An Advance in Indian Monsoon Rainfall Prediction

Transcription

1 An Advance in Indian Monsoon Rainfall Prediction K. Krishna Kumar 1,+,, Martin P. Hoerling 2* and Balaji Rajagopalan 1,3 1 Cooperative Institute for Research in Environmental Sciences, Universty of Colorado, Boulder, CO USA 2 NOAA Climate Diagnostics Center, Boulder, CO USA 3 Department of Civil Environmental and Architectural Engineering, University of Colorado, Boulder, CO USA One Sentence Summary: A popular practice for seasonal forecasting based on uncoupled atmospheric models is shown to be flawed for the Indian region resulting in abysmal prediction skill, and we demonstrate a surge in skill by treating the yearly swings in Indian monsoon rains as a coupled ocean-atmosphere problem. Submitted to Science - 29th April, Permanent Affiliation: Indian Institute of Tropical Meteorology, Dr. Homi Bhabha Road. Pune India *Corresponding Author Martin.Hoerling@noaa.gov Page No. 1

2 An Advance in Indian Monsoon Rainfall Prediction K. Krishna Kumar 1,+, Martin P. Hoerling 2,* and Balaji Rajagopalan 1,3 1 Cooperative Institute for Research in Environmental Sciences, Universty of Colorado, Boulder, CO USA 2 NOAA Climate Diagnostics Center, Boulder, CO USA 3 Department of Civil Environmental and Architectural Engineering, University of Colorado, Boulder, CO USA A convergence of remarkable advances in climate observing systems and computing capacity have propelled dynamical methods for prediction to the forefront. Yet, these emergent approaches have yielded inexplicably low skill when applied to seasonal Indian monsoon rains. We show that the practice of using sea surface temperatures as specified forcing for atmospheric general circulation models is flawed for the Indian monsoon, and artificially suppresses skill. We demonstrate significant improvements in the skill of Indian monsoon predictions for atmospheric models coupled to, and fully interactive with the underlying ocean. Of particular importance is the proper representation of air-sea energy exhange over the Indian Ocean warm pool. Forecasting Indian summer monsoon rainfall has an unparalleled lineage within the history of climate prediction. Routine efforts were born of the great 1877 Indian famine that resulted from failed monsoon rains. Subsequent statistical models based on correlation methods developed by Sir Gilbert Walker (1) form the foundation of modern Indian monsoon predictions (2,3). Such empirical approaches are skillful, and have now been placed upon a firm physical basis owing to abundance of observations on the land, ocean and within the free atmosphere (4,5). Such progress notwithstanding, notable failures in Indian monsoon rainfall forecasts for recent years (6) compels an assesment of the prospects for new prediction methods. Page No. 2

3 Year-to-year swings in monsoon rainfall are linked to global forcing functions that evolve slowly compared to the length of the monsoon season itself, and oceanic effects appear to be primary sources (7,8,9). An important question concerns the upper bound of monsoon predictive skill within a scenario of perfectly forecast ocean conditions. Such inquiries have spurred recent atmospheric general circulation model (AGCM) studies in which the observed global sea surface temperatures (SSTs) have been specified, and the atmospheric responses to those have been compared to observations (10,11,12,13). Tantalizing among AGCM results are their intimation for high prospects in furthering Indian monsoon prediction skill. To quantify this potential capability of dynamical techniques, we calculate Indian summer monsoon rainfall predictability in 10 different AGCMs (14). Each model is subjected to the observed evolution of monthly global SSTs during , and numerous simulations are produced for each AGCM (15). Skill is measured by the 50-year averaged correlation between June-September Indian monsoon rainfall occurring in one model realization with that occurring in individual parallel runs of the same model (16). The results for all models are summarized by plotting the probability density function (PDF) of correlations. As shown in Fig. 1 (red curve), a 0.7 median correlation indicates that 50% of the year-to-year Indian monsoon rainfall variations are oceanic controlled, indicating a high predictability within this perfect model scenario. Note further that the vast majority of the individual model skill estimates exceed an estimate of skill using empirical methods (Fig. 1, black dashed line (17)). Such high correlation skill indicates the AGCMs 50- Page No. 3

4 year Indian summer monsoon rainfall time history is reproducible from one integration to another. This behavior is further illustrated by plotting the monsoon rainfall time series for each realization of a particular model (Fig. 2). The individual runs are near facsimiles of one another. Predictability in this perfect model scenario is thus ostensibly judged to be high to the extent the SSTs could themselves be accurately forecast. Paradoxically, the actual skill of dynamical predictions of Indian monsoon rainfall have proven to be woefully inadequate. This is visible in Fig. 2 by the poor correspondence between fluctuations of the observed Indian monsoon rainfall anomalies (blue dots) and the AGCM s ensemble mean values (yellow dots). Climate prediction centers have also reported little skill in their dynamical attempts to foretell Indian summer monsoon rain (18). The practice has employed a two-tier approach in which the future state of global SSTs are first predicted, and then specified as boundary forcings for AGCM integrations. Their low skill is unlikely due to poor forecasts of the SST states themselves --- AGCM simulations of Indian monsoon rainfall forced with the actual observed global SSTs also have low skill. The temporal correlation between each of our 10 models ensemble mean Indian monsoon rainfall and observations is again summarized with a PDF (Fig. 1, blue curve). The near-zero median value indicates that the AGCMs explain virtually none of the observed Indian monsoon rainfall variations, and thus their skill is effectively zero. It is evident that the high, theoretical perfect model skill scores are not being achieved. This is almost certainly not due to an insensitivity of the AGCMs to SST Page No. 4

5 variations in light of the models capacity to reproduce their own Indian monsoon rainfall time series. One can also discount the possibility that the actual Indian monsoon variations originate from internal atmopsheric dynamics alone (and thus predictability should be low), in so far as empirical approaches based on an index of El Niño have demonstrable skill (Fig. 1, vertical line). Instead, the Indian monsoon signal resulting from the models SST sensitivity is different from that occurring in nature, and is furthermore unsynchronized in time with the variations of the observed ocean-forced signal. Comparison of AGCM (Fig. 3, top) and observed (bottom) monsoon season rainfall and surface temperature anomalies during El Niño/Southern Oscillation (ENSO) illustrates a characteristic feature of model failure, one that also provides a clue for possible causes. Much of India suffers drought during ENSO's warm phase, and this is embedded with a larger scale drying spanning the entire Intertropical Convergence Zone of the Indian Ocean (bottom left). Anomalously warm surface temperatures coincide with this drying over land and sea (bottom right), conditions known to result from the atmospheric feedback by anomalous winds that act to suppress rainfall as well as upwelling in the cross-equatorial flow region (19,20). Not only does the AGCM fail to reproduce Indian drought, it actually generates a large scale increase in rainfall over the warmest SSTs of the Indian Ocean and Arabian Sea (top left). It is our contention that this misrepresentation of air-sea energy exchange, also identified in other model studies, is directly responsible for degrading skill (21,22). Page No. 5

6 Various other theories have been previously proposed to explain such low skill. One involves poor simulation of the Indian monsoon mean climate (11,12). We however, do not find a strong relation between simulation skill and the realism of the mean climatology for the 10 AGCMs studied herein. Another possibility is an inability of the AGCMs to realistically simulate the ENSO-Indian monsoon statistical relation (23,24). We find that the absence of simulation skill among the 10 AGCMs cannot be explained by an absence of an ENSO link, and in fact several of the poor performing AGCMs possess realistic ENSO-Indian monsoon relations. It has also been proposed that so-called super-ensemble methods are important for harvesting the AGCM skill (25). We have found that the correlation skill of Indian monsoon rainfall is not greatly increased by combining all of the available model runs into a single 100 member ensemble, the correlation skill of which is still close to the zero median value of Fig. 1. Instead, we discover skill can be materially increased by treating the year-to-year swings in Indian monsoon rains as a coupled ocean-atmosphere problem. Such interactions are not represented in atmospheric GCMs wherein the two-tiered design presupposes Indian monsoon variations to be described as a purely forcing-response system. This popular approach to Indian monsoon forecasting masks the true skill of dynamical models, since it is calculated for a fictitious system in which no air-sea coupled interactions exist. The absence of simulation skill in the AGCMs (Fig. 1, blue curve) is thereby unlikely to be an accurate measure for the prospects of dynamical monsoon predictions. We demonstrate below that coupling is required to generate realistic air-sea interactions that are essential for Indian monsoon prediction. Page No. 6

7 In a companion set of experiments using one of the general circulation models, SSTs in the tropical eastern Pacific are specified as before for , but SSTs elsewhere in the World Ocean are free to evolve according to coupled air-sea interactions with a mixed layer ocean model (26). These simulations thus retain the ENSO variations of the eastern Pacific. SST variations over the so-called warm pool reaches of the Indian Ocean and seas adjacent to Asia are permitted to interact fully with the fluctuating monsoon circulation system. The scatter diagrams of Fig. 4 (right) compare the 50 years of modeled versus observed Indian summer monsoon rainfall using both uncoupled (top) and coupled (bottom) approaches. The relationship is purely random in the uncoupled simulations, and the average correlation skill is 0.0. A strong linear relationship emerges in the coupled simulations, and the skill increases to This remarkable change in skill is highlighted by the magenta circles and arrow in Fig. 1, and it can be seen that the coupled model skill now exceeds every member of the 10 AGCMs (27). The robustness of these findings was confirmed with a second set of identically designed coupled experiments using a different model, the results of which are highlighted by the blue circles and arrow in Fig. 4 (28). The spatial plots of Fig. 4 illustrate the correlation between Indian summer monsoon rainfall and SSTs in the uncoupled model (top-left), the coupled model (middle-left), and the observations (bottom-left). The large-scale pattern has not greatly changed by introducing coupling in this particular model. One can discern, however, a change in SST correlations over the Arabian Sea and Bay of Bengal in the coupled model that now agree more closely with observations. This re-emphasizes that the realism of Page No. 7

8 ENSO-monsoon relations alone is an insufficient condition for realizing predictive skill, and that correct treatment of local warm pool air-sea interactions is of great importance. Our results reveal that the ENSO contribution to Indian monsoon skill can be masked in AGCMs, most probably by misrepresenting air-sea interactions over the warm pool regions of the Indo-west Pacific ocean. The warming (cooling) of the Indian Ocean, Arabian Sea, and Bay of Bengal during El Niño (La Niña) is typically associated with reduced (increased) cloud cover, yet such SST anomalies prescribed in AGCMs tend to excite increased (decreased) cloud cover and convection that encompass the adjacent continent. During warm ENSO events, surface evaporation over the anamolously warm western Indian Ocean and Arabian Sea is observed to decrease, yet is simulated to increase in the AGCMs thereby providing an additional erroneous moisture source for monsoon rains (21). The specified warm pool SSTs in AGCMs thus yield an unrealistically strong negative feedback on the monsoon that competes with the remote ENSO influence as reported in other recent studies (24,25,29). The practical implication of our study is the ensuing complete forfeit of skill using uncoupled dynamical approaches. Skill can be dramatically recovered by even simple treatments of the coupled nature of air-sea interactions using mixed layer ocean models. A systematic intercomparison of monsoon predictability throughtout the tropics in coupled and uncoupled models, and a re-evaluation of dynamical methods, thus is clearly required. Page No. 8

9 References and Notes 1. G.T. Walker, Memoirs of India Meteorological Department, 24, 275 (1924). 2. S. Hastenrath, Climate Dynamics of the Tropics, Kluwer Academic Publishers, Dordrecht, The Netherlands, 488 pp. (1991). 3. K. Krishna Kumar, M.K. Soman and K. Rupa Kumar, Weather, 50, 449 (1995). 4. J. Shukla, in Monsoons, J.S. Fein and P.L. Stephens, Eds. (Wiley, New York), 399 (1987). 5. M. Rajeevan, Curr. Sci., 81, 1451 (2001). 6. S. Gadgil, et al., Curr. Sci., 83, 394 (2002). 7. J. G. Charney and J. Shukla, in Monsoon Dynamics, J. Lighthill, Eds (Cambridge University Press), 99 (1981). 8. T. N. Palmer, Proc. Indian Natn. Sci. Acad., 60, 57 (1994). 9. P.J. Webster et al., J. Geophys. Res., 103, C7, (1998). 10. W.L. Gates et al., Bull. Amer. Meteor. Soc., 80, 29 (1999). 11. K.R. Sperber, T.N. Palmer, J. Clim., 9, 2727 (1996). 12. S. Gadgil and S. Sajani, Climate Dyn., 14, 659 (1998). 13. I. Kang, et al., Climate Dyn., 19, 383 (2002). 14. Nearly 50-year long integrations beginning 1950 from 10 AGCMS are utilized. The models and their ensemble sizes are: 1. MPI-ECHAM4 (24), 2. MPI-ECHAM3 (10), 3. GFDL AM2 (10), 4. NASA (9), 5. Scripps-ECPC (7), 6. NCEP (13), 7. ARPEGE (8), 8. NCAR-CCM3(12), 9. NCAR-CAM2 (15) and 10. GFDL R30 (4). The models typically have a ~300km horizontal resolution. These model ensembles, obtained by forcing them Page No. 9

10 with observed SSTs, are referred as GOGA runs. 15. Each of the ensemble members differ by their atmospheric initial conditions while the same observed SSTs are used as lower boundary forcing. 16. Correlations are computed between monsoon rainfall time series from each ensemble member with those from other members. This is repeated for all the models, resulting in more than 1000 correlation coefficients that form the basis for the PDF (red curve) in Figure 1. It should be noted that this is a conservative estimate of the upper bounds. Higher estimates occur if the correlation is computed between each member and the ensemble mean of remaining members. The median correlation would increase to 0.8. Monsoon rainfall from the models is the total rainfall over 8-30N and 70-90E. PDFs are estimated using kernel density estimators (Bowman and Azzalini, Applied Smoothing Techniques for Data Analysis, Oxford University Press, New York, 1997). These are data-driven techniques with ability to capture any arbitrary density function with no prior assumptions, unlike, parametric methods. 17. The skill of empirical models is taken to be the correlation between observed monsoon rainfall (June-September) and the NINO3.4 index, defined as SSTs averaged over 5ºN-5ºS and 170ºW-120ºW. Observed monsoon rainfall index is based on Parthasarathy et al., Theor. Appl. Climatol., 49, (1994) and available from ( 18. A.G. Barnston, et al, Bull. Amer. Met. Soc., 84,1783 (2003). 19. S.A. Klein, B.J. Soden and N-C. Lau, J. Clim., 12, 917 (1999). 20. M. Alexander et al., J. Clim., 15, 2205 (2002). Page No. 10

11 21. S. C. Clemens and R. J. Oglesby, Paleoceanography, 7, 633 (1992). 22. A. Kumar and M.P. Hoerling, J. Geophys. Res., 103, 8901 (1998). 23. N-C. Lau, M.J. Nath, J. Clim., 13, 4287 (2000). 24. Wu and B. Kirtman J. Clim., (in press) (2004). 25. T.N. Krishnamurti, et al., J. Climate, 13, 4196 (1999). 26. SSTs are prescribed in the Tropical Pacific between 15ºS and 15ºN, and from 172ºE to the South American coast in the GFDL R30 AGCM. For all remaining ocean grid points that are ice-free through out the year, the SST conditions were predicted by a simple mixed-layer model (MLM). Further experimental details are described in N-C., Lau, M.J. Nath, J. Clim., 16, 3 (2003). 27. All possible combination of ensembles of size 4 is drawn from the 16 member MLM runs. The ensemble mean of each combination is correlated with the observed rainfall. The 5th and 95th percentile of the resulting correlation coefficients form the confidence interval. The correlation coefficient from the 4 run sample of GOGA falls outside of this interval, indicating that the increase in correlation in the MLM runs is independent of ensemble size. 28. A. Giannini, R. Saravanan and P. Chang, Clim. Dyn. (in press) (2004). 29. B. Wang, I. Kang and J. Lee, J. Clim., 17, 803 (2004). 30. The funding for the first author via CIRES Visiting Fellowship is thankfully acknowledged. Funding provided to the second author by NOAA s Office of Global Programms is gratefully acknowledged. We also acknowledge the contribution by T. Xu and J. Eischeid of CDC. Page No. 11

12 Figure Captions: Figure 1. Probability density functions (PDFs) of correlation skill of June-September Indian monsoon rainfall based on a theoretical perfect model analysis (red curve), and based on the actual skill compared to observed all Indian monsoon rain (blue curve). Analysis is based on 10 AGCMs forced with specified global SSTs of Tick marks denote the correlation between individual model realizations on the Indian monsoon rain time series. Circles denote the correlation between ensemble GCM and observed Indian monsoon rain times series. Closed, colored circles denote the skill of two of the AGCM coupled to a mixed layer model. Arrows denote the change in skill between pairs of uncoupled and coupled GCM simulations. Figure 2. The time series of June-September Indian monsoon rainfall occurring in each of 12 NCAR/CCM3 AGCM simulations (black lines). Red line indicates the ensemble mean. The observed rainfall is shown as blue dots which can be compared to the AGCM s ensemble mean values shown by yellow dots. Figure 3. Monsoon season June-September warm minus cold El Niño/Southern Oscillation climate signals of, (A) ensemble mean rainfall from NCAR/CCM3 AGCM, (B) same as A but for surface temperatures, (C) Same as A but for satellite estimated rainfall derived from outgoing long-wave radiation (OLR) (D) same as B but for observed temperature. Note that SSTs have been prescribed in the AGCM. Period of analysis is , except for OLR. Page No. 12

13 Figure 4. Correlation maps of (A) observed SSTs and monsoon rainfall simulated from uncoupled GFDLR30 (GOGA) model, (B) simulated SSTs and monsoon rainfall from the coupled (MLM) model, (C) observed SSTs and observed rainfall. Scatter plots of observed and simulated monsoon rainfall for each year during from GOGA and MLM are shown in (D) and (E), respectively. Note that in the coupled model, SSTs have been prescribed between 15ºN-15ºS and 172ºE-80ºW. Page No. 13

14 Figure 1. Probability density functions (PDFs) of correlation skill of June-September Indian monsoon rainfall based on a theoretical perfect model analysis (red curve), and based on the actual skill compared to observed all Indian monsoon rain (blue curve). Analysis is based on 10 AGCMs forced with specified global SSTs of Tick marks denote the correlation between individual model realizations on the Indian monsoon rain time series. Circles denote the correlation between ensemble GCM and observed Indian monsoon rain times series. Closed, colored circles denote the skill of two of the AGCM coupled to a mixed layer model. Arrows denote the change in skill between pairs of uncoupled and coupled GCM simulations. Page No. 14

15 Figure 2. The time series of June-September Indian monsoon rainfall occurring in each of 12 NCAR/CCM3 AGCM simulations (black lines). Red line indicates the ensemble mean. The observed rainfall is shown as blue dots which can be compared to the AGCM s ensemble mean values shown by yellow dots. Page No. 15

16 Figure 3. Monsoon season June-September warm minus cold El Niño/Southern Oscillation climate signals of, (A) ensemble mean rainfall from NCAR/CCM3 AGCM, (B) same as A but for surface temperatures, (C) Same as A but for satellite estimated rainfall derived from outgoing long-wave radiation (OLR) (D) same as B but for observed temperature. Note that SSTs have been prescribed in the AGCM. Period of analysis is , except for OLR. Page No. 16

17 Figure 4. Correlation maps of (A) observed SSTs and monsoon rainfall simulated from uncoupled GFDLR30 (GOGA) model, (B) simulated SSTs and monsoon rainfall from the coupled (MLM) model, (C) observed SSTs and observed rainfall. Scatter plots of observed and simulated monsoon rainfall for each year during from GOGA and MLM are shown in (D) and (E), respectively. Note that in the coupled model, SSTs have been prescribed between 15ºN-15ºS and 172ºE-80ºW. Page No. 17