Meteorol Atmos Phys 94, 43–64 (2006) DOI 10.1007/s00703-005-0171-6
1 2
Centre for Atmospheric Sciences, Indian Institute of Technology Delhi, New Delhi, India Department of Earth and Planetary Sciences, NEZU, Bunkyo-ku, Tokyo, Japan
Simulation of Indian summer monsoon: experiments with SST s S. K. Deb1 , H. C. Upadhyaya1 , O. P. Sharma1 , and A. Chakraborty2 With 15 Figures Received May 25, 2005; accepted November 20, 2005 Published online: July 11, 2006 # Springer-Verlag 2006
Summary Hindcasts for the Indian summer monsoons (I S M s) of 2002 and 2003 have been produced from an ensemble of numerical simulations performed with a global model by changing S S T. Two sets of ensemble simulations have been produced without vegetation: (i) by prescribing the weekly observed S S T from E CM W F (European Centre for Medium Range Weather Forecasting) analyses, and (ii) by adding weekly S S T anomalies (S S TA ) of April to the climatological S S T during the simulation period from May to August. For each ensemble, 10 simulations have been realized with different initial conditions that are prepared from E CM W F data with five each from April and May analyses of both the years. The predicted June–July monsoon rainfall over the Indian region shows good agreement with the G P CP (observed) pentad rainfall distribution when 5 member ensemble is taken from May initial conditions. The AllIndia June–July simulated rainfall time series matches favourably with the observed time series in both the years for the five member ensemble from May initial condition but drifts away from observation with April initial conditions. This underscores the role of initial conditions in the seasonal forecasting. But the model has failed to capture the strong intra-seasonal oscillation in July 2002. Heating over equatorial Indian Ocean for June 2002 in a particular experiment using 29th May 12 G M T as initial conditions shows some intra-seasonal oscillation in July 2002 rainfall, as in observation. Further evaluation of the seasonal simulations from this model is done by calculating the empirical orthogonal functions (EOF s) of the GPCP rainfall over India. The first four EO F s explain more than 80% of the total variance of the observed rainfall. The time series of expansion coefficients (principal components), obtained by projecting on the observed E O F s, provide a better frame-
work for inter-comparing model simulations and their evaluation with observed data. The main finding of this study is that the All-India rainfall from various experiments with prescribed S S T is better predicted on seasonal scale as compares to prescribed S S T anomalies. This is indicative of a possible useful seasonal forecasts from a G CM at least for the case when monsoon is going to be good. The model responses do not differ much for 2002 and 2003 since the evolution of SST during these years was very similar, hence July rainfall seems to be largely modulated by the other feedbacks on the overall circulation.
1. Introduction The southwest summer monsoon, occurring every year from June–September, is one of the most well known seasonal phenomena for the Indian subcontinent, which is also a dominant feature of the general circulation of the atmosphere. Although there is some consistency in the seasonal reversal of winds over the Arabian sea that herald the onset of monsoon over India, the oscillations of the monsoon system which are manifested in the rainfall occur both at intra-seasonal and inter-annual scales over and around India during summer (Krishnamurti and Bhalme, 1976; Madden and Julian, 1994). Accurate prediction of these oscillations is intimately linked to a faithful representation of the dynamical and physical processes in numerical models if they are to be employed for extended range
44
S. K. Deb et al
predictions. These predictions range from 10 days to monthly and seasonal scales which can be made either with an atmospheric general circulation model (AG CM ) alone or an AG C M coupled to an ocean model. The medium range forecasting (10-day forecast) is operationally produced at several centres, the monthly=seasonal forecasting is still in the experimental stage. The extended range forecasting (ERF ) of monsoon is done with the help of statistical models by the India Meteorological Department (I M D ) ever since Walker (1923) presented their mathematical foundation in his pioneering work. However, the prospect of seasonal forecasting of the Indian summer monsoon must also be investigated with numerical models because they are fairly complete and sufficiently sophisticated to reproduce the weather and climate features. In this regard, Charney and Shukla (1981) stressed the relative importance of sea surface temperature (S ST ) and land-surface conditions as sources of anomalous forcing for the large scale atmospheric flow. Later Palmer (1994) underscored the role of internal nonlinear dynamics and proposed that inter-annual variability arises from changes in the frequency of preferred flow patterns characterizing intra-seasonal variability. For seasonal forecasting, the S ST s are to be predicted from an ocean model; on the contrary, for medium range predictions, the SS T s are simply held constant during the course of AG C M integrations from an initial state up to the final state. The ERF on a monthly scale in tropics could be attempted with numerical models because of the presence of the 40–50 day MaddenJulian Oscillation (MJ O ) in these latitudes. The resolution that should be used for this purpose could be intermediate or the one that is used for medium-range forecasting. The AG CM s do reproduce the observed variability of the Indian summer monsoon (Sperber and Palmer, 1996; Sharma et al, 1998), but the simulations are sensitive to initial conditions and a non-negligible fraction of the variance of the rainfall may not be dynamically predicted (Palmer et al, 1990; 1992) even though SS T s are prescribed during the model integration. Despite these uncertainties, AG CM s are capable of reproducing the observed rainfall contrast of different monsoon years from prescribed S ST distributions over oceans. Thus, AG C M s constitute an important
tool for seasonal predictions. Here, the seasonal simulations of Indian summer monsoon have been carried out with an AG CM where S ST fields are constructed for the period May–August by blending the observed SS T anomalies of the April month with mean S ST of summer months, i.e., the April S ST anomaly remain persisted during model integration and these simulations are referred to as persisted SS T anomaly experiments in the text. Identical initial conditions have been used for another set of experiments where the observed S ST from E C M W F data has been used. From these two sets of experiments an evaluation of responses of the model could be undertaken, which would guide us to perceive the uncertainty and to do further research for improving the skill of ERF with numerical models. The large differences in the quantum of rainfall received over India during July 2002 and 2003 present an interesting case for hindcasts that may be thoroughly investigated with atmospheric models using S S T fields ‘‘predicted’’ a priori by using S ST anomalies of the latest calendar month which persist during the long-term integration. This technique of S ST anomaly persistence during model integration is commonly followed at various centres to construct a future state of the ocean when seasonal predictions are made using an AG CM alone (Barnston et al, 2003; Goddard and Mason, 2002; Soman and Slingo, 1997). It may be underscored here that for seasonal forecasting a coupled ocean-atmospheric model is necessary in order to simulate the intraseasonal variations that evolve from the changing ocean conditions and are strongly linked to atmospheric dynamical and thermodynamical forcings. The MJ O (40–50 day period) is one such tropical oscillation for which a coupled oceanatmospheric model is needed for modelling some of the features of its variability arising from its speed of propagation over the Indian Ocean (Flatau et al, 2001). In a recent study, Flatau et al (2003) have compared the 2002 and 2003 monsoon onsets over Kerala (India) and emphasized their links to MJO . Observational studies also reveal that though the onsets of these two successive monsoons had striking similarity (though onset dates were different), the rainfall over the northern part of India during 2002 was far below the climatological normal; In the contrary, the 2003 rainfall was normal (Kripalani et al, 2004).
Simulation of Indian summer monsoon: experiments with S S T s
This evidence of strong contrast in the July rainfall of these two successive years indeed presents a formidable challenge for modellers to understand the causes of such extremes in the Indian monsoon system (Sikka, 2003; Gadgil et al, 2003). This study is in a way only preliminary because its results are derived from a limited number of experiments on the response of just one model to changes in S ST. Multimodel super-ensemble approach of Krishnamurti et al (1999), described by Krishnamurti et al (2000) in detail, provides a robust means for seasonal forecasting. This approach has been employed at various well-known centres of operational weather forecasts. Undoubtedly, very intensive numerical simulations are required in this direction from multi-model integrations which begin from different and numerous initial conditions and are driven by various scenarios of boundary forcings. Seasonal hindcasts from various coupled atmosphereocean models have also been reported by Palmer et al (2003). The evaluation of model against observations may be accomplished in several ways but the empirical orthogonal function (E O F ) or the principal component analysis (P C A ) is one particular way to identify and compare latent patterns in observed and simulated data. E OF s provide both spatial and temporal patterns which allow a meteorologist to identify dominant oscillations and non-essential noise in the climate data. However, it does not produce any additional information other than that is contained in the data, but it has a great value in evaluating model simulations and disseminating important information from weather data for any region. Thus further analysis of rainfall variability (both temporal and spatial), has been performed in terms of the empirical orthogonal functions (E O F s) of GP C P pentad precipitation data (Xie et al, 2003; Sharma et al, 2003). Earlier, P C A has been used by Bedi and Bindra (1980) and Rasmusson and Carpenter (1983) to study the variability of monsoon rainfall. Bedi and Bindra used annual rainfall data of 70 evenly distributed stations for a period of 60 years (1911–70) and were able to show the important modes of oscillation using the first four principal components (PCs). These four modes are found sufficient to explain a significant part of the total variance of summer (June and July) monsoon rainfall over India. But Rasmusson and
45
Carpenter used the monsoon season precipitation of 31 Indian subdivisions and mean monthly precipitation of 35 Indian and Sri Lanka stations. They examined the relationship between eastern Pacific S ST and rainfall over India and Sri Lanka. Another study by Watanabe and Shinoda (1996) used the E O F method to investigate the longterm variability of summer monsoon during 1946–88. This study demonstrated that the second mode of sea surface temperature was highly correlated with monsoon rainfall. If a model is able to simulate the observed circulation and its variability with some fidelity, then patterns of principal components of simulated and observed parameters should closely resemble. However, a different approach has been adopted here in which the simulated rainfall from model ensembles has been projected on to the E OF s of the observed rainfall, in order to evaluate temporal evolution of the simulated rainfall with that of the observed rainfall. This procedure has been earlier adopted by Molteni et al (2003). Elimination of these errors from simulations will lead to improvements in the representation of the atmospheric processes in the AG CM s. The contrast monsoons of 2002 and 2003 have given us a valuable opportunity to study the strong intraseasonal variations, observed specially during July, with an AG CM that has been forced with varying SS T. In the next section, a brief description of the model, data and the design of experiments are presented along with an elaboration on SS T and initial data. Section 3 deals with results and discussion. Firstly, winds at 850 hPa and 200 hPa stream function, velocity potential and mean rainfall patterns along with their anomalies have been analyzed and interpreted. Then, empirical orthogonal functions and their corresponding time series are presented for the G P C P rainfall. Next, we present the time series of expansion coefficients of simulated rainfall, which are obtained by projecting the simulated rainfall data on the E O F s of G P C P rainfall. In the last section, the conclusions of this study are described. 2. The model, data and experiments The L M D general circulation model (L M DZ 3.3) is the main tool that has been utilized here to produce the seasonal hindcasts of July month over India for the years 2002 and 2003 with
46
S. K. Deb et al
different initial and boundary conditions. The first version of the LMD-AG CM has been described by Sadourny and Laval (1984). This is a finite difference general circulation model which uses Arakawa’s C-type grid (Arakawa, 1966) for spatial discretization. For the simulations reported here, the grid points are regularly spaced along longitude and latitude; however coordinate stretching is an integral part of the model formulation and could easily be effected in this model for achieving finer resolution in an area of interest for capturing the structures arising from mesoscale activity. The horizontal discretization is achieved by setting 96 points in longitude as against 71 points in the latitude. It uses a hybrid co-ordinate in the vertical with 19 unequally spaced layers. The time integration scheme is a combination of two explicit schemes: the basic time step of 30 minutes is split into one Euler-backward (Matsuno) step followed by four leap-frog steps. Fourier filtering is applied to the longitudinal derivatives in polar latitudes that allows a longer and uniform time stepping during integration. Lateral diffusion is modelled by a mixed bi-Laplacian. The physics of the model is fairly complete. The physical parameterization package includes solar radiation (Fouquart and Bonnel, 1980), long-wave radiation (Morcrette and Fouquart, 1985), large-scale condensation, adjustment for dry convection, cumulus convection (Tiedtke, 1989; Emanuel, 1993), a prognostic equation for cloud water and a gravity wave drag (Lott and Miller, 1997). The transport for water substances is dealt with a Van Leer type scheme (Van Leer, 1977; 1979, Hourdin and Armengaud, 1999). At the surface, eddy fluxes are calculated using a bulk method with drag coefficient varying with vertical gradient properties. An important omission here is the specifica-
tion of vegetation in these simulations. Vegetation will be included later but it is desirable to assess first the capability of model of retaining or reproducing the seasonal signal during long-term integrations. The description of all the parameterizations of physical processes in this model could be found in Le Treut et al (1994). Four sets of ensemble integrations have been produced: two for each year differing only on the specification of S S T s (remains same for each set) as surface boundary condition (Table 1) and different initial conditions (Kumar and Hoerling, 1995; Rowell, 1998; Brankovic and Palmer, 2000). Apart from all these experiments as shown in Table 1, one sensitivity experiment has been done in 2002 using 29th May 2002 as initial condition, providing heat source over equatorial Indian Ocean for 30 days staring from 1st June 2002. Each member of the set contains 10 simulations for which the AGC M has been integrated starting from 10 different initial conditions chosen from April and May analyses. All model experiments, though begin with different initial conditions, end on 30th August. For preparing the initial conditions for the ensemble runs, atmospheric parameters (such as winds, moisture, zonal, meridional and vertical velocity, temperature, geopotential etc.) have been extracted from E C M W F datasets. The S ST data are 1.5 1.5 optimum interpolated monthly fields from E C M W F for 2002 and 2003. Since in the present study April SS T anomaly is used in persisted SS TA experiments, the difference of April S ST from 2003 and 2002 are shown in Fig. 1, clearly showing a dipole with positive and negative anomalies in the western and eastern part of Indian Ocean, respectively. The surface albedo, soil moisture and sea-ice are prescribed from climatology. For the validation of the model results,
Table 1. Design of experiments Experiments
S S T specification as boundary conditions
Initial conditions
SST02
Observed S S T for 2002
ANO02 SST03
April to July mean S S T þApril 2002 S S T anomaly Observed S S T for 2003
ANO03
April to July mean S S T þApril 2003 S S T anomaly
27–30 Apr 02, 01 May 02 28–31 May 02, 01 Jun 02 Same as in SST02 27–30 Apr 03, 01 May 03 28–31 May 03, 01 Jun 03 Same as in SST03
10 experiments for each ensemble
Simulation of Indian summer monsoon: experiments with S S T s
47
Fig. 1. Difference in the April S S T of 2003 and 2002
the 2.5 2.5 GPCP pentad rainfall data for June and July months, interpolated on the model grid, are utilized. 3. Results and discussion The analysis of model output for the months of June and July is presented here. However, circulation features are shown only for July with stream function and velocity potential showing the divergent field of 200 hPa and 850 hPa winds both for analysis and model ensembles. The anomalies and differences of mean fields from analysis and simulations between 2002 and 2003 are also presented. The GPCP mean rainfall over ISM region is given together with simulated precipitation. From daily-simulated precipitation, pentads (5-day averages) of rainfall are calculated from model generated rainfall. Since the model assigns equal number of days to all months of the year, there are thus 6 pentad values for each month. The time series (consisting of 12 pentads) for observed and simulated precipitation are used for a comparative evaluation of the latter. Further analysis and comparison with observed rainfall is carried out by performing a principal component analysis on the simulated and G P CP rainfall time series data. 3.1 850 hPa winds The onset of monsoon occurs with the establishment of the low level equatorial flow, which engenders the Somali jet over the Arabian Sea, and the upper level tropical easterly jet in the tropical atmosphere. The lower level jet brings the much needed moisture over the Indian subcontinent for the monsoon rainfall to occur there. The cross-equatorial flow and the Somali jet may
be noted as the dominant features of the monsoon circulation over the Arabian Sea. The mean July 850 hPa wind fields from all simulations (Figure not shown) compares favorably with the observed mean low-level circulation from NCEP analysis for July 2002 and 2003, insofar as its pattern of Somali jet and cross equatorial flow which is bit weaker in the equatorial Arabian Sea. For these monsoons years, the easterly flow over the south Indian Ocean during July is not very different but Somali jet is relatively stronger (20 m=s) in 2003 than in 2002 (15 m=s). In the first instance, the differences in the strength in the Somali jet during 2002 and 2003 monsoon periods may be thought to be responsible for generating less rainfall over India in July 2002 as compared to that in July 2003 but it appears that its departure from its mean position are crucial. If it drifts equatorwards from its climatological position, the quantum of rainfall could reduce over India even though the jet is strong. The strength of Somali jet is reached 20 m=s (approx) in SST02 and SST03 but 10 m=s (approx) in ANO02 and ANO03 respectively. 850 hPa July mean winds anomaly from N C E P and all four set of ensemble simulations, for both 2002 and 2003 are shown in Fig. 2. The cross equatorial flow in both SST03 and ANO03 is strong as compared to observation. This is evident from their corresponding anomaly with weak easterly along equatorial Indian Ocean. It is also note that a part of Somali jet along 10 N is weak in both SST03 and ANO03, showing easterly wind in their corresponding anomaly but observation shows westerly, where as in SST02 and ANO02 no such phenomenon is noticed. Figure 2 also shows the differences of winds between N C E P analysis, two observed S S T experiments and persisted April S S T anomaly experiments between 2003
48
S. K. Deb et al
Fig. 2. 850 hPa mean July winds anomaly and differences from observations and simulations (a) NCEP02, (b) NCEP03, (c) NCEP03–NCEP02, (d) SST02, (e) SST03, (f) SST03–SST03, (g) ANO02, (h) ANO03, and (i) ANO03–ANO02
and 2002. This clearly shows that in SST03 the Somali jet was bit stronger than the SST02, as in N C E P analysis but it was bit weaker in ANO03 than ANO02. This feature correlated with their differences of rainfall over Western Ghats (Fig. 8f and i). The orientation of the Somali jet in all ensemble experiments is practically east–west. One may noticed that the I T C Z (Intertropical convergence zone) is simulated very well in simulations experiments. Also, all experiments the westward turning of northerly winds over the northwest Africa may be noted but without forming any close circulation; however, a close anticyclonic circulation is noted in the observations over this region. This close circulation prevents the relatively strong northerlies to reach over the Arabian Sea, which may adversely affect the northward movement of the Somali jet. An unrealistic strong westerly flow over equatorial Africa has been simulated in the
ensemble. Despite some unfavorable features in the numerical simulation experiments, SST02 (Fig. 2d) and SST03 (Fig. 2e), the observed S S T may provide a good representation for surface boundary when an atmospheric model alone is used as a tool for seasonal forecasting. Several investigations (Soman and Slingo, 1997; Krishnan et al, 2003) earlier have also come to this conclusion. However, one may use the S S TA from an operational couple ocean-atmosphere model (Ji et al, 1998) or predict them from a canonical correlation analysis in which the recent observations of SS TA in tropical oceans form the predictors (Goddard and Mason, 2002). 3.2 200 hPa stream function and velocity potential The features of the upper-level circulation are analyzed with the help of stream function,
Simulation of Indian summer monsoon: experiments with S S T s
velocity potential and divergent field imposed over velocity potential. The Tibetan High at 200 hPa is the most prominent feature of the upper level circulation during summer monsoon in the northern hemisphere. The mid-Pacific trough, mid-Atlantic trough and the tropical easterly jet over India are the other features that are visible on the climatological map during southwest monsoon. The east–west movement of the Tibetan anticyclone and the strength of the upper-level easterly jet directly affect the performance of the summer monsoon over India in a particular year. It may be noted from the N C E P 200 hPa stream function (Fig. 3a and b) that the anticyclone over Tibet is stronger during July 2003 and that its centre is located east of its climatological position during July 2002. The intensity of the Tibetan High and its location close to its climatological position, appear to be the favorable factors responsible for good rains during July 2003. The position of the upper-level subtropical low over south Indian Ocean also appears to be another very important factor. This upper atmospheric low in 2002 is displaced far too east from its climatological position. On the
49
other hand, it is at it climatological position during July 2002. The intensity of the Tibetan High and its location close to its climatological position, appear to be the favorable factors responsible for good rains during July 2003. The position of the upper-level subtropical low over south Indian Ocean also appears to be another very important factor. This upper atmospheric low in 2002 is displaced far too east from its climatological position. On the other hand, it is at its climatological location in 2003 producing lower pressures there. This implies stronger upper level cross equatorial flow from north to south, suggesting stronger Somali jet and a good monsoon during 2003, which does not seem to be the case during July 2002. A noteworthy observation is that the axis joining the centres of the Tibetan High and the upper level low over Indian Ocean wholly lies over the Bay of Bengal in July 2002 analysis, whereas it has moved over the Arabian Sea during July 2003. This shift in the axis predominantly happened due to large east–west movements of the upper level subtropical low, while the movements of the Tibetan High were just marginal. This fact underscores the
Fig. 3. July 2002 stream function at 200 hPa from observation and simulations (a) NCEP02, (b) NCEP03, (c) SST02, (d) SST03, (e) ANO02, and (f) ANO03. Contour interval 10 106 m2 s1
50
S. K. Deb et al
Fig. 4. July 2002 stream function anomaly at 200 hPa from observation and simulations (a) NCEP02, (b) NCEP03, (c) SST02, and (d) SST03. Contour interval 10 106 m2 s1
importance of the relative movements of the upper level centres of action and their locations in the Asian monsoon region especially over the Indian Ocean. The simulated July 200 hPa stream function and corresponding anomaly fields from ensemble experiments for 2002 and 2003 (Figs. 3 and 4) show a fair agreement with the corresponding observed field though there are some differences in their strengths. The position and pattern of stream functions though resemble well with observation, but the stronger Tibetan High in the simulation of SST02 and ANO02 gives higher stream function anomaly (Fig. 4c) over there in 2002. Figure 5a–c shows the differences of stream function between N C E P, observed S ST
experiments and persisted SS TA for 2003 and 2002, respectively. Strength of Tibetan High in July 2003 is bit weaker as compared to 2002, this impact is clearly reflected in their difference of rainfall (Fig. 8f) over northern part of India. The actual position (strength) of Tibetan High is elongated (weaker) in SST03 (Fig. 3d) as compared to observed, where as in ANO03 (Fig. 3f), the position is very well simulated, though the strength is bit weaker. The weaker Tibetan High in simulations with actual SS T s in 2003 (Fig. 3d) as compared to those with persisted SS TA (Fig. 3f) gives a weaker upper easterly jet over India in this experiment implying less rainfall simulation over India in SST03. The actual
Fig. 5. July 2002 stream function difference (a) NCEP03–NCEP02, (b) SST03–SST02, and (c) ANO03–ANO02. Contour interval 2 106 m2 s1
Simulation of Indian summer monsoon: experiments with S S T s
position of Tibetan High in experiment SST02 and ANO02 are displaced to east from their mean position. The location of Tibetan High in ANO03 is very well simulated as compared to SST03, though the strength is less. The simulated upper level subtropical low centred over the south Indian Ocean from the experiments SST02 and ANO02 agrees fairly well in strength and geographical location with the observations, while in experiment SST03 its position is same as that in the observation. In the experiment ANO03 though it is at its climatological position, but bit elongated eastward. The divergent circulation on the global scale can be best understood from velocity potential fields with divergent field imposed over it. During northern hemisphere summer, the divergent centres are generally located over the warm waters over Indonesia in western Pacific and convergence over the Atlantic Ocean. The July mean 200 hPa velocity potential fields in 2002 and 2003 from N C E P analysis (Fig. 6a and b) have
51
similar large scale structures. The most dominant divergent circulation centre is lying over north of the Equator in the western Pacific and the associated convergence centres are located over the tropical South Atlantic Ocean and south-eastern Pacific. An inspection of the July maps of velocity potential reveal that the convergence over India and south Indian Ocean in 2003 is higher than 2002. This suggests the presence of strong convection over the monsoon region in 2003. The mean July velocity potential fields along with divergent from the ensemble experiments of 2002 and 2003 are shown in Fig. 6c–f. They have a similar large-scale structure like that in the analysis, with principal centres of divergence being simulated by the model but some secondary convergence=divergence centres also appear during the integration. From Fig. 6e and f, one may note that divergence centres are located over the Indian subcontinent in experiments ANO02 and ANO03. On the contrary, low gradient in convergence is simulated by the model in experiments
Fig. 6. July 2002 velocity potential and divergent wind at 200 hPa from observation and simulations (a) NCEP02, (b) NCEP03, (c) SST02, (d) SST03, (e) ANO02, and (f) ANO03. Contour interval 1 106 m2 s1
52
S. K. Deb et al
Fig. 7. July 2002 velocity potential anomaly and divergent wind at 200 hPa from observation and simulations (a) NCEP02, (b) NCEP03, (c) SST02, and (d) SST03. Contour interval 1 106 m2 s1
Fig. 8. July 2002 velocity potential and divergent wind difference (a) NCEP03–NCEP02, (b) SST03–SST02, and (c) ANO03–ANO02. Contour interval 2 106 m2 s1
SST02 (Fig. 6c) and SST03 (Fig. 6d) over this region. The difference of velocity potential and corresponding divergent wind from N C E P, observed S ST experiments and persisted SS TA for 2002 and 2003 are shown in Fig. 8a–c. One may notice that divergent centre is located over India in observed SS T experiments, as in N C E P analysis but convergent is located in persisted S STA differences, showing inter-annual variation between 2002 and 2003. Thus, the fraction of the variance explained by the model seasonal forecast appears to be limited when only the SS T
data are used for forcing the simulations. In other words, prescribing S S T alone may not be just sufficient for successful seasonal forecasts; moreover, it underscores the importance of other feedbacks especially arising from the vegetation cover as the simulations reported here use only bare soil. However, this proposition needs to be examined further by using appropriate vegetation cover in the model. These investigations have been left for the future. One can also notice in Fig. 6c and d, that model simulates a large gradient in divergent field in the equatorial Pacific Ocean in
Simulation of Indian summer monsoon: experiments with S S T s
July 2003 and 2002 but fails to do so in experiments ANO02 (Fig. 6e) and ANO03 (Fig. 6f). The strong overturning associated with the east– west circulation (Krishnamurti, 1971) over monsoon region in experiments SST02 and SST03 resembles the observed overturning, while it is not so strong in experiments ANO02 and ANO03. From the above discussion, it may be concluded that the model has satisfactorily reproduced the average features of large-scale circulation in the simulations. Next, the most sensitive parameter rainfall is analyzed with the help of mean fields and the empirical orthogonal functions analysis. 3.3 Observed and simulated mean rainfall The mean July observed rainfall anomaly from G P CP for the years 2002 and 2003 are shown in Fig. 9a and b, respectively. From an inspection
53
the July 2002 mean rainfall anomaly (Fig. 9a) shows a vertical split in the rainfall pattern about 80 E, poor rainfall region to its west and relatively good rainfall to its east with a maximum in its distribution occurring over the West Bengal and head Bay of Bengal region. On the other hand, rainfall anomaly is homogeneously distributed over most parts of India during July 2003. This is evident from the Fig. 9c showing the difference of mean rainfall for the month of July 2003 and 2002. The mean July summer monsoon rainfall anomaly from SST02 and SST03 are shown in Fig. 9d and e. The mean rainfall anomaly in experiment SST02 (Fig. 9d) shows a fair correspondence over land with observations (Fig. 9a), with poor precipitation over the northwestern part of India and appreciable rainfall over West Bengal, the head Bay region and Tripura. But the model simulates unrealistically
Fig. 9. July rainfall anomaly (mm=day): (a) GPCP 2002 (Observed), (b) GPCP 2003 (Observed), (c) DIFF GPCP (2003–2002), (d) SST02, (e) SST03, (f) SST03–SST02, (g) ANO02, (h) ANO03, and (i) ANO03–ANO02
54
S. K. Deb et al
higher rainfall amounts over the Arabian Sea with another maximum lying west of Western Ghats. In the persisted S STA experiment for ANO02 the model simulated rainfall resembles with observations with less rainfall over Indian subcontinents. The July rainfall from the experiment SST03 (Fig. 9e) has a good correspondence with July 2003 GP CP mean rainfall (Fig. 9b) over most parts of India except Western Ghats, where negative anomaly is present in the simulation. However, there is a strong maximum in the rainfall over equator about 70 E, which is present in GP C P rainfall (Fig. 9b), but bit weaker. Thus, the model produces reasonably well the distribution and rainfall amounts for July 2003, but it does not happen for July 2002. An inspection of the July mean rainfall from experiment SST02 reveals that the model has failed to simulate the observed pattern (Fig. 9c), and one may note some principal deficiencies in the model simulation: a tongue of dry region over northwest India including parts of Gujarat and relatively heavy rainfall over Himalaya and equatorial central Indian Ocean. This tongue is not visible in 2002 experiments but clearly visible in 2003. In both SST03 and ANO03 the rainfall over Western Ghats is not simulated well. This may be due to the poor horizontal resolution viz. 3.8 lon and 2.5 lat (Sabre et al, 2000) and non-inclusion of vegetation in the model. It is also note that in the present simulation climatological soil-moisture, sea-ice and albedo are used, as we all know the importance of evolving soil-moisture and related land-surface parameters for monsoon simulations (Douville et al, 2001; 2002; 2003). Moreover, there is a striking similarity in the morphology of simulated rainfall from experiments ANO02 (Fig. 9g) and ANO03 (Fig. 9h) which is largely attributed to the similarity of April S ST during two monsoon years. On the other hand, simulations forced with observed S ST show resemblance in the rainfall over land regions in 2002 (Fig. 9d) with only minor differences near Western Ghats. It has been found from an inspection of the weekly S S T that there were no significant differences in its evolution during 2002 and 2003 and even the onset of these two monsoons were of similar nature (Flatau et al, 2003). But in reality there were huge differences in the mean July rainfall between 2002 and 2003 seasonal and inter-annual variations. This clearly
shows in the difference of observed rainfall for these two years (Fig. 9c), which is not simulated very well in all the experiments (Fig. 9f and Fig. 9i), though the observed SS T experiments has produced better picture as compared to the SS TA experiments. The mean July 2002 observed and simulated rainfall and their corresponding anomaly (Fig. 12) using 29th May 2002 as initial condition are showing the sensitivity of heat source over equatorial Indian Ocean in June 2002. Heat source over Indian Ocean has improved the simulation quite a lot in July 2002. Figure 12g also shows the difference of July rainfall over Indian subcontinent with and without using heat source. Therefore, it is a very challenging problem for modellers to identify the mechanisms that cause such intense variations in the monsoon system and to understand the extent of their coupling (Krishnamurthy and Shukla, 2000; Sperber et al, 2000; Goswami and Ajaya Mohan, 2001) which may also be responsible for the observed variability both at intra-seasonal and inter-annual scales. In a recent paper, Molteni et al (2003) have examined the impact of S ST anomalies on inter-annual and intra-seasonal variations by performing several predictability experiments for the Asian summer monsoon. Their findings, though of mixed nature, shows a high correlation between observed and simulated SVD -2 index (related to all-India rainfall index) is very encouraging for the prospects of numerical seasonal forecasting. We believe that vegetation in conjunction with observed SS T also plays an important role in modulating the intra-seasonal variations of summer monsoon; however this needs to be quantified from a long series of numerical simulations with the model. A vegetation model due to Ducoudre et al (1993) will be incorporated in the model for future investigations on seasonal forecasting. Further evaluation of simulations is done by calculating the all-India rainfall time series from the June and July rainfall by comparing it with G P C P data ( Figs. 10 and 11) for all members of the ensemble experiments viz. one 10 members and two 5 members ensembles consisting of five initial conditions each from April and May respectively. The 5 members ensembles are denoted by SSTapr02, SSTmay02, ANOapr02 and ANOmay02 for SST02 and ANO02, respectively, and similar index for 2003 also. It is found that
Simulation of Indian summer monsoon: experiments with S S T s
55
Fig. 10. Comparison of simulated and GPCP all-India June–July rainfall time series for observed SST (a) 2002, and (b) 2003
Fig. 11. Comparison ulated and GPCP June–July rainfall time persisted SST anomaly and (b) 2003
of simall-India series for (a) 2002,
56
S. K. Deb et al
Fig. 12. 2002 July mean observed and simulated (29 May 2002 as IC) rainfall and its anomaly (a) G P C P, (b) Normal, (c) Heating, (d) Anomaly G P C P, (e) Anomaly normal, (f) Anomaly heating, (g) Heating normal
the rainfall time series are wide spread in the persisted SS T anomaly experiments as compared to observed S ST experiments for both the year. One may easily notice that the ensemble mean from the May initial condition come more closer to the observation as compared to the April ensemble mean in all experiments for both the years. It is also notice that in 2002 model has not been able to capture the intra-seasonal variability in July. Over a larger part of the given period, except at its beginning and ending por-
tions, all three curves have a very close resemblance in 2003. This is good news for the model, since it has reproduced very accurately the allIndia rainfall time variations of the 2003 summer monsoon. Unfortunately, this has not happened for July 2002 summer monsoon, as it has not been able to simulate the large oscillation in July 2003, which is a bad news for the model. Furthermore, the two sets of simulations only differ in the specification of S S T yet the all-India rainfall is largely spread with persisted SS TA and
Simulation of Indian summer monsoon: experiments with S S T s
57
Fig. 13. Comparison of simulated and GPCP all-India June–July rainfall time series for 2002 from two different simulation (with and without using heat source over equatorial Indian Ocean) (a) Pentad, and (b) Daily
less with actual S ST, it seems plausible to infer that it is mostly caused by deficiencies in the convective parameterization which affect the overall quality of model simulations. This kind of observation has also been made by Molteni et al (2003) for the E C M W F model from predictability experiments for the summer monsoons. But Fig. 13 shows both pentad and daily June–July 2002 all-India rainfall time series for a particular experiment in 2002 where equatorial heat source over Indian Ocean has been applied and compared it with the experiment where no heat source has been added and with observed G P CP. It is found that application of heat source over Indian Ocean has able to capture the strong intra-seasonal oscillation in July 2002, but it has underestimated the June 2002 rainfall quite a bit. Nevertheless, the use of dynamical models for seasonal forecasting does get some vital support from this mixed outcome about the behavior of variations in the simulated all-India summer monsoon rainfall. The all India cumulative rainfall for the month of June and July for 2002 and 2003 from all members of the ensemble and G P CP are shown in Table 2. From the Table
2a, it is evident that so far as cumulative July rainfall are concerned, the total rainfall simulated in ANO02 and ANO03 are close to GP C P but in SST02 and SST03 it is overestimated. For June 2002 (Table 2b) it is found that cumulative rainfall by SST02 is a exactly match with G P C P, but in ANO02 it is underestimated by 13 mm for G P C P. In June 2003 both SST03 and ANO03 have overestimated as compared to G P C P. One major difficulty in evaluating the ensemble evaluation arises due to large differences between the G P C P and I M D rainfall (not given). The G P C P only presents the average large-scale field. So it is appropriate to compare the model results with GP C P. 3.4 The principal components of observed rainfall The empirical orthogonal functions of the G P C P rainfall are calculated over the Indian region which encompasses 56 model grid points. Their time series are also been normalized. The first four E O F s of the pentad rainfall series represent most part of the intra-seasonal variability of ISM
301.57 191.91
275.46 228.97
E-3 (29 apr) 247.40 249.96
E-4 (30 apr)
297.47 182.87
325.35 250.57
215.90 287.97
282.11 192.01
July 2003 All India Cumulative rainfall (mm)
230.74 248.99
E-2 (28 apr)
SST03 ANO03
SST02 ANO02
231.26 193.80
E-2 (28 apr) 222.86 229.78
E-3 (29 apr) 216.82 239.24
E-4 (30 apr)
164.50 112.61
273.93 113.33
199.81 197.94
204.38 133.96
June 2003 All India Cumulative rainfall (mm)
144.73 128.03
E-1 (27 apr)
June 2002 All India Cumulative rainfall (mm)
Table 2b. June 2002 and 2003 All India Cumulative rainfall
SST03 ANO03
SST02 ANO02
E-1 (27 apr)
July 2002 All India Cumulative rainfall (mm)
Table 2a. July 2002 and 2003 All India Cumulative rainfall
210.26 115.66
161.54 139.72
E-5 (01 may)
260.46 232.80
229.28 208.28
E-5 (01 may)
125.667 158.93
109.57 86.07
E-6 (28 may)
213.18 198.26
233.32 167.14
E-6 (28 may)
147.74 153.72
132.075 88.66
E-7 (29 may)
244.59 190.33
217.97 155.38
E-7 (29 may)
122.46 103.76
99.62 75.32
E-8 (30 may)
207.11 179.39
205.1 136.64
E-8 (30 may)
186.46 176.50
134.58 129.71
E-9 (31 may)
220.67 208.209
221.92 183.66
E-9 (31 may)
99.98 138.78
113.25 120.02
E-10 (01 jun)
229.08 187.19
205.52 134.90
E-10 (01 jun)
173.52 140.52
156.63 143.03
ENS-10
249.59 210.96
236.83 190.58
ENS-10
126.3 126.3
156.7 156.7
OBS (GPCP)
214.7 214.7
178 178
OBS (GPCP)
58 S. K. Deb et al
Simulation of Indian summer monsoon: experiments with S S T s
59
Fig. 14. Four leading EOF s of June–July G P C P pentad rainfall for 2002 (a–d), and 2003 (e–h)
60
S. K. Deb et al
that arises from its inherent oscillations or due to the passage of advective=convective systems through this region. The first E O F of GP CP pentad rainfall (Fig. 14a) in 2002 explains a significant part (56%) of the total variance, while the second, third and fourth E O F s (Fig. 14b–d) jointly put together explain 28% (12%, 10% and 6%, respectively) of the total variance of the monsoon rainfall. Thus the first four E O F s explain 84% of the total variance of the rain-
fall. However, the first four E O F s explain 86% (Fig. 14e–h) of the variance of June–July monsoon rainfall. Most importantly, the leading E O F in 2003 is far more dominant as it explains 71% of the variance as compared to 56% by the same in the preceding year. Obviously rainfall was fairly homogenous over the entire country and the effect of model variations on seasonal mean rainfall was much smaller because the variance of rainfall in 2003 explained by second, third and
Fig. 15. Time series principal components of G P C P 2002, SST02 and ANO02 (left panel) and GPCP 2003, SST03 and ANO03 (right panel)
Simulation of Indian summer monsoon: experiments with S S T s
fourth E O F s is only 6%, 5% and 4% respectively; implying the dominance of active intraseasonal variations on the rainfall in this year, while the opposite is true in 2002 monsoon (Kripalani et al, 2004). This evidently gives rise to a strong signal of inter-annual variability in observed monsoon rainfall of two successive years, where intra-seasonal variations have played a key role. The strong role of intra-seasonal variations in determining the seasonal-mean variability of the monsoon has been earlier stressed by Goswami and Ajay Mohan (2001). The E O F 1 of each monsoon has a unimodal pattern with a maximum in correlations (þ0.9). The region of maximum correlations is oriented northeast and lies over West Bengal and Bangladesh during 2002, where as it practically covers entire India in 2003. The homogeneous pattern of E OF 1 is associated with increased or decreased precipitation. The structure of the first E O F 1 for 2002 has a very similar pattern to that is shown in a study by Shukla (1987), there also a zone of higher values of correlations of the first function is noted over the northeast India. He argued that higher values of the E O F 1 are confined to those regions of India, which show large correlations with the Southern Oscillation. This is difficult to verify here with only two monsoon year data and a longer time series is necessary. The corresponding time series (PC1) of E O F 1 of GP CP rainfall (Fig. 15a, e) of two monsoons represent the progress of mean precipitation over India during each year and one may note the contrast in rainfall there. Moreover, there is also a large variation in amplitude of PC1 (Fig. 15a) during July 2002 due to the long break conditions in the month of July; on the contrary amplitude of the PC1 of 2003 (Fig. 15e) gradually increases and peaks during the month of July which depicts the actual behaviour of observed rainfall. The second, third and fourth E OF s of precipitation for 2002 and 2003 also present some interesting results even though they explain much lower proportion of the variance. E O F 2 (Fig. 15b) for 2002 divides ISM region just into two parts, with strong positive correlations (þ0.6) in the western part and negative correlations (0.3) in the around the eastern part of India. This clearly shows an east–west oscillation of the summer rains arising from the intra-seasonal variations of the summertime circulation. The time series PC2
61
(Fig. 15b) shows high frequency variations in the month of July 2002. The E O F 2 (Fig. 14f) for 2003 divides India into three parts with positive correlations in the southern and northern region, and negative correlations lying in between them. This type of pattern in rainfall occurs over India after the onset of a break in monsoon; Fig. 14f shows its amplitude. This is still not a dominant mode since it explains a much smaller (6%) proportion of the variance. The E O F 3 map (Fig. 15c) for 2002 indicates the presence of north–south oscillations in rainfall. The corresponding time series PC3 (Fig. 15c) possesses an oscillation of high amplitude during first week of July. The maps of E O F 4 in 2002 and 2003 show the contribution of weather-scale disturbances (lows and depression etc) to rainfall variability. The time series of E O F 4 (Fig. 15d, h) is apparently composed of rapid oscillations as compared to the time series of other E OF s. In what follows, the rainfall from model ensembles is analyzed in terms of the aforementioned leading components. 3.5 Analysis of simulated rainfall from model ensembles The simulated precipitation from each model ensemble is projected on the E O F s of the G P C P pentad rainfall discussed above. Hence the observed and model simulated rainfall anomalies have common spatial structures defined by the E O F s (of observed rainfall) and consequently, the differences between these two fields shall appear in the corresponding expansion coefficients or the principal component time series of each mode. It then becomes much easier to compare mode wise essential differences over the entire region from expansion coefficients time series of these two fields. The elegant mathematical foundation of a projection method makes it one of the novel methods of evaluating a large body of numerical simulations with greater ease and reliability. This procedure is robust and which could help in identifying the model deficiencies as well. The expansion coefficients series of E O F s of the G P C P and simulated rainfall anomalies will be referred in the text as PCs (observed) and PCs (simulated), respectively. In a recent study, Molteni et al (2003) have used the singular value decomposition (SVD ) modes on which the model anomalies were projected
62
S. K. Deb et al
to analyze 850 hPa winds and precipitation. Bretherton et al (1992) have presented several methods and their intercomparison for finding the coupled patterns in meteorological data which could be used effectively for evaluating the model performance. Moreover, several authors have applied S V D to projections of meteorological fields on suitable subspaces spanned by dominant E O F s for studying the coupled patterns. The pentad averages of precipitation are calculated from daily stored values in the simulations and subsequently the precipitation anomalies are calculated. The two curves (Fig. 15a) of PC1 (simulated) from the experiments SST02 and ANO02 resemble the corresponding PC1 (observed) in 2002, except in July 2002 simulated PC1 in both experiments have not been able to capture the long break condition as like in PC1 (observed). The time series of expansion coefficients obtained by projecting the simulated rainfall anomalies from experiments SST02 and ANO02 on EOF2 (Fig. 15b) is showing an out of phase relationship with PC2 (observed) during July 2002. Interestingly, the curves of PC3 (simulated) from experiments SST02 and ANO02 (Fig. 15c) show greater resemblance with the corresponding curve of PC3 (observed). On the contrary, it may be seen from the curves (Fig. 15d) of PC4 (simulated) that they do not show oscillations of the kind, which are present in the PC4 (observed). In a similar manner, the simulated rainfall anomaly values from model ensembles for 2003 were projected on to the E OF s of the 2003 G P C P rainfall. The curves (Fig. 15e) of expansion coefficients series from experiments SST03 and ANO03 show very good resemble with PC1 (observed) for 2003 monsoon (Fig. 15e). PC2 (observed) and PC2 (simulated) has also very good agreement with each other (Fig. 15f). The curve (Fig. 15g) of PC3 (observed) shows only one oscillation. Notably, the PC3 (simulated) of experiments SST03 and ANO03 agree with observations in June but in July they are out of phase. The PC4 (simulated) resemble each other (Fig. 15h) to some extent with PC4 (observed), which possesses high frequency oscillations (Fig. 15h). From the above discussion, it may be inferred that PC1 (simulated) and PC2 (simulated) resemble very well to PC1 (observed) and PC2 (observed) in 2003 and in 2002 PC2 (simulated) agrees well with PC2 (observed).
4. Conclusions The seasonal hindcasting of Indian summer monsoons of 2002 and 2003 has been studied from ensemble experiments that have been performed using a general circulation model by changing SS T s for different experiments. The model could serve as a tool for seasonal forecasting because certain parameters of interest like precipitation, the low level and upper level features of the large-scale circulation are satisfactorily reproduced in the simulations. The simulated I S M precipitation values in model forced by persisted SS TA , are generally wide spread in comparison to those produced in numerical experiments with prescribed SS T. Moreover, the model simulates a tongue of dry region in northwestern part of India, in contravention to observations, in all simulations and has a tendency to produce higher rainfall over Himalaya. This is indicative of certain deficiencies in the parameterization of convection, which has to be looked upon. Five members ensemble from May initial condition has come very close to the observed as compared to the other five member ensemble from April initial condition in all experiment. This shows that initial condition plays a significant role in the seasonal forecasting simulation. Thus, the probability of a useful seasonal forecast from a model is higher if the summer monsoon is going to be good, possibly because the strong correlations between IS M rainfall and SS T are reproduced in the simulations. However, if intraseasonal oscillations dominate, then model produced rainfall variations may depend on other factors besides S ST, which is evident from SST02 and ANO02 experiment where model fails to produce the long break condition in July 2002. But equatorial heat source over Indian Ocean is successful in producing the long break condition in July 2002 over India, as compared to any member of SST02. This concludes that July 2002 rainfall over India was strongly modulated by the heating over equatorial Indian Ocean. Nevertheless, the observed S S T model ensembles do provide a signal about the overall expected behavior of the monsoon. It has been observed that the time series of expansion coefficients (PCs) obtained by projecting the simulated rainfall on observed E O F s also provide a framework for comparing model simu-
Simulation of Indian summer monsoon: experiments with S S T s
lations and their evaluation with observed data. It has been observed that PC1 (simulated) and PC2 (simulated) resemble reasonably well to PC1 (observed) and PC2 (observed) for 2003 and PC3 (simulated) and PC3 (observed) for 2002. PC2 appears to be very sensitive to the specification of the SS T in the model in 2002. In other words, PC2 may serve as a parameter to design an appropriate measure to evaluate a seasonal forecast. This appears to be a major problem with the model, but requires a separate study based on several monsoon years. A study on the relationship of rainfall of each year with S ST of that year is very much warranted. In conclusion, it may be mentioned that seasonal forecasting from AG C M s has a very bright prospect. Acknowledgement The research has been supported by the Council for Scientific and Industrial Research (CS IR ), New Delhi under the project ‘‘Adoption, development and evaluation of a variable AG CM (V R A M ) for monsoon region’’ under N M ITL I scheme. Laurent Fairhead (L M D ) and Marie-Angile Filiberti (I P S L ) are acknowledged for providing E CM W F initial and boundary conditions data to make this study possible. References Arakawa A (1966) Computational design for long-term numerical integrations of the equations of atmospheric motion. J Comput Phys 1: 119–143 Barnston AG, Mason SJ, Goddard L, DeWitt DG, Ziebiak S (2003) Multimodel ensembling in seasonal forecasting at IRI. Bull Am Meteor Soc 8(4): 1783–1796 Bedi HS, Bindra MMS (1980) Principal components of monsoon rainfall. Tellus 32: 296–298 Brankovic C, Palmer TN (2000) Seasonal skill and predictability of ECMWF PROVOST ensembles. Q J Roy Meteorol Soc 126: 2035–2067 Bretherton CS, Smith C, Wallace JM (1992) An intercomparison of methods for finding coupled patterns in climate data. J Climate 5: 541–560 Charney JG, Shukla J (1981) Predictability of monsoons. In: Monsoon dynamics (Lighthill J, Pearce RP, eds). Cambridge University Press, pp 99–109 Douville H (2002) Influence of soil moisture on the Asian and African monsoons. Part 2: Inter-annual variability. J Climate 15: 701–720 Douville H (2003) Assessing the influence of soil moisture on seasonal climate variability with AGCMs. J Hydro Meteorol 4: 1044–1066 Douville H, Chauvin F, Broqua H (2001) Influence of soil moisture on the Asian and African monsoons. Part 1. Mean monsoon and daily precipitation. J Climate 14: 2381–2403 Ducoudre N, Laval K, Perrier A (1993) SECHIBA, a new set of parameterizations of hydrologic exchanges at the
63
land=atmosphere within LMD atmospheric general circulation model. J Climate 6: 284–273 Flatau MK, Flatau PJ, Daniel Rudnick (2001) The dynamics of double monsoon onsets. J Climate 14: 4130–4146 Flatau MK, Flatau PJ, Schmidt J, Kiladis GN (2003) Delayed onset of the 2003 Indian monsoon. Geophys Res Lett 30(14): 1768 (DOI: 10.1029=2003GL017424) Fouquart Y, Bonnel B (1980) Computations of solar heating of the earth’s atmosphere: a new parameterization. Beitr Phys Atmos 53: 35–62 Gadgil S, Vinayachandran PN, Francis PA (2003) Droughts of the Indian summer monsoon: Role of clouds over the Indian Ocean. Curr Sci 85(12): 1713–1719 Goddard L, Mason SJ (2002) Sensitivity of seasonal climate forecasts to persisted SST anomalies. Climate Dyn 19: 619–631 Goswami BN, Ajai Mohan RS (2001) Intra-seasonal oscillations and inter-annual variability of the Indian summer monsoon. J Climate 14: 1180–1198 Hourdin F, Armengaud A (1999) On the use of finite volume methods for atmospheric advection of trace species: I. Test of various formulations in a general circulation model. Mon Wea Rev 127: 822–837 Ji M, Behringer DW, Leetmma A (1998) An improved coupled model for ENSO prediction and implications for ocean initialization, Part 2: The coupled model. Mon Wea Rev 126: 1022–1034 Kripalani RH, Kulkarni A, Sabade SS, Revadekar JV, Patwardhan SK, Kulkarni JR (2004) Intra-seasonal oscillations during monsoon 2002 and 2003. Curr Sci 87(3): 325–331 Krishnamurti TN (1971) Tropical east–west circulations during northern summer. J Atmos Sci 28: 1342–1347 Krishnamurti TN, Bhalme HN (1976) Oscillations of a monsoon system. Part 1: Observational aspects. J Atmos Sci 33: 1937–1954 Krishnamurti TN, Kishtawal CM, LaRow T, Bachiochi D, Zhang Z, Williford CE, Gadgil S, Surendran S (1999) Improved skills for weather and seasonal climate forecasts from multi-model superensemble. Science (September 3) Krishnamurti TN, Kishtawal CM, LaRow T, Bachiochi D, Zhang Z, Williford CE, Gadgil S, Surendran S (2000) Multi-model superensemble forecasts for weather and seasonal climate. J Climate 13: 4196–4216 Krishnamurthy V, Shukla J (2000) Intra-seasonal and inter-annual variability of rainfall over India. J Climate 13: 4366–4377 Krishnan R, Majumder M, Vaidya V, Ramesh KV, Satyan V (2003) The abnormal Indian summer monsoon of 2000. J Climate 16: 1177–1194 Kumar A, Hoerling MP (1995) Prospects and limitations of seasonal atmospheric GCM predictions. Bull Amer Meteor Soc 76: 335–345 Le Treut H, Li ZX, Forichon M (1994) Sensitivity study of LMD GCM to green house forcing associated with two different cloud water parameterization. J Climate 7: 1827–1841 Lott F, Miller M (1997) A new subgrid scale orographic drag parameterization: its testing in the ECMWF model. Q J Roy Meteorol Soc 123: 101–127
64
S. K. Deb et al: Simulation of Indian summer monsoon: experiments with S S T s
Madden RA, Julian PR (1994) Observations of 40–50-day tropical oscillations – a review. Mon Wea Rev 122: 814–836 Molteni F, Corti S, Ferranti L, Slingo JM (2003) Predictability experiments for the Asian summer monsoon: Impact of SST anomalies on inter-annual and intra-seasonal variability. J Climate 16: 4001–4021 Morcrette JJ, Fouquart Y (1985) On systematic errors in parameterized calculations of long wave radiation transfer. Q J Roy Meteorol Soc 111: 691–708 Palmer TN, Brankovic C, Molteni F, Tibaldi S (1990) Extended range predictions with ECMWF models: Inter-annual variability in operational model integrations. Q J Roy Meteorol Soc 116: 799–834 Palmer TN, Brankovic C, Viterbo P, Miller MJ (1992) Modelling inter-annual variations of summer monsoons. J Climate 5: 399–417 Palmer TN, Alessandri A, Andersen U, Cantelaube P, Davey M, Deque M, Diez E, Doblas-Reyes FJ, Feddersen H, Graham R, Gualdi S, Gueremy J-F, Hagedorn R, Hoshen M, Keenlyside N, Latif M, Lazar A, Maisonnave E, Marletto V, Morse AP, Orfila B, Rogel P, Terres J-M, Thomson MC (2004) Development of a European multimodel ensemble system for seasonal to inter-annual prediction (DEMETER). Bull Amer Meteor Soc 85(6): 853–872 Rasmusson EM, Carpenter TH (1983) The relationship between eastern equatorial Pacific sea-surface temperature and rainfall over India and Sri Lanka. Mon Wea Rev 111: 517–528 Rowell DP (1998) Assessing potential seasonal predictability with an ensemble of multidecadal GCM simulations. J Climate 11: 109–120 Sabre M, Hodges K, Laval K, Polcher J, Desalmand F (2000) Simulation of monsoon disturbances in the LMD GCM. Mon Wea Rev 128: 3752–3771 Sadourny R, Laval K (1984) January and July performances of LMD general circulation model: In new perspective in climate modelling (Berger AL, Nicolis C, eds). Elsevier Press, pp 173–198 Sharma OP, Le Treut H, Seze G, Fairhead L, Sadourny R (1998) Inter-annual variations of summer monsoons: Sensitivity to cloud radiative forcing. J Climate 11: 1883–1905 Sharma OP, Upadhyaya HC, Harzallah A, Sadourny R (2003) Preparing area averaged rainfall on a regular grid
from scattered rain gauge data. J Comp Meth Sci Engg (in press) Shukla J (1987) Inter-annual variability of monsoon. In Monsoons (Fein JS, Stephens PL, eds). New York: Wiley, pp 399–463 Sikka DR (2003) Evaluation of monitoring and forecasting of summer monsoon over India and a review of monsoon drought of 2002. Proc Indian Natl Sci Acad 69(A-5): 479–504 Soman MK, Slingo JM (1997) Sensitivity of the Asian summer monsoon to aspects of the sea-surface temperature anomalies in the tropical Pacific Ocean. Q J Roy Meteorol Soc 123: 309–336 Sperber KR, Palmer TN (1996) Interannual tropical rainfall variability in general circulation model simulations associated with the AMIP. J Climate 11: 2727–2750 Sperber KR, Slingo JM, Annamalai H (2000) Predictability and the relationship between subseasonal and interannual variability during the Asian summer monsoon. Q J Roy Meteorol Soc 126: 2545–2574 Tiedtke M (1989) A comprehensive mass flux scheme for cumulus parameterization in large-scale models. Mon Wea Rev 17: 1779–1800 Van Leer B (1977) Towards the ultimate conservative difference scheme IV: A new approach to numerical convection. J Comput Phys 23: 276–299 Van Leer B (1979) Towards the ultimate conservative difference scheme: V: A second-order sequel to Godunov’s method. J Comput Phys 32: 101–136 Walker GT (1923) Correlations of seasonal variations of weather. VIII: A preliminary study of world weather. Mem India Meteor Dept 24: 75–131 Watanabe M, Shinoda M (1996) Long-term variability of Asian summer monsoon rainfall, during 1946–1988. J Meteor Soc Japan 74: 923–931 Xie P, Janowiak JE, Arkin PA, Adler R, Gruber A, Ferraro R, Huffman GJ, Curtis S (2003) GPCP pentad precipitation analyses: an experimental data set based on gauge observations and satellite estimates. J Climate 16: 2197–2214 Corresponding author’s address: Sanjib Kumar Deb, Centre for Atmospheric Sciences, I. I. T. Delhi, New Delhi-110016, India (E-mail:
[email protected];
[email protected])