On Climate Prediction in the Tropics Department of Meteorology University of Wisconsin-Madison

Stefan Hastenrath On Climate Prediction in the Tropics Department of Meteorology University of Wisconsin-Madison

Abstract 2. Concepts and approaches

Climatic disasters are common in many tropical regions, and rainfall The term prediction should refer to a statement about an anomalies in particular have a severe human impact. Accordingly, event from information on antecedent conditions. Climate both the World Climate Programme and the U.S. National Climate prediction is a statement on the quality of a (rainy) season or Program have identified climate prediction as a major objective. Approaches can be grouped into five major categories! (i) the ex- year as a whole, based on conditions more than a month pre- trapolation of empirically or theoretically deduced periodicities; (ii) ceding. Conversely, where a statement is made from concur- the assessment of statistical relationships between rainfall and var- rent conditions, terms such as specify, determine, deduce, or ious meteorological elements; (iii) the relation between rainfall in the infer, rather than predict, should be used. preseason and at the height of the rainy season; (iv) comprehensive diagnostic studies of climate and circulation anomalies combined Approaches can be grouped into five major categories: (i) with statistical methods; and (v) numerical modeling. the extrapolation of empirically or theoretically deduced pe- Methods pertaining to (iv) indicate the feasibility of empirically riodicities; (ii) the assessment of statistical relationships be- based climate prediction for certain tropical regions. For the tween rainfall and various meteorological elements; (iii) the drought-prone region of Northeast Brazil and Indonesia, in particu- relation between rainfall in the preseason and at the height of lar, it has been demonstrated on independent data sets that almost half of the interannual rainfall variability can be explained from an- the rainy season; (iv) comprehensive diagnostic studies of tecedent departures in the large-scale circulation. Application of climate and circulation anomalies combined with statistical these methods on an operational basis involves two simultaneous methods; and (v) numerical modeling. This classification is input data requirements: 1) they must be available in quasi real time; intended to bring some order into the diversity and is not 2) long (>10 years) homogeneous reference series of internally consistent parameters are needed, while absolute calibration is not es- meant to be dogmatic. sential. The practical benefit of climate forecasts appears to hinge on The following regional-reviews emphasize climate predic- societal and economic factors. tion proper and are limited to published works.

1. Introduction 3. Indian monsoon Large interannual variations of rainfall are an intrinsic part of tropical climate and can have a severe impact on human The work in India is among the earliest climate-prediction ef- activities. It is therefore not surprising that attempts to fore- forts in the tropics (Banerji, 1950; Normand, 1953; Rao and cast anomalous rainy seasons well in advance extend over a Ramamoorthy, 1960; Jaganriathan, 1960; Rao, 1965; Rao, century and that both the World Climate Programme (World 1976; Das, 1984,1986). H. F. Blanford issued monsoon fore- Meteorological Organization (WMO), 1980) and the U.S. casts from 1882 to 1885, calling attention in particular to the National Climate Program (National Climate Program Of- negative relation between southwest monsoon rainfall and fice, NOAA, 1980) declare climate prediction as a central the preceding winter's snowfall in the Himalayas. It was then objective. decided to issue monsoon forecasts annually, but they were Long-range forecasting for the midlatitudes has been re- held confidential in 1902-1906, during World War II, and viewed by Livezey and Jamison (1977), Nicholls (1980), and later (Jagannathan, 1969). Seasonal monsoon forecasting Preisendorfer and Mobley (1982). Charney and Shukla was the motivation for Sir Gilbert Walker's pioneering work (1981) suggested that climatic variability in the tropics on the Southern Oscillation. In 1910 he published a predic- should be more predictable because it was in large part due to tion formula for Indian monsoon forecasting containing slowly varying anomalies of the lower boundary of the at- four predictors (Walker, 1910). By 1924 he differentiated his mosphere. Despite these expectations, Nicholls (1980), in his work for two large regions of India, using two formulae with comprehensive review, was still unable to find climate-pre- a total of five predictors (Walker, 1924). Finally, Walker used diction work proper for the tropics. It is only in the 1980s that 28 predictors in six forecast formulae, differentiated re- major breakthroughs have been achieved. The purpose of the gionally (Banerji, 1950). At the time of Banerji's writing, the present note, based in part on Hastenrath (1985), is to sketch India Meteorological Department still used part of the afore- the history of climate prediction endeavors in the tropics; to mentioned predictors in five formulae, none of which con- summarize the progress made in the 1980s; and to discuss the tained upper-air information. Rao and Ramamoorthy (1960) tasks that lie ahead. After a summary of terminology and report that at the time of their work the India Meteorological methods, climate prediction endeavors are reviewed for var- Department issued three forecasts annually, using upper-air ious tropical regions and finally a synthesis and outlook are information in addition to predictors familiar from Walker's presented. work. Endeavors to forecast the onset of the southwest monsoon are more recent. Reddy (1977) proposes as predictor the © 1986 American Meteorological Society May 50-mb zonal wind component over Singapore.

696 Vol. 67, No. 6, June 1986

Unauthenticated | Downloaded 10/08/21 12:45 AM UTC Bulletin American Meteorological Society 697

FIG. 1. Monsoon rainfall in central India 1958-77. Dots denote the observed totals and open circles the amounts obtained from Kung and Sharif's (1982) regression equations based, successively, on all years except the particular year plotted. From Kung and Sharif (1982).

On the part of the India Meteorological Department (Ja- gannathan, 1960; Rao, 1965; Das, 1984, 1986), the desired standard of performance for forecasting monsoon rainfall has been variously expressed as being an 80 percent chance of success in a three-category scheme in accordance with the desired success rate stated by Sir Gilbert Walker (1928) for a two-category scheme. In appraising forecast power, it is es- FIG. 2. Scatter diagram of all-India rainfall index, period 1951— 70, with numbers indicating the years. Forecast values are obtained sential to distinguish between dependent and independent from regression equations based, successively, on all years except the data sets, but it is not clear whether this distinction has al- particular year plotted. Broken line denotes 45-degree angle. The ways been made. correlation coefficient of +0.92 between forecast and observed Banerjee et al. (1978) present a regression formula for pre- values is significant at the 0.1 percent level. From Wu (1984). dicting the ratio of the number of subdivisions with normal or above-normal rainfall in June-September to the total number of subdivisions (31). The latitude position of the 500- Kung and Sharif (1980, 1982) developed regression meth- mb ridge along 75°E in April serves as sole predictor, and the ods for forecasting both the Indian southwest monsoon rain- performance tested on 16 years of independent data appears fall and on the onset date over Kerala, South India, based on remarkable. April upper-air patterns in the India-Australia region and The important role of antecedent upper-air conditions for sea-surface temperature around India in the preseason. In Indian monsoon rainfall, as recognized in various earlier the regression models (Kung and Sharif, 1982), five predic- studies referred to above, is emphasized in the recent work of tors are retrieved for monsoon onset and six for rainfall. Thapliyal (1981,1982). He used the April latitude position of Kung and Sharif (1982) offer a measure of forecast perform- the 500-mb ridge over India as sole input to an ARIMA ance by comparing the observed rainfall amounts with the (auto-regressive integrated moving average) scheme and values calculated from regression equations based succes- applied the model thus obtained to predict the rainfall over sively on all years except the particular year calculated (Fig. peninsular India during the years 1977-80; the success for 1). It must be recognized that such specification of onset the four years being high. dates and rainfall totals within the 1958-78 time span may It appears that from Walker's time to the 1970s climate provide no fair measure of performance for forecasts proper prediction in India evolved from a primarily statistical basis of conditions beyond the base interval. As in Wu's (1984) (category ii) towards reliance on circulation diagnostics study, Kung and Sharif's (1982) approach relies on diagnos- (category iv). Reddy's (1977) paper is considered statistical tics on the large-scale circulation (category iv). (category ii), while the studies using the latitude position of In Wu's study, observations of important elements were the 500-mb ridge (Banerjee et al., 1978; Thapliyal, 1981, likewise limited to about two decades, namely the period 1982) imply some general circulation understanding (cate- 1951-72. From extensive diagnostic studies in the Indian gory iv). Thapliyal (1982) claims that his technique is super- Ocean sector, plausible predictors were identified including ior to the multivariate method hitherto used by the India Me- preseason temperature and pressure over land; temperature, teorological Department. For the early 1980s, he reports that pressure, wind, and cloudiness in certain areas of the Ara- this technique uses, as predictors, temperature in India, April bian Sea; a Southern Oscillation index; and various upper-air latitude position of the 500-mb ridge, and South American parameters. These elements served as input to a stepwise pressure. Apparently the India Meteorological Department multiple-regression scheme. The resulting regression model never explored the pressure, wind, cloudiness, and sea-sur- expresses the departure of a rainfall index in terms of a linear face-temperature fields in the Indian Ocean as possible pre- combination of departures in selected circulation parame- dictors of monsoon rainfall. This possibility was studied by ters. Then the aforementioned procedure of Kung and Sharif researchers outside India. (1982) was used, whereby values of an annual-rainfall index

Unauthenticated | Downloaded 10/08/21 12:45 AM UTC 698 Vol. 67, No. 6, June 1986

variations in the Indonesian region. He considers that forecasting for the east monsoon rainfall (June-September) holds the better prospect, and that pressure is to be taken as sole predictor. The subsequent decades-long effort by Dutch meteorologists includes most notably the works of Berlage (1927) and de Boer (1947), but the various schemes developed lack verification of performance on an independent data set, so that the practical usefulness of these exercises appears uncertain. Of greatest interest for Indonesian monsoon forecasting is Nicholls' (1981, 1983) direct follow-up of Braak's (1919) ideas more than 60 years earlier, namely to use Darwin pressure in the first half of the calendar year to predict Javan rainfall during the second semester (category iv). Nicholls (1981,1983) constructed a linear single-parameter regression model of Djakarta September-November rainfall versus Darwin August pressure during 1951-69 and then used this relationship to predict the rainfall during each of the years 1970-80. Results are shown in Fig. 3. The model proves cap- able to explain 44 percent of the interannual rainfall variance in the independent data set, a remarkable performance! Expanding on Nicholls' (1981, 1983) work for Indonesia, Nicholls et al. (1982) and Nicholls (1984) developed a scheme (category iv) for predicting the onset of the wet season (around the beginning of November and later) in northern Australia, again based on the preseason pressure at Darwin, FIG. 3. Predicted and observed September-November Djakarta an early onset date being heralded by low pressure. rainfall in 1970-80 with numbers indicating the year. Predictions made from observed Darwin August pressure using linear regression based on 1951-69 data. Correlation coefficient between predicted and observed is +0.66. From Nicholls (1983). 5. El Nifto are calculated for individual years from regression equations The intimate relation of El Nino events to the Southern Oscil- based on all other years of the series. Results are illustrated in lation pressure seesaw suggests that the episodes of warm Fig. 2. The coefficient of correlation between the rainfall de- water and abundant rainfall of the equatorial Pacific could partures thus calculated and the observed anomalies is +0.92 be predicted from surface-pressure observations (categories (significant at the 0.1 percent level), a performance compar- ii and iv). This possibility was explored by Quinn and Burt able to that of Kung and Sharif. It must be anticipated, how- (1970, 1972) and Quinn (1974, 1976). For a dependent data ever, that the performance will be inferior for prediction set they report that certain criteria of the pressure distribu- proper, that is, for years beyond the end of the base period tion allow them to specify between 76 and 88 percent of the used in the construction of regression models. years with heavy precipitation in the equatorial zone. How- Also pertinent to Indian monsoon-rainfall prediction is ever, operational forecasts beyond the dependent data set the recent work by Campbell et al. (1983), who show that a were discouraging. portion of interannual rainfall variability in North India may Barnett (1981, 1984) proposed that sea-surface-tempera- be related to the periodic solar-lunar tidal variations and that ture anomalies off Peru can be predicted from the antecedent this component of rainfall variability can be predicted a year wind field in the equatorial Pacific. Inoue and O'Brien (1984) in advance (category i, theoretically deduced periodicity). drove an ocean numerical model with observed winds and While the effect accounts for only about 10 percent of the concluded that El Nino could be predicted one to three total variance, this portion may in part be independent from months in advance from knowledge of the wind field alone. the variance accessed by other methods. Accordingly, me- Inoue and O'Brien's (1984) paper appears a historically im- chanical tidal forcing appears a useful addition to long-range portant step towards climate prediction by numerical model- prediction. ing (category v).

4. Indonesian rainfall 6. The S§cas of Northeast Brazil The interannual variability of rainfall over Indonesia very early attracted the attention of Dutch meteorologists. In par- The Secas, or droughts, of Northeast Brazil (Nordeste) rep- ticular, Braak (1919) already recognized the essential re- resent a climate problem which is unusually well defined in lationship between long-term pressure, wind, and rainfall the large-scale circulation setting, and which has an extraor-

Unauthenticated | Downloaded 10/08/21 12:45 AM UTC Bulletin American Meteorological Society 699 dinary economic and social impact. The possibility of predicting the droughts has accordingly long been a central concern. Walker (1928) proposed a method for predicting Northeast Brazil rainfall anomalies, based on statistical relationships with meteorological elements at distant locations (category ii). This line of approach was also attempted by Sampaio Ferraz (1929) and Serra (1956). Freise (1938), Markham (1967), and Serra (1973a,b) considered the relation of preseason rainfall to the precipitation amount at the peak of the rainy season (category iii). Extrapolations from the sunspot cycle (category i) were proposed by Sampaio Ferraz (1950) and recent unpublished work in Brazil draws on quasi-periodicities in Nordeste rainfall (category i). Hastenrath et al. (1984) found that the extrapolation of time-series characteristics of Nordeste rainfall may be useful for the prediction of five-year mean departures, but not for the rainfall anomalies of individual years. The general circulation background of the rainy season, concentrated around March-April, has been elucidated by the analysis of long-term ship observations in the tropical Atlantic (Hastenrath and Lamb, 1977), and these also formed the basis for the general circulation diagnostics of interannual rainfall variability (Hastenrath and Heller, 1977). Extensive diagnostic investigations into the interannual climate variability served to identify plausible candidates for climate prediction (Hastenrath et al., 1984). These were input FIG. 4. Scatter diagram of March-September Nordeste rainfall index based on observations through January. Regression period is to a stepwise multiple-regression scheme, and the resulting 1921-41,1946-56, and forecast period 1958-72, with numbers indi- regression equations were then used in a predictive mode on cating the years. Broken line denotes 45-degree angle. Correlation an independent data set. Results are illustrated in Fig. 4. Has- coefficient r = +0.68 is significant at the one percent level. tenrath et al. (1984) demonstrated that 46 percent of the interannual variance of March-September rainfall can be predicted from the antecedent general circulation departures in For southern Africa, Tyson and Dyer (1978,1980) extrap- the tropical Atlantic sector (category iv). This performance is olated the time-series characteristics of summer rainfall to comparable to that of Nicholls' (1981,1983) experiments for suggest the likelihood of above-normal rainfall in the 1970s Indonesia. and of drought in the 1980s, but they refrain from statements about individual years and caution against a possible non- stationarity of time-series behavior (category i). For the central Kenya highlands, Sansom (1955) described 7. Africa a procedure for the prediction of rainfall anomalies in the March-May and September-December rainy seasons based The widespread publicity given to the Sahel drought, which on statistical relationships with meteorological elements at began in the late 1960s, and its severe social and economic distant locations (category ii). Results presented for three consequences was followed by various papers proposing to years of independent data appear encouraging, but no in- forecast climatic disasters in this part of the world. Winstan- formation seems available concerning any operational appli- ley (1973a,b, 1974) ventured an extrapolation into the third cation after 1954. Since 1984, the Kenya Meteorological De- millenium from the rainfall record at eight stations spread partment has issued forecasts for the rainy seasons in the first between the Atlantic and India and a prediction of June-Sep- and second semesters of the year. The method has not been tember totals from June rainfall, an exercise queried by Bunt- published, but relies on a form of extrapolation of rainfall ing et al. (1975). Wood and.Lovett (1974) proposed rainfall time series of Kenya stations (category i). forecasts for Ethiopia based on the 11-year sunspot cycle. Faure and Gac (198 la,b) extrapolated a 30-year periodicity in the Senegal River runoff to infer a termination of the cur- rent drought around 1985. While the aforementioned papers 8. North Atlantic hurricanes (categories i and ii) give little attention to the underlying general circulation conditions, a recent note by Adedokun Drawing on his extensive investigations into the large-scale (1979) seems a step in the right direction. Drawing on the an- environmental factors for tropical storm formation, Gray nual-cycle latitudinal shifts of the intertropical discontinuity (1983,1984a,b) studied the general circulation diagnostics of and the associated steep meridional precipitation gradient interannual variability in the frequency of North Atlantic over West Africa, Adedokun (1979) suggests seasonal rain- hurricanes, and described a prediction method. Although the fall prediction for sub-Saharan West Africa based on pre- regression models developed on this basis have not yet been season precipitation at stations in the south (category iii). tested on an independent data set, Gray applies these diag-

Unauthenticated | Downloaded 10/08/21 12:45 AM UTC 700 Vol. 67, No. 6, June 1986 nostic results to the seasonal forecasting of North Atlantic on the basis of sound general circulation diagnostics. The so- seasonal hurricane activity (category iv). These experimental phistication of the statistical method chosen in the prediction forecasts are regularly distributed to interested colleagues, so exercise should be commensurate with the quality of the di- that a verification of performance over the years will be pos- agnostic analysis and of the input observations. The per- sible in due course. formance of the empirical method combining general circulation diagnostics and statistics should be verified on an independent data-set. Various results of this class of approach indicate that empirically based climate prediction 9. Hong Kong climate with a few months lead time is possible for certain tropical climate problems. In fact, studies reviewed above have dem- Bell (1976a,b, 1977) noted a correlation between the June- onstrated on independent data sets that almost half of the in- September rainfall at Hong Kong and the preceding Janu- terannual rainfall variance in Northeast Brazil and Indonesia ary's pressure difference, Irkutsk minus Tokyo. Bell (1976a) can be explained from antecedent circulation departures in a reports that successful forecasts of Hong Kong summer rain- predictive mode. In the further development of this method, fall based on these results have been issued for many years the following topics merit attention: secular variation of pre- (category ii), although there is a considerable secular varia- dictability; optimal length of the observation period on tion in these relationships (Bell, 1977). which the regression model is based; and updating of this regression base period. Application of these methods (category iv) on an operational basis rests on two simultaneous conditions: (a) input 10. Synthesis and outlook observations must be available in quasi real time; and (b) long (>10 years) homogeneous time series of internally con- In a large part of the tropics, climate prediction is of far sistent parameters are needed, while absolute calibration is greater practical importance than daily weather forecasting, not essential. Conventional ship and land observations, as a with the notable exception of such severe-weather systems as rule, comfortably satisfy criterion (b), but not (a). Satellite tropical storms and squall lines. It is then not surprising that sensing appears an obvious choice regarding criterion (a), attempts at forecasting the quality of rainy seasons well in but an internally consistent, satellite-derived climatology advance have a long tradition in some densely populated (sea-surface temperature, low-level wind, cloudiness, radia- agricultural lands of the tropics, and that climate prediction tion), which would have to extend over more than ten years has been declared a central objective of both the World Cli- (criterion b), has not yet been created and is well beyond the mate Programme (WMO, 1980) and the U.S. National Cli- reach of a small research group. Indeed, this would involve mate Program (National Climate Program Office, NO A A, the re-evaluation of satellite measurements accumulated in 1980). In the tropics, the interannual variability of rainfall is the course of the past decade. of primary practical concern. Substantial progress has been Ground-based monitoring of key atmospheric and oce- made in the 1980s in climate-prediction research for the trop- anic elements at strategic locations also merits attention for ics. It seems appropriate to consider the potential of the var- climate prediction (category iv). An encouraging example is ious approaches. the real-time monitoring of wind and pressure on St. Peter Prediction from extrapolation of time-series behavior and St. Paul Rocks (0°55' N, 29°26' W) in the equatorial (category i) hinges on two criteria, namely (a) stationarity, Atlantic under the auspices of the SEQUAL (Seasonal Re- and (b) percentage of variance explained by the quasi-peri- sponse of the Equatorial Atlantic) program (Garzoli and odic variation. Because of the latter aspect, this class of meth- Katz, 1984; Committee on Climatic Changes and the Ocean, ods seems more appropriate for the average of several years, UNESCO-ICSU, 1985) for the past several years. Drawing rather than for an individual year. Potentially attractive on earlier investigations into the general circulation diagnos- about this approach is the circumstance that forecasts can be tics of interannual rainfall variability in Northeast Brazil, it is envisaged more than a year in advance, so that the timely found that these recordings would be useful for predicting availability of input information poses no practical problem. the rainfall anomalies of Brazil's Nordeste. Accordingly, this Both theoretically and empirically deduced periodicities monitoring effort initiated with oceanographic objectives have recently been proposed for climate prediction. merits support also because of its potentially important role Purely statistical exercises (category ii) do not appear satis- in climate prediction. factory regarding our understanding of the general circula- Furthermore, numerical models (category v) have been tion, and they do not seem to have proven successful in the suggested as a possible tool in climate prediction. The appli- work of the 1980s. cation of an ocean numerical model in a predictive mode, as Seasonal persistence of rainfall (category iii) merits atten- reviewed above in relation to the El Nino phenomenon, must tion for prediction in certain regions, and has in fact been be regarded as a milestone towards climate prediction by de- combined to advantage with approach (iv) for Brazil's Nor- terministic methods. However, while some of the empirical deste (Hastenrath et al., 1984). methods described here could already be translated into op- Considerable prospects for rainfall prediction with a few eration at least for certain tropical regions, numerical models months lead time are offered by an approach (category iv) may achieve a comparable performance only in the more- which combines extensive diagnostic investigations into the distant future. At any rate, a sound empirical understanding interannual variability of circulation and climate with statis- seems a prerequisite for the design of realistic coupled at- tical methods. Most essential is the choice of good predictors mosphere-ocean numerical models.

Unauthenticated | Downloaded 10/08/21 12:45 AM UTC Bulletin American Meteorological Society 701

Climate prediction is intended to be practical and useful. Weather, 31, 208-212. In order to compare the predictive potential of methods it , 1976b: Seasonal forecasts and Northern Hemisphere circula- seems desirable to express the performance (obtained on an tion anomalies. Weather, 31, 282-292. independent data set) in the first place as percentage of ex- , 1977: Changes in sign of the relationship between sunspots and plained variance; in subsequent practical applications it may pressure, rainfall and the monsoons. Weather, 32, 26-32. Berlage, H. P., 1927: East-monsoon forecasting in Java. Koninklijk be appropriate to present amounts by categories, or other Magnetisch en Meteorologisch Observatorium te Batavia, In- formulations for the user. Various questions arise: What per- donesia, Verhandelingen, No. 20, 42 pp. [available from Royal centage of interannual rainfall variance should be explained Netherlands Meteorological Institute, De Bilt, 3730 AE, Nether- for the forecast to be useful? Should predictions be publicized lands] or held confidential? What forms of forecasts and what lead Braak, C., 1919: Atmospheric variations of short and long duration in times are needed for practical application? In fact, what use the Malay Archipelago and neighboring regions, and the possibility will be made at all of the climate predictions? In his attempt to forecast them. Koninklijk Magnetisch en Meteorologisch Ob- at ascertaining the benefit of a climate forecast for sub-Sa- servatorium te Batavia, Indonesia, Verhandelingen No. 5, 57 pp. haran West Africa, Glantz (1977) concluded that, due to var- [available from Royal Netherlands Meteorological Institute, De ious deficiencies in the infrastructure, the value of a long- Bilt, 3730 AE, Netherlands] range forecast would have been limited. However, the lesson Bunting, A. H., M. D. Dennett, J. Elston, and J. R. Milford, 1975: Seasonal rainfall forecasting in West Africa. Nature, 253,622-623. to be learned is not that climate forecasts would be intrinsi- Campbell, W. H., J. B. Blechman, and R. A. Bryson, 1983: Long-pe- cally useless, but rather that it will take some difficult work to riod tidal forcing of Indian monsoon rainfall: an hypothesis. J. render them useful. With a view towards the practical appli- Atmos. Sci., 22, 287-296. cation of climate forecasts, Lamb (1981) suggests that the Charney, J., and J. Shukla, 1981: Predictability of monsoons. Mon- human activities most severely affected by climatic fluctua- soon dynamics, Sir J. Lighthill and R. P. Pearce, Eds., pp. 99-104, tions should be identified and that the flexibility of regional Cambridge University Press, 735 pp. economies should be assessed, in particular their ability to Committee on Climatic Changes and the Ocean, UNESCO-ICSU, adjust sufficiently to benefit substantially from climate fore- 1985: CCCO panel on tropical Atlantic Ocean climate studies, casts. For many countries of the Third World the handicaps third session, 9-13 Sept. 85, Rio de Janeiro, Paris, 100 pp. referred to by Glantz (1977) and Lamb (1981) are consider- Das, P. K., 1984: The monsoons—A perspective. India National able. To develop a societal flexibility and responsiveness Science Academy, Perspective Report Series 4, New Delhi, 52 pp. , 1986: Monsoons. Fifth IMO lecture, World Meteorological appears an important task also for ensuring the practical Organization, Geneva, in press. benefit of climate predictions. These demands go beyond mete- de Boer, H. J., 1947: On forecasting the beginning and the end of the orological expertise. Clearly, the wisdom of agriculturists, dry monsoon in Java and Madura. Koninklijk Magnetisch en Me- economists, and other planners, is called upon, not only in teorologisch Observatorium te Batavia, Indonesia, Verhandelin- the eventual application, but also in the very design of the gen No. 32, 20 pp. prediction effort. Finally, the meteorological community Faure, H., and J. Y. Gac, 1981a: Will the Sahelian drought end in must give serious consideration to the credibility of climate 1985? Nature, 291, 475-478. forecasts: publication of the method and verification of re- ,and , 1981b: Senegal river runoff, reply. Nature, 293,414. sults are imperative. Freise, F. W., 1938: The drought region of northeastern Brazil. Geogr. Rev., 28, 363-378. Garzoli, S. L., and E. J. Katz, 1984: Winds at St. Peter and St. Paul Rocks during the first SEQUAL year. Geophys. Res. Lett., 11, 715-718. Acknowledgments. This work was supported by National Science Foundation Grant ATM84-13575. I thank Peter Lamb for com- Glantz, M. H., 1977: The value of a long-range weather forecast for ments on a draft version of this paper. the West African Sahel. Bull. Amer. Meteor. Soc., 58, 150-158. Gray, W. M., 1983: Atlantic seasonal hurricane frequency. Part 1: El Nino and 30 mb QBO influences; Part 2: Forecasting its variability. Atmospheric Sciences Paper No. 370, Dept. of Atmospheric Sciences, Colorado State University, 48 pp. References , 1984a: Atlantic seasonal hurricane frequency. Part 1: El Nino and 30-mb quasi-biennial oscillation influences. Mon. Wea. Rev., Adedokun, J. A., 1979: An empirical method for forecasting the 112, 1649-1668. April-August precipitation in parts of West Africa. Pre-WAMEX , 1984b: Atlantic seasonal hurricane frequency. Part 2: Fore- Symposium on the West African Monsoon, D. O. Adefolalu Ed., casting its variability. Mon. Wea. Rev., 112, 1669-1683. Nigerian Meteorological Services, 600 pp. Hastenrath, S., 1985: Climate and Circulation of the Tropics, Reidel, Banerjee, A. K., P. N. Sen, and C. R. V. Raman, 1978: On foreshad- Dordrecht, Boston, Lancaster, Tokyo, 455 pp. owing southwest monsoon rainfall over India with mid-tropo- , and L. Heller, 1977: Dynamics of climatic hazards in northeast spheric circulation anomaly of April. Indian J. Meteor. Hydrol. Brazil. Quart. J. Roy. Meteor. Soc., 103, 77-92. Geophys., 29, 425-431. , and P. Lamb, 1977: Climatic Atlas of the Tropical Atlantic and Banerji, S. K., 1950: Methods of foreshadowing monsoon and winter Eastern Pacific Oceans. University of Wisconsin Press, 112 pp. rainfall in India. Indian J. Meteor. Geophys., 1, 4-14. , M.-C. Wu, and P.-S. Chu, 1984: Towards the monitoring and Barnett, T. P., 1981: Statistical relations between ocean-atmosphere prediction of Northeast Brazil droughts. Quart. J. Roy. Meteor. fluctuations in the tropical Pacific. J. Phys. Oceanogr., 11, Soc., 110, 411-425. 1043-1058. Inoue, M., and J. J. O'Brien, 1984: A forecasting model for the onset , 1984: Prediction of the El Nino of 1982-83. Mon. Wea. Rev., of a major El Nino. Mon. Wea. Rev., 112, 2326-2337. 112, 1403-1407. Jagannathan, P., 1960: Seasonal forecasting in India, a review. Me- Bell, G. J., 1976a: Seasonal forecasts of Hong Kong summer rainfall. teorological Office, Poona, 79 pp.

Unauthenticated | Downloaded 10/08/21 12:45 AM UTC 702 Vol. 67, No. 6, June 1986

Kung, E. C., and T. A. Sharif, 1980: Regression forecasting of the Basu, P. R. Pisharothy, K. R. Ramanathan, and U. K. Bose, Eds., onset of the India summer monsoon with antecedent upper air New Delhi, Hindi Union Press, 271 pp. conditions. J. Appl. Meteor., 19, 370-380. Rao, Y. P., 1976: Southwest monsoon. India Meteorological De- , and , 1982: Long-range forecasting of the Indian summer partment, Meteorological Monograph, Synoptic Meteorology, monsoon onset and rainfall with upper air parameters and sea sur- No. 1/1976, 307 pp. face temperature. J. Meteor. Soc. Japan, 60, 672-681. Reddy, S. J., 1977: Forecasting the onset of Southwest monsoon Lamb, P. J., 1981: Do we know what we should be trying to fore- over Kerala. Indian J. Meteor. Hydrol. Geophys., 28, 113-114. cast—climatically? Bull. Amer. Meteor. Soc., 62, 1000-1500. Sampaio Ferraz, J. de, 1929: Sir Gilbert Walker's formula for Ceara Livezey, R. E., and S. W. Jamison, 1977: A skill analysis of Soviet droughts: suggestion for its physical explanation. Meteor. Mag., seasonal weather forecasts. Mon. Wea. Rev., 105, 1491-1500. 64, 81-84. Markham, C. G., 1967: Climatological aspects of drought in north- , 1950: Iminencia duma "Grande" Seca Nordestina. Rev. Brasil. eastern Brazil. Ph.D. dissertation, University of California, Berke- Geogr., 12, 3-15. ley, 352 pp. Sansom, H. W., 1955: Prediction of the seasonal rains of Kenya by National Climate Program Office, NOAA, 1980: National Climate means of correlation. Weather, 10, 81-86. Program. Washington, D.C., 101 pp. Serra, A., 1956: As Secas do Nordeste. Bol. Geog., 14, 269-270. Nicholls, N., 1980: Long-range weather forecasting: Value, status, , 1973a: Previsao das secas nordestinas—Testes estatisticas. Bol. and prospects. Rev. Geophys. Space Phys., 18, 771-788. Geogr., 32, 78-104. , 1981: Air-sea interaction and the possibility of long-range , 1973b: Previsao das secas nodestinas. Banco do Nordeste do weather prediction in the Indonesian Archipelago. Mon. Wea. Brasil, Fortaleza, 55 pp. Rev., 109, 2435-2443. Thapliyal, V., 1981: ARIMA model for long-range prediction of mon- , 1983: Prospects for empirical long-range weather prediction. soon rainfall in peninsular India. India Meteor. Monograph, Clima- Proc. of the WMO/CAS/JB expert study meeting on long-range tology, 12, India Meteorological Department, Poona, India, 22 pp. forecasting, Princeton, WMO, Programme on Weather Prediction , 1982: Stochastic dynamic model for long-range prediction of Research, Long-Range Forecasting Research Publication Series, monsoon rainfall in peninsular India. Mausam, 33, 399-404. Geneva, 238 pp. Tyson, P. D., and T. G. J. Dyer, 1978: The predicted above-normal , 1984: A system for predicting the onset of the North Austral- rainfall of the seventies and the likelihood of droughts in the eight- ian wet season. J. Climatol., 4, 425-435. ies in South Africa. South African J. Science, 74, 372-377. , J. L. McBride, and R. J. Ormerod, 1982: On predicting the , and , 1980: The likelihood of droughts in the eighties in onset of the Australian wet season at Darwin. Mon. Wea. Rev., South Africa. South African J. Science, 76, 340. 110, 14-17. Walker, G. T., 1910: Correlation in seasonal variations of weather Normand, C. W. B., 1953: Monsoon seasonal forecasting. Quart. J. II. Mem. Indian Meteor. Dept., 21, 22-45. Roy. Meteor. Soc., 79, 463-473. , 1924: Correlation in seasonal variations of weather, VII: The Preisendorfer, R. W.,and C. D. Mobley, 1982: Climate forecast veri- local distribution of monsoon rainfall. Mem. Indian Meteor. Dept., fications, U.S. mainland, 1974-82. NOAA Technical Memoran- 23, 23-40. dum ERL PMEL-36, Seattle, Washington, 235 pp. Walker, G. T., 1928: Ceara (Brazil) famines and the general air move- Quinn, W. H., 1974: Monitoring and predicting El Nino invasions. J. ment. Beitr. Phys. d. freien Atm., 14, 88-93. Appl. Meteor., 13, 825-830. Winstanley, D., 1973a: Recent rainfall trends in Africa, the Middle , 1976: An improved approach for following and predicting East and India. Nature, 243, 464-465. equatorial Pacific changes and El Nino. Proceedings of the NOAA , 1973b: Rainfall patterns and the general atmospheric circula- Climate Diagnostics Workshop, 4-5 Nov. 1976, Washington, tion. Nature, 245, 190-194. D.C., article 21, 5. , 1974: Seasonal rainfall forecasting in West Africa. Nature, 248, , and W. V. Burt, 1970: Prediction of abnormally heavy precipi- 464. tation over the equatorial Pacific dry zone. J. Appl. Meteor., 9, Wood, C., and R. R. Lovett, 1974: Rainfall, drought, and the solar 20-28. cycle. Nature, 231, 594-596. , and , 1972: Use of the Southern Oscillation in weather World Meteorological Organization, 1980: Outline plan and basis prediction. J. Appl. Meteor., 11, 616-628. for the World Climate Programme 1980-1983. WMO-No. 540, Rao, K. N., 1965: Seasonal forecasting—India. WMO Technical Geneva, 64 pp. Note No. 66, WMO No. 162 TP. 79, Geneva, pp. 17-30. Wu, M.-C., 1984: On the interannual variability of the Indian mon- , and K. S. Ramamoorthy, 1960: Seasonal (monsoon) rainfall soon and the Southern Oscillation. Ph.D. dissertation, University forecasting in India. Symposium on Monsoons of the World, S. of Wisconsin, Madison, Wis., 110 pp. •

Unauthenticated | Downloaded 10/08/21 12:45 AM UTC