Monsoon watch

Researchers of PIK, Germany, claim that their novel method, using the Eastern Ghats and north Pakistan as the “tipping elements”, is better in predicting the onset and withdrawal of the Indian summer monsoon than the existing methods.

The monsoon at its peak over Kava village in Palakkad. A file picture.

IN a paper published online on April 20 in the journal Geophysical Research Letters (GRL), researchers from the Potsdam Institute for Climate Impact Research (PIK), Potsdam, Germany, have claimed that with their new method they can predict the onset and withdrawal dates (OD and WD) of the Indian summer monsoon (ISM)—the south-west monsoon (June 1 to September 30)—two weeks and one and a half months earlier respectively than with the existing methods.

It is known that the arrival of the monsoon is characterised by an abrupt transition to a weather system causing a sudden increase in rainfall over Kerala that is sustained over a period. This clearly identifiable feature has been used for more than 100 years by Indian meteorologists to forecast the onset of the monsoon. The historically determined normal date for the onset is June 1, with a statistical margin of a week (±7 days). However, no precise quantitative definition of the onset can be given in terms of physical variables that result in this abrupt transition to heavy rainfall days over Kerala.

While there have been several attempts to predict the OD and the WD (mainly the OD) over different time scales up to two weeks, including the statistical model of the India Meteorological Department (IMD), which it uses operationally about two weeks ahead, the PIK researchers’ claim is that their novel method can do better in terms of the lead time of prediction for both the OD and the WD.

In the paper titled “Tipping elements of the Indian monsoon: Prediction of onset and withdrawal”, Veronika Stolbova and her associates have argued that before the onset, meteorological variables such as temperature (T) and relative humidity (RH) exhibit signs of critical behaviour and the onset is characterised by their crossing their respective threshold values. They claim to have identified the critical transition in these variables in the case of the Indian monsoon on the basis of the time trends of these variables in the past years (1951-2015) in two particular regions on the subcontinent, namely the Eastern Ghats (20° N, 80° E) and north Pakistan (32° N, 72° E).

These two regions are where the researchers found the T and the RH to exhibit the maximum variance in the run-up to the onset, and have called these geographic regions the “tipping elements of the monsoon”.

Tipping elements The researchers have argued that when criticality is achieved in both these regions together, the physical conditions become favourable for the arrival of the monsoon in the central part of the Indian subcontinent. Using this basic concept, forecast of the OD (at the Eastern Ghats) is done by looking at the pre-monsoon data and extrapolating their corresponding trends. The use of the terminology “tipping elements” for the variables that cause an abrupt onset of the monsoon is, however, misleading and is a misnomer. The phrase is generally used for factors causing irreversible changes, particularly in the context of climate change.

The PIK work has drawn a lot of media attention. Interestingly, it has also drawn the attention of several Members of Parliament (nine to be precise) who on May 11 wanted the Minister of Science, Technology and Earth Sciences to answer the (unstarred) question: “Whether the government is in touch with the institute or proposes to make use of this technology and, if so, the details thereof and if not, the reasons thereof?”

Monsoon forecast is crucial for farmers to plan their sowing and crop strategy for the season. But what needs to be remembered is that the monsoon over Kerala, irrespective of whether it is early or delayed, has no bearing on the further progress of the monsoon across the subcontinent, its performance in terms of the total quantum of rainfall for the season, and the rainfall’s spatial and temporal distributions across the country. The operational forecast of the OD by the IMD is for the monsoon over Kerala, and not for any interior part of India. However, a delay in the onset of the monsoon is generally associated with a delay in the onset over the southern States (and up to Mumbai), which could have an impact on the strategy of farmers of the region for the season.

In response to the question raised in the Lok Sabha, Minister for State for Science and Technology Y.S. Chowdary stated that the methodology adopted by the PIK researchers had limitations because it forecast the onset over the Eastern Ghats and withdrawal from north Pakistan only. “As the methodology has not been tested for the monsoon onset over south Kerala, no tangible gain is expected for our operational onset monsoon forecast. However, the IMD issues the monsoon onset forecast based on an indigenously developed model. The forecasts for monsoon onset for the last 10 years have been found correct,” the written reply of the Minister stated. As is clear from the reply, “onset” as defined in the PIK work is not the same as what is widely accepted following its definition by the IMD and how their finding is related to the monsoon over Kerala is not clear.

Monsoon transition In their 2009 paper “Summer monsoon onset over Kerala: New definition and prediction”, D.S. Pai and M. Rajeevan of the National Climate Centre (NCC) at IMD, Pune, said: “The onset of monsoon represents significant transitions in the large-scale atmospheric and oceanic circulation in the Indo-Pacific region. There is no widely accepted definition of this monsoon transition. However, at the surface, the onset of the monsoon is recognised as a rapid substantial and sustained increase in rainfall.”

But it is well known that the onset phase of the monsoon is associated with large-scale changes in the regional and planetary scale circulation features across the Indian monsoon region. The summer monsoon season is a result of the northward migration of the Inter-tropical Convergence Zone (ITCZ) from the equatorial region, which results in the large-scale transition of deep convection from the equatorial to the continental region. As Pai and Rajeevan have pointed out, at the onset date, a deep band of deep convection (low outgoing long-wave radiation, or OLR) is seen passing through the southern tip of India, which manifests as the zone of maximum cloudiness that moves northwards as the season progresses.

During the onset phase, major changes are seen in the atmospheric wind flow at all levels, particularly the vertical extension of the flow of the westerlies up to about four kilometres (600 hPa (hectopascals), or mbar (millibar) pressure level). The RH of the air also increases at least up to an altitude of 5.5 km (500 mbar level). Following the monsoon over Kerala, the vertically integrated moisture transport at individual stations over the peninsula begins to increase sharply with the appropriate time lag after the onset. In fact, in 2003, J. Fasullo and P.J. Webster proposed a hydrological definition of the onset and withdrawal in terms of the vertically integrated moisture transport. They argued that using rainfall over Kerala alone could lead to “false” or “bogus” onsets that were related to propagating intra-seasonal and non-monsoonal atmospheric disturbances, as it happened, for example, in 2002 and 2004.

In order to avoid such pitfalls of “bogus” onsets, following the 2003 work of P.V. Joseph, former IMD Director, in 2006 the IMD adopted a new three-tier operational definition for the monsoon over Kerala in terms of meteorological conditions that include, besides rainfall over Kerala, wind field and OLR, which together signify the beginning of a large-scale monsoon flow over the Indian subcontinent:

“a) Rainfall: If after May 10, 60 per cent of the available 14 stations enlisted, that is, Minicoy, Amini, Thiruvananthapuram, Punalur, Kollam, Alappuzha, Kottayam, Kochi, Thrissur, Kozhikode, Thalassery, Kannur, Kasargode and Mangalore, report rainfall of 2.5 millimetre or more for two consecutive days, the monsoon over Kerala be declared on the second day, provided the following two criteria are also satisfied in concurrence:

b) Wind field: Depth of the westerlies should be maintained up to a height of about 4.2 km (600 mbar level) in the box equator to 10º N latitude and 55º E to 80º E longitude. The zonal wind speed over the area bounded by 5-10º N latitude and 70-80º E latitude should be of the order of 15-20 knots at a height of about 900 m (925 mbar pressure) based on satellite-derived wind data.

c) OLR: INSAT (Indian National Satellite System)-derived OLR value should be below 200 W/m ² in the box defined by 5-10º N and 70-75º E latitude.”

Rajeevan and Pai also developed a statistical model, on the basis of the parameters relating to physical processes, to forecast the OD about two weeks ahead, which the IMD has now adopted, and accordingly issues its forecast by mid May every year. Analysing the variability of the OD (for Kerala), they arrived at a set of predictive signals derived from thermal, convective and circulation patterns evolving over the Asia-Pacific region in association with the monsoon over Kerala.

On the basis of the time series of the OD for the period 1971-2007 (using the 2006 criteria) and data of wind, mean sea level pressure, land surface minimum temperature, rainfall over south India and the OLR over the Asia-Pacific region, they developed a model using the statistical technique of principal component analysis (PCA). The aim was to arrive at a set of predictors that indicate large-scale circulation patterns as well as intra-seasonal oscillations, which, according to them, have a bearing on the monsoon over Kerala. The PCA technique basically identifies the independent variables from the larger variable set, which have least inter-correlation among them, and thus arrives at a minimal set of predictors.

The operational model of the IMD has six predictors thus identified and, using their observed data during the pre-monsoon period, predicts the OD. The six predictors are: i) Minimum temperature over north-west India; ii) Pre-monsoon rainfall peak over the southern peninsula; iii) OLR over the South China Sea; (iv) Lower tropospheric zonal wind over the south-east the Indian Ocean; (v) Upper tropospheric zonal wind over the east equatorial Indian Ocean; and, (vi) OLR over the south-west Pacific region. The OD forecast is declared about a fortnight ahead, with a model error of ± 4 days, which is better than the natural variability of ± 7-8 days.

The PIK researchers have, however, argued that the existing prediction techniques for the OD, such as that used by the IMD, are based on the analysis of rainfall time series, “which are often poorly measured and modelled…. We use near-surface T and RH, which are well represented in both measurements and models and show clear indicators of the upcoming monsoon.”

They use the daily values of near-surface T and RH (at the height of about 110 m) and wind fields at about 3 km from “re-analysis” gridded datasets of the European Centre for Medium-Range Weather Forecasts (ECMWF), United Kingdom, and the National Centre for Environmental Prediction and the National Centre for Atmospheric Research (NCEP/NCAR), United States, for the period 1958-2001 and 1951-2015 respectively. With a spatial resolution of 2.5°, for the monsoon region (6.25°-97.5° E, 5.0°–40° N) this yields a grid of 225 grid points. The dataset for the OD and the WD for comparison of their predictions have been taken from the IMD (for the two identified critical geographic regions of the Eastern Ghats and north Pakistan respectively).

When a time series analysis of the T and RH data is done in all the grid points, the researchers detect the greatest amount of variance in the Eastern Ghats and north Pakistan, which they call the “tipping elements” (Fig. 1). They also find that there is an important relation between the two regions: the intersection of the average time series in the regions occurs twice (Fig. 2), which, they claim, coincide with mean values of the OD and the WD, as declared by the IMD for the Eastern Ghats within a window of ±5 days.

For prediction, they use the 14-year-data before the year of prediction as the training period. They find that the OD coincides with the data when the T in the Eastern Ghats and north Pakistan become equal. That is, for both the T and the RH, the training period suggests that, close to criticality, the trends in the two regions are opposite, thus leading to cross-overs, once for the OD and once for the WD.

Therefore, forecasting the OD amounts to forecasting when the temperature in the Eastern Ghats will abruptly decrease and coincide with the temperature in north Pakistan. If you use the RH, the RH in the Eastern Ghats has to increase to meet the decreasing RH in north Pakistan. Observed trends in the T and the RH in the pre-monsoon period in both the critical regions, as well as the trends discerned from the training period, are judiciously used for predicting the OD. For prediction of the WD, the Eastern Ghats is not used but the scheme relies entirely on the symmetry of T changes in north Pakistan during the year.

In terms of performance of the PIK prediction scheme, it is regarded successful if the time difference between the predicted and the real OD is less than a week (≤ 7 days) and for the WD it is taken as ≤ 10 days. (The authors, however, emphasise that successful prediction is not the same as accuracy of prediction, which is ±4 days just as in the case of the IMD model.) Prediction success (for the 50-year period from 1965 to 2015) is 73 per cent if based on trends in the T and 63 per cent if based on trends in the RH, when the forecast is made on the 125th day of the year, which is May 5. The lower success rate using the RH, according to the authors, is because of high variability in the RH and associated difficulties in approximating trends (Fig. 3).

The prediction scheme for the WD succeeds in 84 per cent of the years when the forecast is made on the 205th day of the year (July 25) with ±5.6 days accuracy (Fig.3). The IMD, however, does not predict the WD but declares approach to withdrawal on or later than September 1. “We don’t have a model to predict withdrawal. We only say that the monsoon is withdrawing from a certain area. Over north-west India we see the rainfall and moisture. Rainfall should come down for a week or so and moisture should also drop significantly. In upper air, a pressure high or anti-cyclone forms, which are indicators of withdrawal,” Rajeevan said.

Prediction of the WD is useful in only as far as knowing the extent of the monsoon season itself for water storage and hydropower generation purposes. In his reply to the question in the Lok Sabha, the Minister in fact pointed out that the PIK had demonstrated poor skill in predicting the WD between 2010 and 2014.

“The notion of criticality may be a good concept but 73 per cent is a poor success rate for the OD especially when you have such a wide window to define success, which is as much as natural variability,” said Rajeevan, who is now Secretary, Ministry of Earth Sciences (MoES). “Once the monsoon begins, moisture builds up everywhere. We have not noticed such threshold effects in the interior. We see the abrupt change only over the Kerala coast. In the interior, transitions are slow and gradual,” he said.

“Our forecast models are based on physical processes, such as the building up of the westerlies, the building up of moisture, and the building up of convection,” he said. In the past 11 years, the IMD’s prediction of the OD has been accurate on 10 occasions (except for 2015) (Table 1), which is indicative of good prediction skill of the model. This year, the IMD has forecast the OD to be delayed by a week, which means the monsoon over Kerala would be around June 8 (±4 days). Joseph, too, has predicted a delayed onset on June 15 on the basis of historical data pertaining to the ODs following the El Nino years.

El Nino effect Even though El Nino has nearly ended over the Pacific, there is a lingering effect of it over the Indian Ocean, rendering the region still very warm, which is delaying the northward migration of the convection zone over the subcontinent, Rajeevan said. “Before the monsoon comes, one spell of rain comes, which does not move beyond 12° N or so. That spell, this year, has come late. But there is a lot of convection over the Indian Ocean itself. So a depression or cyclone will form and the next spell has to come. In fact, the whole seasonal transition is late,” Rajeevan said.

As indicated in their paper, the 125th day of the year, or May 5, forecast of the OD (at the Eastern Ghats) for 2016 has been stated as June 13 (Fig.4). Within their window of success of seven days, if the monsoon reaches the Eastern Ghats by June 20, the method would be called successful! Now, the normal OD, according to the IMD, is June 10 when the monsoon over Kerala is on June 1, meaning a lag of about nine days. With a lag of nine days, a delayed monsoon over Kerala around June 7-8 would still not be able to tell if the PIK technique has good skill or not.

Be that as it may, the very concept of threshold effect or criticality conditions attained by some meteorological variables being the causative mechanism of monsoon onset over the subcontinent may itself be of interest scientifically. “Though we have not done such an analysis of variance in the gridded data of T and RH across the country, with so many changes and large-scale circulation accompanying the northward migration of the monsoon regime, it is quite possible that such sharp changes in some variables occur, because, after all the onset over Kerala is an abrupt transition,” Pai said.

But could there be a causal connection between such critical transitions and the physical processes driving the monsoon? “With surface air temperature and moisture time series at only two locations as inputs in their method, it would be difficult to formulate a verifiable physical mechanism giving cause-and-effect relations,” Joseph pointed out in an email correspondence. “A thick population of deep cumulonimbus [thunder] clouds giving large amounts of rainfall is a characteristic of the monsoon. For the formation of such clouds, low-level air temperature and humidity should be high over the large monsoon area,” he added.

Raghu Murtgudde of the University of Maryland, U.S., told the well-known environmental journalist Darryl D’Monte for the online Indiaclimatedialogue.net: “The best use of these results would be to understand why these tipping points are able to better capture the onset and withdrawal [if they indeed do] and how they can be translated into dynamical understanding,… what it actually means for the precipitation process, which brings us back right back to having to getting the rainfall right in the model anyway.”