COVER STORY

The forecasting failure

The behaviour of the monsoon this year has been so peculiar that not one of the groups around the world that do monsoon forecasting, including the India Meteorological Department, got it right.

R. RAMACHANDRAN

MIDWAY through the four-month period of the monsoon season (June 1-September 30), despite the apparent revival of the monsoon in some areas in early August, it is clear that the country is heading for a major drought. The India Meteorological Department's (IMD) long range forecast (LRF) in May - that the country will have a "normal" monsoon for the 14th year running - has proved to be an embarrassing failure.

While there are serious methodological flaws in the IMD's 16-parameter statistical LRF model (Frontline July 21, 2000 and November 9, 2001), the behaviour of the monsoon this time has been so peculiar that not one of the groups that carry out monsoon forecasts around the world, using diverse numerical and physical models, got it right. "What one can do at this point of time is only a diagnosis based on the symptoms of a failed monsoon. What has caused this is not easy to understand because this has happened despite favourable conditions of atmospheric forcings that are normally linked to the monsoon," points out D.R. Sikka, former Director of the Indian Institute of Tropical Meteorology (IITM) in Pune.

According to the mid-season review released by the IMD on August 5, the cumulative rainfall across the country from June 1 to July 31 presents a grim situation. Of the 36 meteorological sub-divisions, the rainfall was scanty (-60 per cent or less of the sub-divisional long-term average, LTA) in five sub-divisions, deficient (-20 per cent to -59 per cent of the sub-divisional LTA) in 21 sub-divisions, normal (+19 per cent to -19 per cent of the sub-divisional LTA) and excess (+20 per cent or more of the sub-divisional LTA) in one sub-division.

The rainfall deficit for the whole country in the first half of the monsoon is a high 30 per cent; in July alone it is 49 per cent. Against a mid-season LTA of 47 cm (Table 1), the country has received only about 33 cm of rainfall. Considering that the LTA for the entire season is 88 cm, there should be about 55 cm of precipitation in the second half, as against an LTA of 41 cm, to make up the deficit. That would mean a highly unlikely rainfall quantum of about 134 per cent in July-September.

Is this situation unprecedented and abnormal? In the past 130 years, only four times - 1877, 1918, 1972, and 1987 - has the rainfall in June-July been less than this year's. There have been 21 drought years (less than 90 per cent of LTA), 92 normal rainfall years and 18 excess (more than 110 per cent of the LTA) rainfall years.

Data also show that the frequency of drought years has varied on a decadal scale. While there were 10 droughts during the period 1965-87, the last 13 years have had a string of normal monsoons, a situation seen only twice before in the last 130 years - during 1878-90 and 1921-32.

According to Sulochana Gadgil of the Centre for Atmospheric and Ocean Sciences (CAOS), Indian Institute of Science (IISc), Bangalore, this year's failed monsoon should not, therefore, be viewed as an unprecedented catastrophe. "It is a part of the natural variability of the monsoon system, very close to the lower limit of the observed variation in mid-season cumulative rainfall, just as the 13 years of normal monsoon too is," she said.

While there has been a recent spurt in monsoon activity with appropriate atmospheric systems forming, past data suggest that when the rainfall in the first half of the season tended to be low, it was so in the second half as well. In fact, the four years mentioned above turned out to be major drought years.

According to Sulochana Gadgil, analysis of the temporal variation of rainfall shows that if the deficit in June-July is more than 10 per cent, the probability of rainfall becoming normal is 0.33 and of it being a drought year is 0.67. Past data also show that there has hardly ever been excess rainfall during August-September; it is mostly normal or slightly below normal. In any case, the agricultural drought that has already set in cannot be reversed, notwithstanding the possibility of enhanced rice transplantation activity and a sustained revival of the monsoon.

Where did the IMD go wrong in its long range prediction? Actually, from its perspective the IMD did nothing wrong. It did exactly what it has been doing for the last 13 years. The fact that as many as nine times this exercise went off the mark in quantitative terms did not attract sufficient attention because the monsoon was normal in the gross and the sub-divisional scale dry spells and localised droughts did not have much of an impact on the economy. But this time the monsoon failed the IMD, and that too massively.

To expect the model to do well every time is unrealistic. Besides its inherent flaws, it is a statistical model and, by the very nature of its construction, should be expected to predict only the average behaviour correctly. Since the monsoon is normal about 70 per cent of the time, any reasonable statistical model is likely to do well most of the time. "The test of a forecast model is in the weather extremes - drought or floods. Statistical models are unlikely to predict such extreme events," points out J. Srinivasan of CAOS. This year has clearly been one such extreme event. All the tinkering with parameters that the IMD has been doing for the last few years is only to do this average job somewhat better.

The IMD, based on the 16-parameter Power Regression Model it has used since 1988 for operational forecasts, predicted the total seasonal rainfall this year to be 101 per cent of the LTA of 88 cm. (Monsoon for the whole country is said to be "normal" if it is within a 10 per cent window around the LTA. The model is said to have an inherent plus or minus 4 per cent error.) The 16 parameters include regional and global scale temperature, wind, pressure and snow related meteorological variables (Table 2). The total seasonal rainfall is related to these 16 parameters through an empirically derived equation based on historical monsoon data. The seasonal rainfall for the three "homogeneous rainfall regions" of the country - the northwestern region, the peninsular region and the northeastern region - were predicted to be 104 per cent, 99 per cent and 100 per cent respectively of the regional LTA. The regional forecasts use three different subsets of the 16 parameter set and have an inherent error window of 8 per cent.

Of the 16 parameters, 11 (69 per cent) were said to be favourable for a normal monsoon. According to the authors of the model, historical monsoon data suggest that whenever more than 60 per cent of the parameters are favourable, it results in a normal monsoon. This year, the five unfavourable parameters were: El Nino (the same year), El Nino (the previous year), Southern Oscillation Index (March to May), Indian Ocean Equatorial Pressure (January to May) and Eurasian Snow Cover (December). Apart from the Indian Ocean Equatorial Pressure, the other four are related to the El Nino phenomenon - the anomalous warming of the Pacific waters - which is believed to affect adversely the monsoon, among its various impacts in different parts of the world.

However, monsoon data over the years seem to suggest that there is no one-to-one correspondence between the amount of rainfall and El Nino. This was most dramatically seen in 1997, which recorded the strongest El Nino of the 20th century as well as an especially good monsoon. Indeed, it is to the credit of the IMD that it had predicted a normal monsoon in contrast to what was being generally predicted by other monsoon forecasting exercises around the world.

A recent investigation into the weakening influence of El Nino over the Indian monsoon showed that higher surface temperatures over Eurasia during the preceding winter seemed to offset the adverse impact of El Nino. Conversely, increased Eurasian snow cover - implying unfavourable lower surface temperatures - would not be able to counter El Nino effectively. Equatorial pressure could also be linked to El Nino. A study by scientists of the IITM some years ago suggested that higher equatorial sea surface pressure is often a precursor to El Nino.

In the wake of a highly deficit rainfall, some misplaced arguments have been made about the possible causes. R.K. Pachauri, director of the Tata Energy Institute (TERI), New Delhi, and Chairman of the Intergovernmental Panel on Climate Change (IPCC), reportedly attributed the failed monsoon to global warming. A scientist involved in the recently released Third Assessment Report of the IPCC reportedly ascribed the below normal rainfall to the increased aerosol intensity over Afghanistan following the U.S. war against that country. Both these arguments are dismissed easily.

"An isolated drought cannot be attributed to long-term changes such as global warming," said L.S. Rathore of the National Centre for Medium Range Weather Forecasting. "Climatic changes would result in gradual changes in the quantum of seasonal rainfall and in the frequency of droughts and floods. Moreover, all climate-change models have predicted increased monsoon rainfall," he pointed out. As for the aerosol argument, Rathore said that the monsoon system involved the release of enormous amounts of energy and would not be affected by localised changes in aerosol density. Further, as Gadgil pointed out, aerosols present in winter were washed away by the first monsoon rain and model simulation of soot particles in the atmosphere resulted in increased heating of the atmosphere and enhanced rainfall.

The important thing about this year's monsoon is that it failed every scientific exercise at prediction undertaken by different meteorological research groups around the world. These included not just statistical models but dynamical and numerical (with specified initial conditions from observed data) models as well.

The three-parameter model of the IITM, which seems to reproduce past data as well as the IMD's 16-parameter model, predicted a 110 per cent rainfall. In fact, the several empirical models developed by the IITM over the years produced a consensus forecast of 105 per cent. Similarly, the Space Application Centre, Ahmedabad, which has developed an LRF model that uses satellite-derived global data, made a forecast of 104 per cent. The neural network-based model of the Centre for Mathematical Modelling and Computation, a unit of the Council of Scientific and Industrial Research, forecast 99 per cent rainfall.

More interesting are the predictions of the numerical models and the General Circulation Models (GCMs) of pre-eminent international institutions such as the European Centre for Medium Range Forecasts (ECMWF) and the International Research Institute for Climate Prediction (IRI). The ECMWF, using its sophisticated numerical model with initial conditions in May, forecast near-normal rainfall over India from June to August except for some shortfall in the southwestern region. The IRI's May forecast (based on an ensemble of GCMs) had predicted a normal rainfall all over the country.

However, the forecasts based on initial conditions in June altered these scenarios. Predictions for July-August-September showed major rainfall deficits over India, particularly in the northwestern region. This is perhaps indicative of the development of some meteorological systems in June, which were absent in May. "This could be due to some unexpected large-scale changes in the atmospheric circulation in May, which the initial conditions in June picked up. What exactly they are and what caused them needs a full-scale analysis of the June inputs of the ECMWF and IRI models," said Sulochana Gadgil.

Empirical models such as the IMD model, on the other hand, use predictors derived from past data which fail to predict an unusual event such as the failure of this year's monsoon. "The number of extreme events has not been more than 15 during the last 130 years. Hence it is not easy to develop statistical methods based on a small sample of extreme years. Since we have had normal or above normal rainfall for the past 14 years, it may have led to the placing of greater faith on statistical models than one should really have," said Srinivasan. While the atmospheric variables that historical data show as being conducive to a good monsoon were generally favourable in May, appropriate systems that drive the monsoon apparently failed to develop.

As Rathore pointed out, while the cross-equatorial flow, which normally brings in moisture-carrying winds from the Indian Ocean over the southwestern peninsula, was strong enough, its direction seemed to have been skewed towards South-East Asia and the Pacific. The force that drives the cross-equatorial winds, known as the Mascarenes High over Madagascar, seems to have been positioned farther from the normal position, according to him. The favourable warm sea surface temperatures (SSTs) over the southern Indian Ocean (parameter 3) in February-March seem to have persisted well into June-July causing much of the precipitation to occur close to the Equator itself. Also, the pressure gradients along the West Indian coast that accentuate the monsoon wind flow over land and lead to rains over the Western Ghats and the Arabian sea, did not develop.

Similarly, even though conditions in the mid troposphere (at a height of 5 to 6 km) were favourable early on, appropriate wind regimes did not form during the monsoon period. The Tibetan High, a dominant feature of the monsoon related to mid-tropospheric flows, seems to have been at an anomalous position. With the monsoon trough - the zone of low pressure that extends from the east coast in the northwestern direction facilitating the flow of monsoon winds back over the Gangetic plains from the Bay of Bengal - anchored at the foothills of the Himalayas, no easterly jets were seen till late July.

Usually the cyclonic storms over the Pacific traverse across South-East Asia to form depressions over the Bay of Bengal, which serve as cloud-bearing systems as they weaken over land and move along the monsoon trough. Strangely, this year, while the frequency of typhoons in the northern Pacific has been higher than normal, these have not resulted in depressions over the Bay of Bengal.

SO despite a better understanding of global-scale meteorological processes, sophisticated computational tools and the ever-increasing number crunching power of computers, do we understand the monsoon at all in all its detail? What is the cause and what is the effect? The latter question is pertinent in the context of this year's monsoon and El Nino. Contrary to predictions early this year of a strong El Nino, it has been weak to moderate until now. But a weakened monsoon seems to be resulting in El Nino becoming stronger. "This is not surprising," says Sikka. "Such a response of the Pacific to the monsoon has been seen in the past as well."

Indeed, even this strengthening El Nino seems to have a strange pattern. For one, it is accompanied by anomalous warm waters over the entire region. Towards the east, it extends from the southern Indian Ocean up to the central Pacific where the warming is over a slender region and not the usual large region characteristic of El Nino. Towards the north, both the Bay of Bengal and the Arabian Sea have warm SSTs. More interestingly, the Peruvian coast is still cool and non-El Nino-like. The classic El Nino phenomenon first warms up the Peruvian coastal region (known as Nino 1+2), caused by the upwelling of warm waters from below, which then spread westward to the Nino 3 and Nino 4 regions in the central Pacific. This time it is as if warm waters have been sprayed on the central Pacific over a thin region extending right up to the Indian Ocean towards the west.

The 16-parameter model uses the temperature anomaly in the 1+2 region as the predictor, though El Nino researchers believe that the 3+4 region should have greater influence. Indeed, an analysis by V. Thapliyal (the chief author of the 16-parameter model) and M. Rajeevan of the IMD, found that the El Nino tendency over 3+4 was better correlated to the Indian monsoon than that over 1+2. However, this change in the parameter has not yet been made in the operational model. Of course, even if this warming of the central Pacific is sustained, it is likely to peak only after the monsoon is over. So while this may not have a direct link to the current monsoon, the same atmospheric circulation that has caused this strange and delayed warming of the Pacific has also perhaps caused a weakened monsoon over the Indian land mass.

This year's monsoon has certainly thrown up a whole gamut of questions. Krishna Kumar of the IITM believes that it is time for a re-look at all the predictors because of the increasing evidence of their weakening, and sometimes even negative, correlation with seasonal rainfall. Sulochana Gadgil, on the other hand, believes that while these need to be closely monitored, a single instance such as this does not necessarily mean that in future these models will fail. What one needs, perhaps, is a mid-course correction based on the observations in June. The IMD, which has the unenviable task of providing an operational forecast with its associated political and economic implications, is unlikely to embark on such a continuing exercise through the monsoon. But the monsoon of 2002 has brought home the complex interplay of planetary-scale atmospheric circulation that scientists are far from being able to understand.