ASSESSMENT OF THE PREDICTABILITY OF SEASONAL RAINFALL IN RATNAPURA USING THE SOUTHERN OSCILLATION AND ITS TWO EXTREMES

: An attempt is made to assess the predictability of seasonal rainfall at Ratnapura, namely, first intermonsoonal convectional rains (FIM), second intermonsoonal convectional rains (SIM), south-west monsoon rains (SWM) and north-east monsoon rains (NEM) by using Southern Oscillation and occurrence of El Niiio and La Nifia events. Seasonal rainfall data of Ratnapura and three monthly averaged Southern Oscillation Index (SOI) values for 118 consecutive years were analysed for lhg con-elation coefficients (CCs). The link between seasonal rainfall with El Nifio or La Niiia events was evaluated using binomial analysis. The association between FIM rains and the SO1 slrows a weak positivc relationship. All the other seasonal rains, SWM, SIM and NEM are negatively related to the SOI. However, these relationships are _ weak and therefore cannot be used for any predictive purposes without wide margins of eilurs. The link of FIM rains (March to April) either with El Nifio or La Nifia events was not clear. Nevertheless, it is apparent that there is a strong tendency for above normal SIM rainfall (October to November) during El Nifio years. If a n 4 Nifio year is immediately followed by a La Nifia year, both the FIM and SIM rains would produce above normal rains. It is interesting to note that the occurrence of "drougl~t" conditions duringthe SWM season, May to September, is remote in a La Nifia year. Significant associations between NEM rains (December to Fehmary) and El Nifio or La Niiia events were not observed.


INTRODUCTION
Ratnapura is an important area of the Wet zone of Sri Lanka where agricultural land use accounts for both field and plantation crops. Rainfall plays a vital role in maintaining these crops especially the plantation crops such as tea, rubber and coconut. The characteristic feature of the climate of Ratnapura is the year round rainfall. The annual rainfall ranges from 2800 mm to 5200 mm. I t i s a result of two intermonsoonal convectional rainy seasons; southwest monsoon rainy season and to a lesser extent the northeast monsoon rainy season. Being solely dependent on t h e rainfall, t h e agricultural production of the area i s badly affected by abnormal weather conditions. The late onsets of the four different rainy seasons and the lack of anticipated amount of total rains from them cause widespread crop failures and poor yields. In addition, Ratnapura has gained much attention as one of the few areas of the country where floods are very common. Over the years, hundreds of lives have been lost through flooding and land slides. Rehabilitation and reconstruction has been expensive. Early prediction of rainfall can lead to precautionary actions to mitigate the possible catastrophic consequences.
There has been increasing recognition recently that the most important key to the earth's year to year climatic variability is the El Niiio/Southern Oscillation phenomenon which is oRen referred to jointly as the ENS0 phenomenon. The Southem Oscillation is the see-saw pressure pattern between the Indian-western Pacific (Indonesian low) and central-east Pacific oceans (south Pacific sub-tropical high). On average, pressure is low, relative to the zonal mean, over the Indian-western Pacific oceans and tends t o be high over the central-east Pacific ocean. A simple index, the Southern Oscillation Index (SOI), is often used to study these pressure variations. This index is the difference between normalized monthly mean atmospheric pressures at Darwin (12OS, 131°E), normally low, and Tahiti (1g0S, 150°W), normally high. Extreme anomalies in this pattern involve dislocations of the rainfall distribution in the tropics, bringing drought to some regions and torrential rains to others1. El Niiio events occur during periods when sea surface temperatures (SSTs) are warmer and the trade winds are weaker than normal in the central and eastern Pacific, and SSTs are cooler than normal in the eastern Indian ocean and western Pacific oceans2. The opposite extreme, when the east Pacific is cool and pressure there is higher than the normal, is called anti-ENS0 events or La Niiia episodes. El Niiio event usually starts early in the year and reaches its maximum intensity between the following November and January, returning to normal over the next few months" The SO1 is negative during the ENS0 events and positive during the anti-ENS0 events. ENS0 phenomenon directly affect the climate of a t least half of the planet and in many instances result in heavy loss of life and resources.
In previous studies4-', the association of the ENS0 phenomenon and the spatial and temporal variability of rainfall of the major climatic zones of Sri Lanka have been reported. The strength of the relationship between SO1 and rainfall varies depending on season and location. Therefore, the impact of ENS0 phenomenon need t o be assessed for individual locationsx. However, an in depth investigation of the predictability of rainfall a t Ratnapura using ENS0 phenomenon has not been undertaken despite its usefulness in contingency planning. Hence, this paper investigates the association of seasonal rainfall at Ratnapura with the Southern Oscillation and the occurrence of El Niiio and La Niiia events in the Pacific region.

METHODS AND MATERIALS
-Monthly rainfall data of Ratnapura for 118 consecutive years (1876-1993) were used for the analysis. These data were converted t o four different rainfall seasons before being used in the study: first intermonsoonal convectional rainy season, FIM, (March-April), south-west monsoon rainy season, SWM, (May-September), second intermonsoon rainy season, SIM, (October-November) and north-east monsoon rainy season, NEM, (December-February). The status of each year, either El Nifio or La Niiia, during the period from 1876 to 1993 was identified using published information and was verified against the SO1 data taken from the Bureau of Meteorology, Australia9. Although some of the relationships between the SO1 and climatic fluctuations are not linear, it has been reported that the linearity assumption seems to work well for many areas in the worldlo. Assuming such a linear relationship, the association between the SO1 and the normalized seasonal rainfall of Ratnapura was determined by lag-correlation analysis. The long term mean and the standard deviation used for normalization of the seasonal rainfall were calculated only using rainfall in normal years excluding the years with extreme phases of the SOI, i.e. El Niiio and the La Niiia years.
In analysis with the El Nifio and the La Niiia events, a year has been considered from March of the current year to February of the subsequent year. This period includes the peak months of the El Niiio and La Niiia events during October to December. Binomial probability analysis was used to determine whether the observed deviations of seasonal rainfall during El Nifio and La Nifia years are due t o chance occurrencel1. For each season, the number of above (below) normal years were counted in El Niiio years. Let Xohs be the observed number of above (below) normal rainfall years among the N number of El Nifio years. Then, p = Pr ob {X2Xohsl, the probability of number of above (below) normal rainfall years, is at least as great as that observed during El Niiio years. If this value is less than the significance level, it is unlikely that the observed event is a chance occurrence. A similar analysis was carried out for the La Niiia years. A "Dry" season was defined when the normalised seasonal rainfall is less than -1 whereas seasons with +1 or greater normalised rainfall were considered as "Wet" seasons. In a data set, the interval from -1 to +1 standard deviation contains approximately 68 per cent of the data values. Therefore, data beyond + 1 boundary can be safely interpreted as extreme cases of rainfall, droughts or floods.

Relationship between the Southern Oscillation a n d seasonal rainfall
The lag-correlations were computed to investigate whether the seasonal rainfall anomalies at Ratnapura would indicate a teleconnection to the SOI. Table 1 summarises the principal results of the lag-correlation analysis for four different rainy seasons a t Ratnapura. The three upper case letters that denote the three months period stand for the first letter of each month, i.e. JFM means average SO1 of January through March. The three months periods are always from the current year unless specified otherwise. The mean SO1 of NDJ and DJF periods were calculated using SO1 of November and December of the preceding year along with the SO1 of January and February in the current year.
In general, the correlation between first intermonsoon rains (FIM) and the mean SO1 of preceding three months periods were positive and weak. The highest recorded correlation was 0.14 with the mean SO1 of January through March. The lag-correlations of SO1 with the south-west monsoon rains (SWM) was generally negative except with mean SO1 of immediately preceding March to May period. The recorded highest correlation was -0.17 with the mean SO1 of January to March. Unlike the relationships with the FIM and SWM rains, the association between the rainfall of the second intermonsoonal rains (SIM) and the mean SO1 of the preceding months showed a relatively strong negative correlation ( Table 1). The strongest link of SIM rains was associated to the mean SO1 of May through July period. Approximately equal correlations of SIM rains were also recorded with the mean SO1 of AMJ, JJA, JAS, A S 0 and SON time periods. The north-east monsoonal rains were negatively correlated with the mean SO1 of the preceding months. The highest negative correlation of -0.18 was observed with the mean SO1 of September through November period (SON). The influence of the El Nifio and La Niiia episodes on seasonal rainfall Table 2 shows t h e occurrence of above normal rainfall a t four different rainy seasons during El Njfio and La Niiia years a t Ratnapura. The two integers within the brackets denote the number of "Wet" and "Dry" rainy seasons, the extreme situations, occurred during El Nifio and La Niiia years. Out of 30 El N 5 o events during the study period, 14 events caused above normal FIM rains while 16 years resulted in below normal rainfall. m e r e was a little contrast of rainfall anomalies in the FIM season with La Nifia episodes. Out of 23 La Nifia episodes during the study period, 15 episodes caused above normal rainfall while there were on1 y 8 years with below normal rainfall. Both El Nifio and La Nifia events caused 14 years of above normal rainfall during the SWM season out of 30 and 23 such years, respectively (Table 2). These occurences were not significant a t the 5% probability level.
There was a significant teleconnection between occurrence of El Niiio events and the SIM rains a t Ratnapura. Thirty El Niiio events occurred during the last 118 years causing 22 years of abdve normal SIM rains a t Ratnapura ( Table 2). The binomial analysis showed that the occurrence of above normal rainfall during the SIM season is significant at the 5% probability level. Out of 22 years of above normal SIM rains, five occurences were 'Wet" years. The influence of the La Niiia events on the SIM rains was not significant. There were only 1 1 years with above normal SIM rains out of 23 La Nifia events that occured during the last 118 years.
Out of 12 below normal SIM rainy seasons during La Nifia years, eight years were "Dry" years (Table 2). ' significantly different at the 5% probability level. T l~e two integers within the brackets denote the number of Wet and Dly rainy seasons, the extreme situations, occui~ed during El Nifio and La Nifia years.
The anomalies-during the NEM do not show a clear pattern with either El Niiio or La Niiia episodes. Fifteen El Niiio events were associated with above normal NEM rains resulting in the same number of years with below normal NEM rains. The influence of the La Niiia episodes on the NEM rains was also not evident. Out of 23 La Niiia episodes t h a t occurred during the last 118 years, 10 years resulted in above normal NEM rains while 13 years caused below normal NEM rains a t Ratnapura. None of these anomalies were significant a t t h e 5 % probability level. In addition, i t is worth noting t h a t the possibility of extreme NEM events of either drought or flood conditions, occurring during La Niiia years is quite low ( Table 2).
During the last 118 years, there have been 11 El Nifio events immediately followed by La Nifia events. Such ENS0 events showed a reasonable forecasting possibility of FIM and SWM rains (Table 3). La Niiia events preceded by El Niiio events caused 8 years of above normal FIM rains, out of which three years were in the 'Wet" extreme. However, this link was not significant a t the 5% probability level. The effect of occurrence of El Nifio and La Niiia events in a row was mBre prominent with the SWM rains where 9 years produced above normal SWM rains. The binomial analysis showed that this occurrence is not due to chance. Out of these 9 years, 7 years were in the 'Wet" extreme of SWM rains. Thus, if a La Niiia event is immediately preceded by an El Niiio event, it is more likely to receive above normal rainfall in both FIM and SWM seasons a t Ratnapura. In the case of SWM rains, it would be more likely to experience flood conditions. However, this relationship was not clear with the SIM and NEM rains (Table 3).

No. of above normal years
No. of below normal years Season

DISCUSSION
In general, association between the SO1 and the seasonal rainfall at Ratnapura was weak. All the correlation coefficients were below 0.27 indicating that only a very small portion (less than 7%) of the total variation of seasonal rainfall can be explained by the SOI. When the correlation coefficient of a relationship is less than 0.70, it has little or no predictive value being unable to explain at least 50% of the total variation of the system12. Thus, these findings suggest that the possibility of predicting the seasonal rainfall at Ratnapura by SO1 alone has limited potential. The weak association between seasonal rains at Ratnapura and the SO1 could be attributed to the micro scale convective activity which could mask the influences of large scale circulation changes on rainfall variations. Such small scale convection of the atmosphere is quite common in the Wet zone of Sri Lankati. Nevertheless, it is noteworthy that most of the correlation values, obtained in this analysis, that are relatively high in magnitude were associated with the SIM rains.
The influence of the occurrence of El Niiio or La Nifia episodes on the first interrnonsoon rains (FIM) at Ratnapura was not evident. The binomial analysis showed that there was no evidence t o conclude t h a t either El Nino or La Niiia episodes cause above or below normal rains during the FIM season in Ratnapura and therefore, the observed rainfall anomalies are due to chance. The lack of any significant link between the occurrence of El Niiio or La Nifia events and the FIM rains could be attributed to the fact that both El Niiio or La Niiia events are at their early stages of development in the east and central Pacific oceans when the FIM season is effective over Sri Lanka. The magnitude of the increased (decreased) sea surface temperature over the Pacific ocean with a newly developed El Nifio (La Nifia) event is rather small. Such a small increase (decrease) of sea surface temperature would not be strong enough to influence the general circulation of the atmosphere and thereby the tropical rainfall.
The link between SWM rains and the occurrence of El Niiio or La Nifia events was also not distinct. It has been shown that the relationship between SWM rains and the occurrence of El Nifio or La Niiia events is weak in the Wet zone compared to the Dry zone". However, it was interesting to note that likelihood of having "Wet" SWM rains at Ratnapura is higher if La Niiia events cause above normal SWM rains ( Table 2). Out of 14 La Niiia years which resulted in above normal SWM rains, 12 years produced 'Wet" SWM rains. Nine La Nina events which caused below normal rainfall resulted in only two years with "Dry" SWM rainy season. Thus, the possibility of drought conditions during May to September in La Niiia years is rather low.
The significant relationship between El Niiio events and SIM rains suggests that the increased SIM rains during El Niiio years at Ratnapura are not due t o chance and probably due to changes in the general circulation. It also points towards a clear forecasting ability of five to six months in advance. There is more than 70% of chance of above normal SIM rains during an El Nifio year. Although El Nifio events have caused 8 below normal SIM rainy seasons during the last 118 years none of them were below -1 standard deviation (Table 2). Therefore, it is unlikely that El Nifio would cause drought conditions during SIM seasons at Ratnapura. Generally, rainfall anomalies of SIM rains during El Niiio years are positive over entire Sri Lanka and may range from +10 to + 30 per cent5. Apossible explanation for such positive anomalies is t h a t warm sea surface temperature anomalies during El Nifio years may enhance the convection with cyclonic circulation near Sri Lanka5.
Although the NEM season coincides with the peak period of El Niiio and La Nifia episodes, the rainfall anomalies do not show a clear contrast as in the case of the SIM season. This could be attributed to the f a d that the NEM winds mainly bring rainfall to the windward side (eastern part) of the central highlands and little or no rains to the leeward side where Ratnapura is located.