THE most remarkable thing about the recent cyclone, Phailin, which struck Gopalpur in Odisha on October 12 and caused widespread destruction in the coastal areas of Odisha and Andhra Pradesh, was the minimal loss of human lives. At last count it was just 38. This was the result of the successful evacuation of nearly a million people well before Phailin struck the coast. Unlike the situation during the Odisha super-cyclone of October 1999, which left 10,000 or more dead, the planned evacuation on a massive scale this time was possible thanks to the advanced and accurate forecast of the cyclone intensity, landfall location, time of landfall and the post-landfall adverse weather conditions by the India Meteorological Department (IMD).
The IMD had predicted the cyclonic system to become a very severe cyclonic storm (VSCS) when it was intensifying into a depression over the North Andaman Sea three days before it struck even as weather agencies in the United States estimated it to be the more devastating super-cyclone (Table 1). In the case of the 1999 super-cyclone the lead time available was only 24 hours. The earlier models for track and intensity prediction were not good and the forecast confidence was also less. With better models, improved observation systems and greater forecast skills, IMD scientists predicted not only the cyclone intensity accurately but also the landfall at Gopalpur even as individual model predictions of the location varied from Visakhapatnam to Balasore.
M. Mohapatra, scientist at the Cyclone Warning Division (CWD) of the IMD, said one model predicted that the landfall would be between Chandbali and Balasore and another predicted that it would be near Visakhapatnam. He added that the Japan Meteorological Agency (JMA) said it would be Visakhapatnam; the IIT Bhubaneshwar model predicted Balasore; the U.K. Meteorological Office (UKMO) said it would go near north of Gopalpur; the Global Forecast System (GFS) from the U.S. predicted Paradip; the European Centre for Medium-Range Weather Forecasting (ECMWF) said it would cross between Kalingapatnam in Andhra Pradesh and south of Gopalpur. “So there are a lot of variations. It is not that models are done and forecasters can predict. You need human expertise,” said Mohapatra.
“At the end of the day,” said L.S. Rathore, Director-General of the IMD, “as a forecaster you have to take a call. Take the intensity forecast, for example. Western agencies were intensifying it. For a forecaster, it is easy to be swept away, particularly if it makes him safer. By over-prediction the forecaster is always safe, both in terms of intensity and landfall.” On October 8, the IMD forecast that the cyclone was headed towards north Andhra Pradesh and Odisha. On October 9, the forecast zeroed in on Gopalpur. “This time we had a lot of internal debate,” Rathore said, “because at a given point of time the system does not move straight, as it has its own energy and dynamics, though finally of course the track appears straight. So when it seems to move a little away from a straightforward track, your heartbeat increases. You really need to have guts and stick to what you had predicted. And so it was on the 12th evening, the landfall was right at Gopalpur, and with the wind speed only about 210 km/hr, it was not a super-cyclone either.”
Practically zero error What was the final error in the landfall prediction? “To be honest,” said Rathore, “if I say it was zero, you would laugh at it. We had internal discussions and ultimately arrived at an error of 3 km, which is practically zero.” As against the average landfall error of 50-100 km in 24-hour forecasts in recent years, this is indeed a phenomenal achievement. According to him, with the confidence in the IMD’s prediction, the inter-agency coordination meetings decided on evacuation in five districts, with the largest number of evacuees being from Ganjam because maximum damage was expected there. As the cost of evacuation per km is huge, the accuracy of prediction of the strike location helped minimise the costs as well, he pointed out. Thus, the IMD predictions were vindicated ultimately and the low death toll would suggest that the preparedness measures taken by the disaster management agencies, particularly the Odisha State Disaster Management Authority (OSDMA), proved adequate and effective.
“This is not a point success or an isolated success,” emphasised Mohapatra. “It is a gradual success.” Indeed, the predictions about the Thane cyclone of December 2011, which was also a cyclone of the same category, a VSCS, with wind speeds of 120-140 km/hr, and which struck the Tamil Nadu coast between Cuddalore and Puducherry, the track and landfall predictions were equally accurate. The error in the 24-hour landfall prediction was 20 km. Here, too, timely and accurate predictions greatly helped minimise the number of lives lost, which was only 46. So were the predictions about Mahasen, the cyclonic storm that struck Chittagong in Bangladesh in May 2013.
However, neither Thane nor Mahasen attracted the media and public attention that Phailin did. This was mainly because of the greater public awareness and media focus created by the IMD’s outreach efforts, with regular media briefings and press conferences for three days before the event, and by the warning by Western agencies that the IMD was underestimating the cyclone’s severity and that it was likely to be a potentially far more devastating super-cyclone.
Satellites ushered in a revolution in cyclone monitoring and characterisation in 1960 when the U.S. launched TIROS, the first polar orbiting weather satellite. The IMD did make use of TIROS images, though not in real time. Some images and products could be accessed under some bilateral arrangement. Meteorological observations with geostationary satellites began in 1974 and products from these were also made available to India. According to Mohapatra, the U.S. established India’s first satellite picture receiving station in Mumbai for an Indo-U.S. satellite experiment, which was followed by more stations in other places in the country.
INSAT boost The launch of the INSAT system with INSAT-1B in 1983 brought in a paradigm change in the IMD’s cyclone monitoring efforts. INSAT images began to be received every half hour, since the on-board camera took about 23 minutes to complete one scan, and India did not have to depend on other countries any more. Satellite-based weather monitoring got a further boost with the launch, in 2002, of Kalpana, a dedicated meteorological satellite. It was used for monitoring Phailin as well (Figures 1a and b). The criticality of satellite imagery in monitoring cyclones in the region, and the associated track, intensity and landfall forecast was painfully brought home during the super-cyclone when the very high resolution radiometer (VHRR) aboard INSAT-2E failed and the read-out software for CCD images was still not in place at the IMD, resulting in poor track and landfall forecast beyond 24 hours ( Frontline , November 26, 1999).
A major change in the cyclone observational systems came with the installation of Cyclone Detection Radars (CDRs) in the 1970s. Ten CDRs were installed, which covered the entire coast. Only these analog-type CDRs were in use until 2002 when the first Doppler Weather Radar (DWR) was installed. Today there are about 22 S-band DWRs in the radar network across the country, but still there are only five on the coast (Figure 2).
The DWRs are more useful because they provide data on wind distribution in the cyclone, which the analog CDRs cannot give. Once there is information on wind speed, a quantitative assessment of the cyclone intensity becomes possible. Since CDRs do not give wind speeds, they enable only a qualitative assessment of the intensity. The DWRs, which have a range of 400-500 km, help in not only monitoring the location, the intensity of the system and wind speed, but also provide some features that help the meteorologist know about the likely movement of the cyclone.
As an aside it may be noted that there is no DWR on the Odisha coast, though that is not so critical as long as there is at least one available in the range for the cyclone to be tracked and monitored as it approaches the coast. The DWR data for Phailin actually came from the DWR at Visakhapatnam.
Network of buoys One of the important improvements in the instrumentation for measuring cyclone parameters, in particular sea surface temperature (SST), is the use of a network of buoys, the establishment of which began in 1997 under the National Data Buoy Programme. Buoys were not available for the 1999 super-cyclone, for instance. As on date there are 13 moored buoys and one drifting buoy (which give atmospheric data on air pressure, air temperature, humidity and wind speed and direction, and ocean-related data on current speed, current direction, water temperature, SST, salinity and wave height, direction and period). The network is maintained by the National Institute of Ocean Technology (NIOT), Chennai, and data collection and analysis are carried out at the Indian National Centre for Ocean Information Sciences (INCOIS), Hyderabad. Together with buoys placed for some international programmes, there are in all about 25 in the network.
Introduction of Automated Weather Stations (AWSs) for ground observations was another important development during the 1990s. After the IMD modernisation phase from 2009 to 2012, their number has increased greatly and now there are 675 of these land-based stations across the country. “Whatever data we are collecting, it is a limited data set,” pointed out Rathore. “When there is a cyclone, because there is a cloud mass it will prevent taking good data. So very good data from the point of view of models is not there; it reduces, in fact. A single buoy close to the system can give immense information. You can then adjust other information accordingly because it is the most reliable data you will have. As regards satellite data, Kalpana does not have sounders. INSAT-3D will have, which will give temperature and humidity profiles as well as SST, wind, cloud top temperature, etc. Then we will be able to simulate the atmosphere better. Cloud top temperature is a very useful thing; the temperature at the eye of the cyclone can also tell us how intense the cyclone is.”
Satellite-based ocean data also form part of the data collection for forecast purposes. Satellites are used to provide cyclone wind data beyond the DWR range of 400-500 km. After the Indian Space Research Organisation (ISRO) launched Oceansat-2 in 2009, this capability has been enhanced greatly.
Oceansat provides sea surface wind and ocean surface data based on the principle of scatterometry, which involves measuring the backscattered or reflected component of the microwave beam sent from the satellite. However, Oceansat is a polar orbiting satellite and, therefore, scatterometric data can be obtained only once or twice a day, and it need not necessarily cover the cyclone area during a pass, pointed out Mohapatra. Two other foreign satellites, WINDSAT and ASCA, also provide scatterometric data. In addition, the IMD also obtains data on SSTs, ocean thermal energy, sea surface height anomalies and sea condition from other satellites, in particular the open access satellites of the National Oceanic and Atmospheric Administration (NOAA) of the U.S.
‘Genesis’ and after Given the above sources of observational data, how does the operational system for a cyclone forecast function? Under the IMD, there is a National Weather Forecasting Centre (NWFC) and the CWD. “The CWD,” said Mohapatra, “comes into the picture only when a low pressure area develops in the region. If it forms over the North Indian Ocean (NIO) region, we become more watchful.”
In the case of Phailin, the IMD’s monitoring of the NIO showed on October 3 that a low pressure system was likely to form over the North Andaman Sea around October 6. Regular monitoring on the basis of data from the IMD’s island stations and some observations from Thailand showed that a low pressure area had indeed developed.
On October 7, there were signs of intensification, and the IMD’s all-India weather forecast bulletin forecast that the system would concentrate into a depression on October 8 and intensify into a cyclonic storm by October 9, which will head towards north Andhra Pradesh and Odisha. Depression is when the wind speed is at least 17 knots (~32 km/hr; 1 kt = 1.85 km/hr). “Wind distribution,” said Mohapatra, “is determined from Oceansat, available buoys and sea and coastal observations.” He added: “On October 8, the system did become a depression and that is what we call ‘genesis’. Then from visible, IR [infrared] and enhanced IR satellite imageries we estimate the intensity based on the Dvorak’s T-scale (Table 1).”
Once the genesis is determined, it is the job of the forecaster to decide on the “best landfall location” and “best intensity”. There is a standard operational procedure (SOP) laid out by the IMD to arrive at these decisions. The SOP is updated regularly on the basis of experience; the last update was, in fact, as recently as in July 2013. The SOP has a checklist for what information is available and what is not. The forecaster takes all observations—satellite, radar and coastal—and modulates them using scatterometric data, buoy data or ship data if they are available and arrives at the best track of the system. All this is done every three hours. There is a decision support system to aid the forecaster in making his decisions, which includes a consensus approach for critical predictions.
On October 9, the IMD had forecast that the cyclonic storm would cross between Kalingapatnam and Paradip by October 12 night as a VSCS. On October 10, the track prediction was narrowed down and it was forecast that the VSCS would cross close to Gopalpur, with a sustained speed of 175-185 km/hr and a storm surge of 1.5-2.0 m. On October 11, the day before the strike, the IMD forecast that the VSCS with a higher sustained speed of 210-220 km/hr would cross the coast with an accompanying storm surge of 3-3.5 m, and the landfall location was maintained as Gopalpur.
One set of forecasts can be made only 24 hours before the event on the basis of synoptic observations, satellite observations and sea observations. On the basis of observations alone a forecast cannot be given earlier than 24 hours though qualitatively one can say what would be the likely wind, pressure and location. For forecast beyond 24 hrs, numerical weather prediction (NWP) models are used. The NWP models are of various kinds, global and regional. There are various global models in use around the world.
Forecasting models The IMD uses the global forecast system (GFS) adapted from the National Centre for Environmental Prediction (NCEP) in the U.S. It is being run at the IMD and at the National Centre for Medium Range Weather Forecasting (NCMRWF). The difference between the original NCEP-GFS and the IMD-GFS is the addition of local data. Global models from other countries provided under bilateral arrangements, such as the JMA model; the Unified Model of the UKMO, which is running at the NCMRWF; and a model from Meteo-France are also used. These models, according to Mohapatra, are available readily on a digital platform. “We also used to purchase some open access algorithms and image products from the ECMWF but now we do not. It is very costly,” he said
Then there are regional models. The IMD uses the Weather Research and Forecasting (WRF) model, which is a mesoscale model developed in the U.S. and is in the public domain. Mesoscale model parametrises systems at scales of a few kilometres to hundreds of kilometres. (A cyclone can be 100-1,000 km with eye size being 10-55 km. Phailin was 500-1,000 km and its eye size was 15-20 km.)
There is also a cyclone-specific regional model called Hurricane WRF (HWRF), which is an atmosphere-ocean coupled model also adapted from the U.S. For a cyclone, ocean conditions are very important and this model parametrises both the atmosphere and the ocean conditions. However, according to Mohapatra, at present only the atmospheric component of this model is being run and the ocean component is expected to be given to INCOIS in the next two years. In the case of Phailin, therefore, the oceanic component of HWRF was not used for prediction purposes. So the HWRF prediction should improve when the ocean component is also incorporated into the model. The WRF and HWRF models can be run at two spatial resolutions, 27 km and 9 km. In the HWRF, the cyclone’s periphery is parametrised in a 27-km grid and the inner domain is parametrised in a 9-km grid. The resolution in global models is 25 km.
“The GFS is a model with complete data assimilation, which we run routinely twice a day for a seven-day forecast,” said S.K. Roy Bhowmik, Deputy Director-General in charge of the Numerical Weather Prediction (NWP) division in the IMD. “The GFS gives all the parameters like rainfall, wind field and wind circulation and gives an idea in the medium-range time scale if a circulation is forming or not.” This model, he said, gave an indication seven days ahead that this system from the Pacific was developing into a low pressure area. The WRF model gives a three-day forecast and also gives rainfall and circulation data.
The “moving nest” All the models except the HWRF are run (on the IMD supercomputer IBM P6) twice a day, at 0 UTC (coordinated universal time) and 12 UTC, for routine weather forecast. The HWRF is run only when there is a cyclonic system developing. “This model has a provision for vortex relocation,” said Bhowmik. “When we run the model, from the synoptic observational data of the location of the cyclone centre, the initialisation data will accordingly be modified to the new position of the centre. The model integrates over five hours for every initialisation. The parametrised inner domain thus moves along with the cyclone. It is, therefore, called ‘the moving nest’.” The HWRF gives a five-day forecast of track, intensity, rainfall distribution and surface wind distribution.
“If we have five models,” said Bhowmik, “they will give five different results, which will confuse the forecaster. But with some in-house developments we have been able to give products that a forecaster can readily use. We have developed a multi-model ensemble (MME) technique. So if I have five models, the ensemble is constructed with weights assigned to each based on their previous performance. This is the main strength of our cyclone forecasting system.”
The MME method is an objective way of minimising the uncertainty among the models and provides appropriate guidance to the forecaster. The five models used by Bhowmik’s group for constructing the MME are: the IMD-GFS, the WRF, the JMA and the NOAA’s Quasi Lagrangian Model (QLM). As Figures 3 (a) and (b) show, the predictions made using the IMD-MME were fairly accurate and the final official forecast of track and landfall on October 9 was nearly the same as the MME’s predicted track and rainfall.
Many low pressure areas form over the sea but only a few intensify into a cyclonic storm. To identify which of the lows can potentially turn into a cyclone, Bhowmik’s team has developed what is called the Genesis Potential Parameter (GPP), which is an index defined as the product of four dynamical variables. “At the early [stage] itself we should get an indication of this intensification,” said Bhowmik. “Based on past cyclones, the threshold value for the index has also been defined. If the GPP is more than 8.0 in T1.0 (a pressure low), there is a chance that it will become cyclonic. We have found it to be very useful. In the case of Phailin, right from the beginning the GPP had a very high value, and the alert on October 7 was given on this basis. We have been using this index for the last five-six years.”
At present the skill of the dynamical models in predicting intensification is quite low. Post-processing (quantitative value addition) of model outputs improves the skill of intensity prediction. The IMD has developed a statistical model for intensity prediction called the Statistical Cyclone Intensity Prediction (SCIP) model. It uses a multiple linear regression technique whose parameters have been determined from the database of 62 cyclones that developed over the Bay of Bengal during the period 1981-2000. The model output values for these parameters are the predictors for determining the intensity via the regression equation.
“Using SCIP, our prediction was quite accurate,” said Bhowmik. “But this only goes as an input to the forecasters for guidance. They may not always go by this but this has proved to be very accurate in the last few years. This time, too, the SCIP model and the observed values matched.” There is another parameter that the IMD has defined, called the Chances of Rapid Intensification (CRI). Rapid intensification is when intensity increases by 30 knots (55 km/hr) in 24 hours. For Phailin, the probability jumped from 9.4 per cent on October 9 to 73 per cent on October 10.
Storm surge Once track and intensity forecasts are correct, forecasting the adverse weather impact of a cyclone is also important, and the most adverse weather system is the storm surge, which is an abnormal rise of water over and above the predicted astronomical tide as the cyclone crosses the coast. It depends on the intensity of the cyclone and shore bathymetry, with a shallow coastline generating higher surges. With the track and intensity forecasts, pressure drop becomes known and, using that, prediction of storm surge heights can be made.
The IMD uses a combination of nomograms and a storm-surge prediction model developed at Indian Institute of Technology Delhi. Nomograms, which are basically a coast-specific relationship between peak surge heights and pressure drops, are derived by a combination of empirical methods and numerical analysis. Bathymetry data provides the coast-specific input to the numerical model. “Coast nomograms,” said Bhowmik, “are very consistent but models may not always be accurate. The forecaster uses both to make predictions for a given cyclone and coast. For Phailin, the forecast was 3-3.5 m, which turned out to be correct.”
The other adverse weather factor that needs to be forecast is the radius of maximum rain (RMR)—a cyclone’s size and wind distribution determine that. This information is obtained from scatterometric data on wind distribution and cloud top temperature (CTT) obtained from satellites. The CTT profile has a relationship with pressure and wind. So if CTT is known, wind distribution can be known and from wind distribution the RMR can be determined.
If CTT is not available, as is the case at present for the IMD, the wind distribution can be estimated from the diameter of the middle of the eye wall. The wall has a width of about 100 km and the intensity drops on both sides of its centre. So, usually, the RMR is about 30-40 km. The forecaster estimates the correct value of the RMR. As the cyclone approaches closer, Doppler radar derived wind distribution can also be used to confirm the RMR.
According to Mohapatra, in the case of Phailin, the IMD also attempted to predict coastal inundation, though this was not made public. Only a detailed post-cyclone survey will verify this experimental prediction.
The accuracy of these various predictions in the case of Phailin has demonstrated the growing predictive skill of the IMD scientists behind this complex process of cyclone detection, monitoring, modelling and forecasting of its evolution and decay.
This expertise and maturity in the IMD system has not come overnight. Planned and focussed programmes over the years have helped achieve that. In particular, the cyclone Forecast Demonstration Project (FDP), a multi-institutional programme that has been ongoing since 2008, has contributed in no small measure towards that.
COMMents
COMMents
SHARE