PERS will help to assess the role played by the ocean–atmosphere coupling. A comparison between VEPS and MOFC forecasts will help to evaluate the impact of the atmospheric horizontal resolution on the prediction of Indian rainfall. Finally the comparison between MOFC and ML will help to assess the impact of the oceanic vertical resolution on the Indian precipitation forecasts. The four experiments will be described in more detail below.

Each experiment consists of 15-member ensemble integrations starting on each 15 May from 1991 to 2007. The current integrations of the operational monthly forecasting system are 32 days long. The integrations in this series of experiments have been extended to 46 days to cover the full month of June. The integrations have been repeated 15 times for each starting date, with slightly different initial conditions to represent the errors in the initial conditions. The atmospheric perturbations are produced using the singular vector method (Buizza and Palmer 1995). They include perturbations in the extratropics and perturbations in some tropical areas by targeting tropical cyclones (Puri et al. 2001). The oceanic initial conditions in the coupled runs are perturbed the same way as in System 3: SST initial conditions are perturbed by adding random perturbations, which are constructed by taking the difference between two SST analyses. In addition, wind perturbations are also applied during the ocean data assimilation to produce five slightly different oceanic initial conditions. The wind perturbations are also computed from the difference between two wind analyses (see Vialard et al. 2005 for more details). To take into account the uncertainties in the model subgrid-scale parameterizations, the tendencies in the atmospheric physics are randomly perturbed during the model integrations (Buizza et al. 1999). The experiments have been performed with the ECMWF atmospheric model Integrated Forecast System (IFS) Cycle 32R2, which was operational from June to November 2007.

VEPS has the same configuration as the new version of the monthly forecasting system, which has been operational since March 2008. In this configuration, the atmospheric model has a horizontal resolution of TL399 (about 60 km) until day 10 and TL255 (about 80 km) afterward. The atmospheric model is forced by persisted SST anomalies until day 10 and coupled to an OGCM at day 10. The OGCM is the Hamburg Ocean Primitive Equation (HOPE) model from the Max Plank Institute (Wolff et al. 1997). The ocean initial conditions at day 10 are taken from the last day of a 10-day OGCM integration, which is initialized from the ECMWF analysis (Balmaseda et al. 2008) and forced by the fluxes from the atmosphere-only integration. A detailed description of the VEPS system can be found in Vitart et al. (2008). PERS has the same configuration as VEPS, but this time, the atmospheric model is forced by persisted SST anomalies during the whole integration.

MOFC has a lower horizontal resolution than VEPS-T159 (about 120 km). The atmosphere and ocean are coupled from day 0. This configuration is similar to the previous version of the ECMWF monthly forecasting system (see Vitart 2004) and to System 3, but for the model physics and the hourly ocean–atmosphere coupling instead of daily coupling in System 3.

The ML is identical to MOFC, but, this time, the atmospheric model is coupled to an ocean mixed layer model rather than to an OGCM. This experiment was motived by results from Shinoda and Hendon (1998), Bernie et al. (2005), and Shinoda (2005) who have shown in 1D mixed layer modeling experiments forced by observed fluxes from the Woods Hole Oceanographic Institution (WHOI) mooring during the Tropical Ocean and Global Atmosphere Coupled Ocean–Atmosphere Response Experiment(TOGA COARE) that the modulation of the diurnal cycle of SST by the intraseasonal variability in surface fluxes has a significant impact on the intraseasonal variability in SST. In particular, using the TOGA COARE data, Bernie et al. (2005) found that failing to properly represent the diurnal cycle of mixing reduced the intraseasonal variability by as much as 30%. They concluded that a vertical resolution of about 1 m in the upper ocean and a coupling frequency of at leat 3 h are necessary to capture 95% of the intraseasonal variability. Woolnough et al. (2007) have shown that an increase of the vertical resolution of the upper ocean improved the prediction of the Madden–Julian oscillation in the ECMWF monthly forecasting system. A comparison of ML and MOFC will show if this is also the case for the Indian precipitation in June.

The mixed layer model is the same as the one described in Woolnough et al. (2007). It is based on the K-profile parameterization (KPP) vertical mixing scheme of Large et al. (1994). It has a vertical domain of 200 m with 29 vertical levels. The vertical grid is stretched so that the top model level is 1.4 m thick, instead of 10 m in the OGCM, with 16 levels in the top 30 m (instead of 3 levels in MOFC). In the horizontal direction, the mixed layer model uses only half the resolution of the full dynamical model. The mixed layer model extends to 44° of latitude, poleward of which the SST is provided by persistence of the initial conditions. To produce a smooth transition between the fully coupled and uncoupled regions, the SST passed to the atmosphere between 40° and 44° is a weighted average of the modeled SST and the initial SST. Table 1 summarizes the four experiment settings.

3. Verification data

The skill of the numerical model to predict the monsoon precipitation over India is assessed by computing the precipitation averaged over the domain (5°–30°N, 70°–90°E), which includes most of the Indian subcontinent. The area averaging is performed only over land points. The precipitations predicted by the model are averaged over June and a series of 5-day periods (pentads) in June.

In sections 4 and 5, the model predictions are validated against an analysis from Indian station data (Rajeevan et al. 2006). However, the 40-yr ECMWF Re-Analysis (ERA-40) will be used for verification in section 6, since the diagnostics in this section require values over sea points. The Indian station data, available in a 1° × 1° grid is based on the outputs of a few hundreds stations covering India. Figure 1 shows the interannual variability of June precipitation from this dataset, along with the interannual variability obtained from two other possible verification data: GPCP and 24-h forecast of precipitation from ERA-40 (Uppala et al. 2005) until 2002 followed by ECMWF operational analysis. The three datasets show some broad agreement in terms of interannual variability. There are, however, some significant differences with ERA-40 displaying too little precipitation in 1993 and 1994 relative to the two other datasets. The three datasets disagree in June 1997: according to GPCP there was more precipitation than average that year, whereas the station dataset indicates a dryer June 1997 than usual. However, none of the datasets displays the strong negative anomaly, which would be expected as a response to the El Niño event of the century, as already discussed in several studies (e.g., Slingo and Annamalai 2000; Shen and Kimoto 1999). Slingo and Annamalai (2000) explain this unusual impact of the 1997 El Niño event by the fact that the atmospheric circulation over the Indian Ocean can enter a different regime depending on the strength of ENSO. The discrepancy between the datasets may be due to the use of satellite information in the GPCP dataset. Overall ERA-40 seems to be more consistent with the station data than GPCP: the linear correlation with the interannual variability of Indian precipitation anomalies from station data is 0.82 for ERA-40 instead of 0.77 for GPCP. The root-mean-square error (RMSE) is 0.51 for ERA-40 instead of 0.57 for GPCP.

4. June precipitation over India

The interannual variability of precipitation over India has been computed for each experiment and averaged over the 15 ensemble members. After a few days of integrations, the model starts to drift toward its own climate. Therefore the model integrations need to be calibrated. In this study, the forecasts are calibrated using a cross-validation technique: for each experiment and for a given year the model climate is computed by averaging all the ensemble forecasts of this experiment excluding the forecasts associated to this specific year. The model climate is then removed from the ensemble forecast to produce precipitation anomalies. The interannual variability of the ensemble-mean precipitation anomalies over India is then compared to anomalies computed from the Indian station data.

Table 2 shows the linear correlation and RMSE between the interannual variability of precipitation anomalies over India predicted by each experiment and analysis. Results suggest that the current ECMWF monthly forecasts (VEPS) has some skill in predicting the interannual variability of June precipitation over India, with a linear correlation of 0.57 (Fig. 2a). The model seems to have more skill in the 1990s than in the more recent years. In particular, in 2007 the model failed to predict above-average precipitation in June over India. Using persisted SST anomalies instead of coupling the atmospheric model to an ocean model after day 10 (experiment PERS) produces similar anomaly correlations, but the RMSE is significantly larger. This is due to the fact that PERS produces unrealistically large precipitation anomalies. MOFC displays significantly less skill than VEPS (Fig. 2b), which is likely due to the lower horizontal resolution. According to a 10 000-iteration bootstrap resampling procedure (Efron 1981) the difference in skill between VEPS and MOFC is significant within the 1% level of confidence. However, the same low-resolution atmospheric model as in MOFC coupled to an ocean mixed layer model with a 1-m vertical resolution in the upper layer (experiment ML) displays significantly better skill than the atmospheric model coupled to an ocean GCM (MOFC; Fig. 2c). The difference is also significant within the 1% level of confidence. The ML also displays better skill than VEPS, despite its lower horizontal resolution. In particular, the period 2000–07 seems better predicted by ML than VEPS. In summary, both the atmospheric resolution and vertical resolution of the upper levels of the ocean model have a positive impact on the skill of the monthly forecasting system to predict June precipitation over India.

The scores obtained with VEPS are also significantly higher than those obtained with the ECMWF seasonal forecasting System 3 starting on 1 May and available on 15 May (therefore available at the same time as VEPS). The linear correlation between the station data and System 3 starting on 1 May is only 0.29. The RMSE between the station data and System 3 starting on 1 May is 1.13, which is significantly higher than the RMSE obtained with VEPS (0.92). This is not that surprising since VEPS starts 15 days later than System 3 forecasts starting on 1 May. However, VEPS also outperforms System 3 starting on 1 June (15 days later than VEPS). The difference is smaller (correlation of 0.57 for VEPS instead of 0.5 for System 3), but statistically significant. According to Table 2, the difference is most likely due to the higher horizontal resolution of VEPS since System 3 has the same resolution as MOFC-T159 (about 120 km). This result suggests that high-resolution extended monthly forecasts could be useful to give a better outlook of the next few months than low-resolution long-range seasonal forecasts.

The onset process of the Indian monsoon is associated with “northward propagation” of precipitation from the near-equatorial Indian Ocean to the Indian subcontinent (Yasunari 1980). To check if the numerical model is able to simulate this important aspect of the Indian monsoon, the total precipitation predicted by the model has been averaged along the longitudinal band 70°–85°E, and its time evolution as a function of latitude has been compared to ERA-40 for 1992 and 2001 (Fig. 3). ERA-40 was chosen because the Indian station dataset covers only land points, and the years 1992 and 2001 were chosen because they represented the driest and wettest Junes, respectively, from 1991 to 2002 (ERA-40 does not cover years beyond 2002). Figure 4 suggests that VEPS produces northward propagations of precipitation as in the analysis. In addition, VEPS captures well the difference between the years 1992 and 2001. In particular, the late onset of 1992 is well predicted, but the 2001 onset is predicted too early. The northward propagation in one individual ensemble member (middle panel of Fig. 3) is as chaotic and, at times, as abrupt as in the analysis.

5. Prediction of Indian precipitation averaged over 5 days (pentads)

In the previous section, we showed that the ECMWF monthly forecasting system displayed some skill in predicting the Indian precipitation averaged over the month of June. In this section, we investigate if the model has also some skill to predict higher-frequency variability of Indian precipitation in June by averaging Indian precipitations over pentads. This should indicate if the model has some skill to predict the onset of the Indian monsoon.

In this study, we consider the 6 following pentads in June: 1–5, 6–10, 11–15, 16–20, 21–25, and 26–30 June. As for section 3, the ensemble-mean forecasts are compared to the analysis from daily station data. The experiment VEPS displays some strong skill to predict the interannual variability of Indian precipitation averaged over the period 1–5 June (Fig. 4a) and 6–10 June (Fig. 4b), which correspond to a time range of days 17–21 and 22–26, respectively. The model also displays some moderate skill for the two following pentads (11–15 and 15–20 June; Fig. 5), with an RMSE still lower than the RMSE obtained with climatology. Interestingly, VEPS seems to have some skill to predict the changes in precipitation anomalies from the 1–5 and 6–10 June pentads. The linear correlation of the analyzed interannual variability of precipitation between those two pentads (dotted lines in Figs. 4a,b) is only 0.5. Therefore, the fact that the model has strong skill to predict the Indian precipitation for the pentad 6–10 June cannot be attributed to persistence only. In particular, the model successfully predicts that the Indian precipitation anomalies would increase from 1–5 to 6–10 June in 1991, and decrease in 2001. Overall those results suggest that VEPS displays some useful skill to predict the onset of the Indian monsoon up to about 30 days.

A comparison with MOFC (Fig. 5) suggests that the increased horizontal resolution of VEPS has a significant positive impact on the forecasts of the precipitation pentads of 6–10, 11–15, and 16–20 June. A comparison between ML and MOFC (Fig. 5) also confirms the conclusion of section 3 that an increase in the vertical resolution of the upper ocean has a positive impact on the monsoon prediction. Interestingly, the decrease of skill with time is slower in ML than in the other experiments. Indeed, ML is the experiment displaying the highest skill for the pentads 16–20 and 21–25 June. Finally a comparison between PERS and MOFC suggests that the ocean–atmospheric coupling has also a significant impact on the scores. In particular the RMSEs are much larger with PERS than with VEPS (not shown) as was the case in section 3 with the mean precipitation over the full month of June.

6. Discussion

The better scores obtained with the mixed layer model, compared to the coupled GCM, in sections 4 and 5 may be explained by a better prediction of SST anomalies over the Bay of Bengal and Arabian Sea (Fig. 6), although the coupled GCM displays better SST forecasts at the equator. The lower scores near the equator are not surprising since, unlike the oceanic component of MOFC, the mixed layer model, which consists just of a 1D thermodynamic model, does not have the dynamics to predict the propagation of equatorially trapped waves. Annamalai et al. (2005) showed that SST anomalies in the southwest Indian Ocean (SWIO) have a significant impact on the onset of Indian monsoon, and most especially on its timing. However, the better scores obtained with ML compared to MOFC cannot be attributed to better SST forecasts in SWIO, since the skill to predict SST anomalies there is lower in ML than in MOFC according to Fig. 6.

The ML seems to produce a more realistic interannual variability of SSTs in the Gulf of Guinea (Fig. 6), which could have an impact on the skill of the model to predict Indian rainfall (Kucharski et al. 2008). Kucharski et al. (2008) found that the SST anomalies in the Gulf of Guinea produce heating anomalies that excite a Rossby wave response in the Indian Ocean, impacting the time-mean monsoon circulation.

The better scores obtained with ML compared to MOFC could also be explained by the faster northward propagation of precipitation in ML than in MOFC (Fig. 7). According to Fig. 7, ML displays a more abrupt propagation particularly between 15° and 20°N than MOFC and VEPS. The propagation in MOFC and VEPS seems slower than in the analysis (Fig. 7a). Figure 8 highlights the difference of northward propagation of precipitation between ML and MOFC in individual years (1994 and 1996). Climatologically, during the first pentad in June, the South Kerala region receives a sudden burst of monsoon rainfall associated with the onset of the monsoon. Figure 9 suggests that the ML experiment is able to capture this maximum of climatological precipitation in the first pentad of June (days 16–20), although it may not be as sudden as in observations. On the other hand, MOFC has a maximum about a week later. This result confirms that the northward propagation is more realistic in ML than in MOFC. This is likely to explain why the scores in ML are significantly higher than in MOFC, particularly for the prediction of the pentads, where predicting the timing of the northward propagation of precipitation is crucial.

The difference in the northward propagation between ML and MOFC (Figs. 7, 8, and 9) could be explained by the enhanced sensitivity of the mixed layer model to the surface fluxes compared to the full dynamical ocean model. Figure 10 shows the time evolution of the tendencies of sea surface temperature in the Arabian Seas as a function of latitude. Both MOFC and ML display similar patterns with a warming preceeding the northward propagation of precipitation, followed by a cooling of SSTs. However, the amplitude of the SST anomalies is significantly different in ML and MOFC with ML displaying much stronger SST anomalies than MOFC. This is consistent with Bernie et al. (2005) who found that a vertical resolution of about 1 m in the upper ocean is necessary to capture 95% of the intraseasonal variability. With a 10-m vertical resolution in the upper ocean, MOFC cannot capture the full amplitude of the intraseasonal variability. This was mentioned in Woolnough et al. (2007) who used experiment settings identical to MOFC and ML, but for the version of IFS. They found that the SST intraseasonal variability in the mixed layer experiment was more consistent with observations from WHOI mooring during TOGA COARE than in the full dynamical ocean model experiment, which displayed a too small intraseasonal variability of SSTs. They also found that the improved response of the mixed layer model was specifically due to the fine vertical resolution.

Several studies have already shown that air–sea interaction processes have an impact on the northward propagation of intraseasonal oscillations (NPISOs; Fu et al. 2003; Fu and Wang 2004a,b; Rajendran et al. 2004; Rajendran and Kitoh 2006). Kawamura et al. (2002) found that this was also the case for the Australian monsoon. MOFC, with its too damped intraseasonal variability of SSTs is likely to strongly underestimate the feedback of the ocean on the northward propagation of precipitation. This may explain the more realistic monsoon precipitation forecasts in ML than in MOFC.

7. Conclusions

The main goal of the present paper is to document the skill of a state-of-the-art monthly forecasting system to predict June rainfall over India. The forecast of June precipitation over India is usually considered as being a difficult problem for seasonal forecasting systems. This study shows that the ECMWF monthly forecasting system starting on 15 May and extended to 46 days to cover the full month of June has some skill to simulate the observed interannual variability of June-mean Indian precipitation during the period 1991–2007. The model has also some skill to predict Indian precipitation averaged over pentads up to a lead time of about 30 days. Since the variability of precipitation from one pentad to another in June is often associated to the onset of the monsoon, this could indicate that the model has some skill to predict the onset of the monsoon a few weeks in advance. Interestingly, the skill of this system exceeds the skill of the ECMWF seasonal forecasting System 3 starting on 1 June, and available on 15 June (therefore available 1 month after the monthly forecasts starting on 15 May). Sensitivity studies indicate that the higher horizontal resolution of the monthly forecasting system is likely to explain this increase of skill. Therefore, high-resolution extended-range forecasts could be useful for the prediction a few weeks in advance of subseasonal events, like the onset of the monsoon. The seamless approach, currently used at ECMWF to produce monthly forecasts (the ECMWF VarEPS-monthly system has a 50-km resolution in the first 10 days and 80 km afterward; see Vitart et al. 2008 for more details) could also be useful for seasonal forecasting as long as there are not enough resources to run 7-month seasonal forecasts at the same resolution as the monthly forecasting system.

Further sensitivity experiments have been performed to assess the impact of high oceanic vertical resolution on the prediction of Indian rainfall. Results show that the prediction of the Indian rainfall is sensitive to the ocean coupling, as it was the case for the MJO (Woolnough et al. 2007). In conclusion, the sensitivity experiments indicate that both the horizontal resolution of the atmospheric model and the representation of the upper ocean have a significant impact on the skill of the monthly forecasting system to predict Indian rainfall.

For technical reasons related to the parallelism of the coupled system (in the ML experiment, the mixed layer model, and the atmospheric model are coupled using the OASIS coupler (Terray et al. 1995)), this model could not be run coupled to the high-resolution atmospheric model. Work is in progress to implement the mixed layer model in the ECMWF atmospheric model Integrated Forecast System (IFS). This will allow us to run experiments with a high-resolution atmospheric model coupled to a high-resolution mixed layer model and verify if the improvements demonstrated in VEPS and ML experiments are additive.

Acknowledgments

We are grateful to Anabel Bowen and Rob Hine, who has helped to improve the quality of the figures. We thank H. Annamalai and an anonymous reviewer whose comments led to substantial improvements in the manuscript.

REFERENCES

  • Adler, S. G., and Coauthors, 2003: The version-2 Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (1979–present). J. Hydrometeor., 4 , 11471167.

    • Search Google Scholar
    • Export Citation
  • Anderson, D., and Coauthors, 2007: Seasonal forecasting system 3. ECMWF Tech. Memo. 503, 58 pp.

  • Annamalai, H., P. Liu, and S-P. Xie, 2005: Southwest Indian Ocean SST variability: Its local effect and remote influence on Asian monsoons. J. Climate, 18 , 41504167.

    • Search Google Scholar
    • Export Citation
  • Balmaseda, M. A., A. Vidard, and D. L. T. Anderson, 2008: The ECMWF Ocean Analysis System: ORA-S3. Mon. Wea. Rev., 136 , 30183034.

  • Bernie, D. J., S. J. Woolnough, J. M. Slingo, and E. Guilyardi, 2005: Modeling diurnal and intraseasonal variability. J. Climate, 18 , 11901202.

    • Search Google Scholar
    • Export Citation
  • Buizza, R., and T. N. Palmer, 1995: The singular-vector structure of the atmospheric general circulation. J. Atmos. Sci., 52 , 14341456.

    • Search Google Scholar
    • Export Citation
  • Buizza, R., M. Miller, and T. N. Palmer, 1999: Stochastic representation of model uncertainties in the ECMWF Ensemble Prediction System. Quart. J. Roy. Meteor. Soc., 125 , 28872908.

    • Search Google Scholar
    • Export Citation
  • Efron, N., 1981: Nonparametrics estimates of standard error: The jackknife, the bootstrap, and other methods. Biometrika, 68 , 589599.

    • Search Google Scholar
    • Export Citation
  • Fu, X., and B. Wang, 2004a: Differences of boreal summer intraseasonal oscillations simulated in an atmosphere–ocean coupled model and an atmosphere-only model. J. Climate, 17 , 12631271.

    • Search Google Scholar
    • Export Citation
  • Fu, X., and B. Wang, 2004b: The boreal summer intraseasonal oscillations simulated in a hybrid coupled atmosphere–ocean model. Mon. Wea. Rev., 132 , 26282649.

    • Search Google Scholar
    • Export Citation
  • Fu, X., B. Wang, T. Li, and J. P. McCreary, 2003: Coupling between northward propagation intraseasonal oscillations and sea surface temperature in the Indian Ocean. J. Atmos. Sci., 60 , 17331753.

    • Search Google Scholar
    • Export Citation
  • Gadgil, S., M. Rajeevan, and R. Nanjundiah, 2005: Monsoon prediction—Why yet another failure? Curr. Sci., 88 , 13891400.

  • Kang, I-S., and J. Shukla, 2006: Dynamical seasonal prediction and predictability of the Asian monsoon. The Asian Monsoon, B. Wang, Ed., Springer-Praxis, 585–612.

    • Search Google Scholar
    • Export Citation
  • Kawamura, R., Y. Fukuta, H. Ueda, T. Matsuura, and S. Iizuka, 2002: A mechanism of the onset of the Australian summer monsoon. J. Geophys. Res., 107 , 4204. doi:10.1029/2001JD001070.

    • Search Google Scholar
    • Export Citation
  • Kucharski, F., A. Bracco, J. H. Yoo, and F. Molteni, 2008: Atlantic forced component of the Indian monsoon interannual variability. Geophys. Res. Lett., 35 , L04706. doi:10.1029/2007GL033037.

    • Search Google Scholar
    • Export Citation
  • Large, W. G., J. C. Mc Williams, and S. C. Doney, 1994: Oceanic vertical mixing: A review and a model with a nonlocal boundary layer parameterization. Rev. Geophys., 32 , 363403.

    • Search Google Scholar
    • Export Citation
  • Molteni, F., F. Vitart, T. Stockdale, L. Ferranti, and M. Balmaseda, 2008: Predictions of tropical rainfall with the ECMWF seasonal and monthly forecast systems. Proc. Workshop on Ensemble Prediction, Reading, United Kingdom, ECMWF. [Available online at http://www.ecmwf.int/publications/library/ecpublications/_pdf/workshop/2007/Ensemble_prediction/Molteni.pdf.].

  • Puri, K., J. Barkmeijer, and T. N. Palmer, 2001: Tropical singular vectors computed with linearised diabatic physics. Quart. J. Roy. Meteor. Soc., 127 , 709737.

    • Search Google Scholar
    • Export Citation
  • Rajeevan, M., J. Bhate, J. D. Kale, and B. Lal, 2006: High resolution daily gridded rainfall data for the Indian region: Analysis of break and active monsoon spells. Curr. Sci., 91 (3) 296306.

    • Search Google Scholar
    • Export Citation
  • Rajendran, K., and A. Kitoh, 2006: Modulation of tropical intraseasonal oscillations by ocean–atmosphere coupling. J. Climate, 19 , 366391.

    • Search Google Scholar
    • Export Citation
  • Rajendran, K., A. Kitoh, and O. Arakawa, 2004: Monsoon low-frequency intraseasonal oscillation and ocean-atmosphere coupling over the Indian Ocean. Geophys. Res. Lett., 31 , L02210. doi:10.1029/2003GL019031.

    • Search Google Scholar
    • Export Citation
  • Reynolds, R. W., N. A. Rayner, T. M. Smith, D. C. Stokes, and W. Wang, 2002: An improved in situ and satellite SST analysis for climate. J. Climate, 15 , 16091625.

    • Search Google Scholar
    • Export Citation
  • Shen, X., and M. Kimoto, 1999: Influence of El Niño on the 1997 Indian Summer Monsoon. J. Meteor. Soc. Japan, 77 , 10231037.

  • Shinoda, T., 2005: Impact of the diurnal cycle of solar radiation on intraseasonal variability in the western equatorial Pacific. J. Climate, 18 , 26282636.

    • Search Google Scholar
    • Export Citation
  • Shinoda, T., and H. H. Hendon, 1998: Mixed layer modeling of the intraseasonal variability in the tropical western Pacific and Indian Oceans. J. Climate, 11 , 22682285.

    • Search Google Scholar
    • Export Citation
  • Slingo, J. M., and H. Annamalai, 2000: The El Niño of the century and the response of the Indian summer monsoon. Mon. Wea. Rev., 128 , 17781797.

    • Search Google Scholar
    • Export Citation
  • Someshwar Das, A., K. Mitra, G. R. Iyengar, and J. Singh, 2002: Skill of medium-range forecasts over the Indian monsoon region using different parameterizations of deep convection. Wea. Forecasting, 17 , 11941210.

    • Search Google Scholar
    • Export Citation
  • Sperber, K. R., and Coauthors, 2001: Dynamical seasonal predictability of the Asian summer monsoon. Mon. Wea. Rev., 129 , 22262247.

  • Terray, L., E. Sevault, E. Guilyardi, and O. Thual, 1995: The OASIS coupler user guide version 2.0. CERFACS Tech. Rep. TR-CMGC 95-46, 123 pp.

    • Search Google Scholar
    • Export Citation
  • Uppala, S., and Coauthors, 2005: The ERA-40 reanalysis. Quart. J. Roy. Meteor. Soc., 131 , 29613012.

  • Vialard, J., F. Vitart, M. A. Balmaseda, T. N. Stockdale, and D. L. T. Anderson, 2005: An ensemble generation method for seasonal forecasting with an ocean–atmosphere coupled model. Mon. Wea. Rev., 133 , 441453.

    • Search Google Scholar
    • Export Citation
  • Vitart, F., 2004: Monthly forecasting at ECMWF. Mon. Wea. Rev., 132 , 27612779.

  • Vitart, F., and Coauthors, 2008: The new VAREPS-monthly forecasting system: A first step towards seamless prediction. Quart. J. Roy. Meteor. Soc., 134 , 17891799.

    • Search Google Scholar
    • Export Citation
  • Webster, P. J., and C. Hoyos, 2004: Forecasting monsoon rainfall and river discharge variability on 20–25-day time scales. Bull. Amer. Meteor. Soc., 85 , 17451765.

    • Search Google Scholar
    • Export Citation
  • Wolff, J. O., E. Maier-Raimer, and S. Legutke, 1997: The Hamburg ocean primitive equation model. Deutsches Klimarechenzentrum Tech. Rep. 13, Hamburg, Germany, 98 pp.

    • Search Google Scholar
    • Export Citation
  • Woolnough, S. J., F. Vitart, and M. A. Balmaseda, 2007: The role of the ocean in the Madden-Julian Oscillation: Implications for MJO prediction. Quart. J. Roy. Meteor. Soc., 133 , 117128.

    • Search Google Scholar
    • Export Citation
  • Xavier, P. K., and B. N. Goswami, 2007: A promising alternative to prediction of seasonal mean all India rainfall. Curr. Sci., 93 , 195202.

    • Search Google Scholar
    • Export Citation
  • Yasunari, T., 1980: A quasi-stationary appearance of 30 to 40 day period in the cloudiness fluctuations during the summer monsoon over India. J. Meteor. Soc. Japan, 58 , 225229.

    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

Interannual variability of June SSTs from GPCP (dotted line), ERA-40 operational analysis (dashed line), and Indian station data (solid line).

Citation: Monthly Weather Review 137, 4; 10.1175/2008MWR2761.1

Fig. 2.
Fig. 2.

Interannual variability of June precipitation anomalies from the ensemble mean of (a) VEPS, (b) MOFC, and (c) ML starting on 15 May (solid line) and analysis from station data (dotted line). The vertical lines indicate 2 std dev.

Citation: Monthly Weather Review 137, 4; 10.1175/2008MWR2761.1

Fig. 3.
Fig. 3.

Hovmoeller diagram of total precipitation averaged between 70° and 85°E. The y axis represents the latitude between 5° and 30°N. The x axis represents the time range from 0 to 46 days. (left) The analysis from ERA-40 for the years (top) 1992 and (bottom) 2001. (middle) The control forecast of VEPS. (right) The ensemble-mean VEPS. The contour interval is 2 mm day−1, with the first contour at 6 mm day−1.

Citation: Monthly Weather Review 137, 4; 10.1175/2008MWR2761.1

Fig. 4.
Fig. 4.

Interannual variability of Indian precipitation anomalies from the ensemble mean of VEPS starting on 15 May (solid line) and analysis from station data (dotted line) for the period (a) 1–5 and (b) 6–10 Jun. The vertical lines indicate 2 std dev.

Citation: Monthly Weather Review 137, 4; 10.1175/2008MWR2761.1

Fig. 5.
Fig. 5.

Linear correlation between the interannual variability between 1991 and 2007 of the ensemble mean of VEPS, MOFC, ML, and PERS with analysis from station data.

Citation: Monthly Weather Review 137, 4; 10.1175/2008MWR2761.1

Fig. 6.
Fig. 6.

The linear correlations between the interannual variability of mean June SST predicted by the model and OI V2 SSTs (Reynolds et al. 2002), which are optimally interpolated observed SSTs, have been computed for the period 1991–2007. The figure shows (a) the scores obtained with ML, (b) the scores obtained with MOFC, and (c) the difference of the scores obtained with ML with the scores obtained with MOFC. In (c), contours indicate higher scores with ML than MOFC.

Citation: Monthly Weather Review 137, 4; 10.1175/2008MWR2761.1

Fig. 7.
Fig. 7.

Hovmoeller diagram of daily total precipitation averaged between 70° and 85°E and over the period 1991–2001 for (a) analysis from ERA-40, (b) ensemble mean of MOFC, (c) ensemble mean of ML, and (d) ensemble mean of VEPS. The y axis represents the latitude between 5° and 30°N. The x axis represents the time range from 0 to 46 days, starting on 15 May. The contour interval is 2 mm day−1, with the first contour at 2 mm day−1.

Citation: Monthly Weather Review 137, 4; 10.1175/2008MWR2761.1

Fig. 8.
Fig. 8.

Hovmoeller diagram of daily total precipitation anomalies (anomalies relative to the period 1991–2001) averaged between 70° and 85°E in (a) 1994 and (b) 1996. The y axis represents the latitude between 5° and 30°N. The x axis represents the time range from 0 to 46 days, starting on 15 May. (left) The analysis from ERA-40, (middle) the ensemble-mean forecasts from MOFC, and (right) the ensemble-mean forecasts from ML. The contour interval is 4 mm day−1, with the first contour at 1 mm day−1.

Citation: Monthly Weather Review 137, 4; 10.1175/2008MWR2761.1

Fig. 9.
Fig. 9.

Total precipitation averaged over the Kerala region: 5°–10°N, 73°–77°E, and over the period 1991–2001. The x axis represents the time range from 0 to 46 days, starting on 15 May. The y axis represents the total precipitation (mm day−1). The gray solid line represents the analysis from ERA-40, the solid black line represents the ensemble mean of ML, and the dotted line represents the ensemble mean of MOFC.

Citation: Monthly Weather Review 137, 4; 10.1175/2008MWR2761.1

Fig. 10.
Fig. 10.

Hovmoeller diagram of SST anomaly tendencies in 1994 (anomalies relative to the period 1991–2001) averaged between 60° and 70°E for (a) the ensemble mean of MOFC and (b) the ensemble mean of ML. The tendencies have been computed by taking the difference between the SSTs at day n and day n − 1. The x axis represents the time range from 0 to 46 days, starting on 15 May. The y axis represents the latitude between 5°S and 30°N. The contour interval is 0.02° day−1, with the first contour at 0.02° day−1. Positive values are contoured.

Citation: Monthly Weather Review 137, 4; 10.1175/2008MWR2761.1

Table 1.

The experiment settings.

Table 1.
Table 2.

Linear correlation and RMSE between the interannual variability of June precipitation anomalies over India simulated by the model and analysis over the period 1991–2007. The numbers in parentheses represent the 99% interval of confidence calculated using a 10 000-iteration bootstrap resampling procedure.

Table 2.
Save
  • Adler, S. G., and Coauthors, 2003: The version-2 Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (1979–present). J. Hydrometeor., 4 , 11471167.

    • Search Google Scholar
    • Export Citation
  • Anderson, D., and Coauthors, 2007: Seasonal forecasting system 3. ECMWF Tech. Memo. 503, 58 pp.

  • Annamalai, H., P. Liu, and S-P. Xie, 2005: Southwest Indian Ocean SST variability: Its local effect and remote influence on Asian monsoons. J. Climate, 18 , 41504167.

    • Search Google Scholar
    • Export Citation
  • Balmaseda, M. A., A. Vidard, and D. L. T. Anderson, 2008: The ECMWF Ocean Analysis System: ORA-S3. Mon. Wea. Rev., 136 , 30183034.

  • Bernie, D. J., S. J. Woolnough, J. M. Slingo, and E. Guilyardi, 2005: Modeling diurnal and intraseasonal variability. J. Climate, 18 , 11901202.

    • Search Google Scholar
    • Export Citation
  • Buizza, R., and T. N. Palmer, 1995: The singular-vector structure of the atmospheric general circulation. J. Atmos. Sci., 52 , 14341456.

    • Search Google Scholar
    • Export Citation
  • Buizza, R., M. Miller, and T. N. Palmer, 1999: Stochastic representation of model uncertainties in the ECMWF Ensemble Prediction System. Quart. J. Roy. Meteor. Soc., 125 , 28872908.

    • Search Google Scholar
    • Export Citation
  • Efron, N., 1981: Nonparametrics estimates of standard error: The jackknife, the bootstrap, and other methods. Biometrika, 68 , 589599.

    • Search Google Scholar
    • Export Citation
  • Fu, X., and B. Wang, 2004a: Differences of boreal summer intraseasonal oscillations simulated in an atmosphere–ocean coupled model and an atmosphere-only model. J. Climate, 17 , 12631271.

    • Search Google Scholar
    • Export Citation
  • Fu, X., and B. Wang, 2004b: The boreal summer intraseasonal oscillations simulated in a hybrid coupled atmosphere–ocean model. Mon. Wea. Rev., 132 , 26282649.

    • Search Google Scholar
    • Export Citation
  • Fu, X., B. Wang, T. Li, and J. P. McCreary, 2003: Coupling between northward propagation intraseasonal oscillations and sea surface temperature in the Indian Ocean. J. Atmos. Sci., 60 , 17331753.

    • Search Google Scholar
    • Export Citation
  • Gadgil, S., M. Rajeevan, and R. Nanjundiah, 2005: Monsoon prediction—Why yet another failure? Curr. Sci., 88 , 13891400.

  • Kang, I-S., and J. Shukla, 2006: Dynamical seasonal prediction and predictability of the Asian monsoon. The Asian Monsoon, B. Wang, Ed., Springer-Praxis, 585–612.

    • Search Google Scholar
    • Export Citation
  • Kawamura, R., Y. Fukuta, H. Ueda, T. Matsuura, and S. Iizuka, 2002: A mechanism of the onset of the Australian summer monsoon. J. Geophys. Res., 107 , 4204. doi:10.1029/2001JD001070.

    • Search Google Scholar
    • Export Citation
  • Kucharski, F., A. Bracco, J. H. Yoo, and F. Molteni, 2008: Atlantic forced component of the Indian monsoon interannual variability. Geophys. Res. Lett., 35 , L04706. doi:10.1029/2007GL033037.

    • Search Google Scholar
    • Export Citation
  • Large, W. G., J. C. Mc Williams, and S. C. Doney, 1994: Oceanic vertical mixing: A review and a model with a nonlocal boundary layer parameterization. Rev. Geophys., 32 , 363403.

    • Search Google Scholar
    • Export Citation
  • Molteni, F., F. Vitart, T. Stockdale, L. Ferranti, and M. Balmaseda, 2008: Predictions of tropical rainfall with the ECMWF seasonal and monthly forecast systems. Proc. Workshop on Ensemble Prediction, Reading, United Kingdom, ECMWF. [Available online at http://www.ecmwf.int/publications/library/ecpublications/_pdf/workshop/2007/Ensemble_prediction/Molteni.pdf.].

  • Puri, K., J. Barkmeijer, and T. N. Palmer, 2001: Tropical singular vectors computed with linearised diabatic physics. Quart. J. Roy. Meteor. Soc., 127 , 709737.

    • Search Google Scholar
    • Export Citation
  • Rajeevan, M., J. Bhate, J. D. Kale, and B. Lal, 2006: High resolution daily gridded rainfall data for the Indian region: Analysis of break and active monsoon spells. Curr. Sci., 91 (3) 296306.

    • Search Google Scholar
    • Export Citation
  • Rajendran, K., and A. Kitoh, 2006: Modulation of tropical intraseasonal oscillations by ocean–atmosphere coupling. J. Climate, 19 , 366391.

    • Search Google Scholar
    • Export Citation
  • Rajendran, K., A. Kitoh, and O. Arakawa, 2004: Monsoon low-frequency intraseasonal oscillation and ocean-atmosphere coupling over the Indian Ocean. Geophys. Res. Lett., 31 , L02210. doi:10.1029/2003GL019031.

    • Search Google Scholar
    • Export Citation
  • Reynolds, R. W., N. A. Rayner, T. M. Smith, D. C. Stokes, and W. Wang, 2002: An improved in situ and satellite SST analysis for climate. J. Climate, 15 , 16091625.

    • Search Google Scholar
    • Export Citation
  • Shen, X., and M. Kimoto, 1999: Influence of El Niño on the 1997 Indian Summer Monsoon. J. Meteor. Soc. Japan, 77 , 10231037.

  • Shinoda, T., 2005: Impact of the diurnal cycle of solar radiation on intraseasonal variability in the western equatorial Pacific. J. Climate, 18 , 26282636.

    • Search Google Scholar
    • Export Citation
  • Shinoda, T., and H. H. Hendon, 1998: Mixed layer modeling of the intraseasonal variability in the tropical western Pacific and Indian Oceans. J. Climate, 11 , 22682285.

    • Search Google Scholar
    • Export Citation
  • Slingo, J. M., and H. Annamalai, 2000: The El Niño of the century and the response of the Indian summer monsoon. Mon. Wea. Rev., 128 , 17781797.

    • Search Google Scholar
    • Export Citation
  • Someshwar Das, A., K. Mitra, G. R. Iyengar, and J. Singh, 2002: Skill of medium-range forecasts over the Indian monsoon region using different parameterizations of deep convection. Wea. Forecasting, 17 , 11941210.

    • Search Google Scholar
    • Export Citation
  • Sperber, K. R., and Coauthors, 2001: Dynamical seasonal predictability of the Asian summer monsoon. Mon. Wea. Rev., 129 , 22262247.

  • Terray, L., E. Sevault, E. Guilyardi, and O. Thual, 1995: The OASIS coupler user guide version 2.0. CERFACS Tech. Rep. TR-CMGC 95-46, 123 pp.

    • Search Google Scholar
    • Export Citation
  • Uppala, S., and Coauthors, 2005: The ERA-40 reanalysis. Quart. J. Roy. Meteor. Soc., 131 , 29613012.

  • Vialard, J., F. Vitart, M. A. Balmaseda, T. N. Stockdale, and D. L. T. Anderson, 2005: An ensemble generation method for seasonal forecasting with an ocean–atmosphere coupled model. Mon. Wea. Rev., 133 , 441453.

    • Search Google Scholar
    • Export Citation
  • Vitart, F., 2004: Monthly forecasting at ECMWF. Mon. Wea. Rev., 132 , 27612779.

  • Vitart, F., and Coauthors, 2008: The new VAREPS-monthly forecasting system: A first step towards seamless prediction. Quart. J. Roy. Meteor. Soc., 134 , 17891799.

    • Search Google Scholar
    • Export Citation
  • Webster, P. J., and C. Hoyos, 2004: Forecasting monsoon rainfall and river discharge variability on 20–25-day time scales. Bull. Amer. Meteor. Soc., 85 , 17451765.

    • Search Google Scholar
    • Export Citation
  • Wolff, J. O., E. Maier-Raimer, and S. Legutke, 1997: The Hamburg ocean primitive equation model. Deutsches Klimarechenzentrum Tech. Rep. 13, Hamburg, Germany, 98 pp.

    • Search Google Scholar
    • Export Citation
  • Woolnough, S. J., F. Vitart, and M. A. Balmaseda, 2007: The role of the ocean in the Madden-Julian Oscillation: Implications for MJO prediction. Quart. J. Roy. Meteor. Soc., 133 , 117128.

    • Search Google Scholar
    • Export Citation
  • Xavier, P. K., and B. N. Goswami, 2007: A promising alternative to prediction of seasonal mean all India rainfall. Curr. Sci., 93 , 195202.

    • Search Google Scholar
    • Export Citation
  • Yasunari, T., 1980: A quasi-stationary appearance of 30 to 40 day period in the cloudiness fluctuations during the summer monsoon over India. J. Meteor. Soc. Japan, 58 , 225229.

    • Search Google Scholar
    • Export Citation
  • Fig. 1.

    Interannual variability of June SSTs from GPCP (dotted line), ERA-40 operational analysis (dashed line), and Indian station data (solid line).

  • Fig. 2.

    Interannual variability of June precipitation anomalies from the ensemble mean of (a) VEPS, (b) MOFC, and (c) ML starting on 15 May (solid line) and analysis from station data (dotted line). The vertical lines indicate 2 std dev.

  • Fig. 3.

    Hovmoeller diagram of total precipitation averaged between 70° and 85°E. The y axis represents the latitude between 5° and 30°N. The x axis represents the time range from 0 to 46 days. (left) The analysis from ERA-40 for the years (top) 1992 and (bottom) 2001. (middle) The control forecast of VEPS. (right) The ensemble-mean VEPS. The contour interval is 2 mm day−1, with the first contour at 6 mm day−1.

  • Fig. 4.

    Interannual variability of Indian precipitation anomalies from the ensemble mean of VEPS starting on 15 May (solid line) and analysis from station data (dotted line) for the period (a) 1–5 and (b) 6–10 Jun. The vertical lines indicate 2 std dev.

  • Fig. 5.

    Linear correlation between the interannual variability between 1991 and 2007 of the ensemble mean of VEPS, MOFC, ML, and PERS with analysis from station data.

  • Fig. 6.

    The linear correlations between the interannual variability of mean June SST predicted by the model and OI V2 SSTs (Reynolds et al. 2002), which are optimally interpolated observed SSTs, have been computed for the period 1991–2007. The figure shows (a) the scores obtained with ML, (b) the scores obtained with MOFC, and (c) the difference of the scores obtained with ML with the scores obtained with MOFC. In (c), contours indicate higher scores with ML than MOFC.

  • Fig. 7.

    Hovmoeller diagram of daily total precipitation averaged between 70° and 85°E and over the period 1991–2001 for (a) analysis from ERA-40, (b) ensemble mean of MOFC, (c) ensemble mean of ML, and (d) ensemble mean of VEPS. The y axis represents the latitude between 5° and 30°N. The x axis represents the time range from 0 to 46 days, starting on 15 May. The contour interval is 2 mm day−1, with the first contour at 2 mm day−1.

  • Fig. 8.

    Hovmoeller diagram of daily total precipitation anomalies (anomalies relative to the period 1991–2001) averaged between 70° and 85°E in (a) 1994 and (b) 1996. The y axis represents the latitude between 5° and 30°N. The x axis represents the time range from 0 to 46 days, starting on 15 May. (left) The analysis from ERA-40, (middle) the ensemble-mean forecasts from MOFC, and (right) the ensemble-mean forecasts from ML. The contour interval is 4 mm day−1, with the first contour at 1 mm day−1.

  • Fig. 9.

    Total precipitation averaged over the Kerala region: 5°–10°N, 73°–77°E, and over the period 1991–2001. The x axis represents the time range from 0 to 46 days, starting on 15 May. The y axis represents the total precipitation (mm day−1). The gray solid line represents the analysis from ERA-40, the solid black line represents the ensemble mean of ML, and the dotted line represents the ensemble mean of MOFC.

  • Fig. 10.

    Hovmoeller diagram of SST anomaly tendencies in 1994 (anomalies relative to the period 1991–2001) averaged between 60° and 70°E for (a) the ensemble mean of MOFC and (b) the ensemble mean of ML. The tendencies have been computed by taking the difference between the SSTs at day n and day n − 1. The x axis represents the time range from 0 to 46 days, starting on 15 May. The y axis represents the latitude between 5°S and 30°N. The contour interval is 0.02° day−1, with the first contour at 0.02° day−1. Positive values are contoured.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 242 88 1
PDF Downloads 117 29 1