Forecasting Tropical Cyclones

By: M Mohapatra
Forecasting deals with the prediction of genesis, location/track and intensity of a tropical cyclone. It also aims at predicting associated adverse weather such as heavy rains, gales, high waves, storm surges and coastal inundation.
Planning n Mitigation

Losses caused by a tropical cyclone (TC) depends on hazard and vulnerability analysis, preparedness and planning, early warning, prevention and mitigation. Early warning is a major component of disaster management in the south Asian region. This involves monitoring and prediction of a cyclone, effective warning and dissemination of relevant information to people, coordination with emergency response units and public perception regarding the credibility of official warning signals (Fig. 1). Continuous upgradation of an early warning system is necessary for effective disaster management of TCs.

Fig. 1: Monitoring and forecasting of a tropical cyclone
Fig. 1: Monitoring and forecasting of a tropical cyclone


Standard operation procedure

The TC analysis, prediction and decision making process is made by blending conceptual models, dynamic and statistical models, meteorological data sets, technology and expertise. A decision support system (DSS) in a digital environment is used to plot and analyse different weather parameters, using satellite, radar and Numerical Weather Prediction (NWP) data. In this hybrid system, a synoptic method could be overlaid on a NWP model supported by modern graphical and geographical information system (GIS) applications to produce high quality analyses and forecasting, prepare past and forecast tracks up to 120 hrs, depict uncertainty in track forecast and to forecast wind directions and speed in the different sectors of a TC. Additional help is taken from radar network involving neighbouring countries, satellite imageries and products and analysis and forecast products from various national and international centres. Of late, automation has increased the efficiency of these systems.

Prediction of cyclogenesis     

If a depression with maximum surface wind (MSW) of 17-27 knots forms over the northern Indian Ocean, a cyclogenesis is said to have occurred. Prevailing environmental conditions such as sea surface temperature (SST), ocean thermal energy, vertical wind shear, low level vorticity, middle level relative humidity, upper level divergence, moist static stability, and low to upper level winds are considered for predicting cyclogenesis. These fields in the NWP model analyses, taken along with characteristic satellite and radar observations, are evaluated to monitor the cyclogenesis, step-wise.

A genesis potential parameter (GPP), for the northern Indian Ocean (NIO) is developed as the product of four variables, namely vorticity at lower levels, middle tropospheric relative humidity, middle tropospheric instability, and the vertical wind shear. The GPP is operationally used for predicting cyclogenesis in the early stages. This probabilistic forecast is issued in terms of nil, fair, good, high and very high probability corresponding to 0, 1-25, 26-50, 51-75 and 76-100 per cent probability of occurrence for the next 72 hours.

Fig. 2: (a) Observed and forecast track of depression that became the cyclone Thane, (b) A graphical presentation of quadrant wind forecast during Thane, 2011
Fig. 2: (a) Observed and forecast track of depression that became the cyclone Thane, (b) A graphical presentation of quadrant wind forecast during Thane, 2011


TC track forecasting

After determining the initial location and intensity, it is attempted to forecast the track and intensity. While the synoptic, statistical and satellite/radar guidance help in short range forecast (upto 12/24 hours), the NWP guidance is mainly used for 24-120 hour forecasts. Consensus forecasts that gather all or part of the numerical forecast and use synoptic and statistical guidance are utilised to issue the official forecast.

An example of Thane (Fig. 2) shows the ‘cone of uncertainty’ (CoU), also known as the ‘cone of death/probability/error’, representing the forecast track of the centre of a TC and the likely error in the forecast track based on past predictive skills.

TC intensity forecasting

A statistical-dynamical model has been implemented for real time forecasting of 12 hourly intensities up to 72 hours. For real-time forecasting, model parameters are derived based on the forecast fields of the IMD Global Forecast System (GFS) model. In the satellite method, a region of maximum reflectivity and mesoscale vortices are assumed to be associated with higher winds. In the radar technique, the direct wind observations are available through radial velocity measurements. The wind estimates from satellite, radar and other observations are extrapolated to forecast the wind. MSW is also available from other sources like scatterometry wind from satellite, buoy, and ships, apart from estimates by the Dvorak technique. Consensus forecasts that gather all or part of the numerical forecast and use synoptic and statistical guidance are utilised to issue the official forecast.

Fig. 3: Official track forecast of root mean square error, during 2003-13 over the northern Indian Ocean
Fig. 3: Official track forecast of root mean square error, during 2003-13 over the northern Indian Ocean


Cyclone wind radii forecasting

Representing the maximum radial extent of winds reaching 34, 50 and 64 knots in each quadrant (NW, NE, SE, SW) of TC, the cyclone wind radii are generated as per the requirement of ships. The initial estimation and forecast of wind radii of a TC is rather subjective and strongly dependent on data availability, climatology and analysis. The subjectivity and reliance on climatology is amplified in the absence of aircraft observations. The consensus forecast issued by IMD is based on a numerical forecast using NWP models and synoptic and statistical guidance (Fig. 2).

Adverse weather forecasting

In addition to the forecast of cyclone track, intensity and landfall location, IMD also provides specific forecast of following adverse weather phenomena associated with cyclones.

Heavy rainfall: The forecast/warning of heavy rainfall includes time of commencement, duration, area of occurrence and intensity of heavy rainfall. The methods for prediction of heavy rainfall include synoptic, climatological, satellite, radar, and NWP techniques. While NWP models provide prediction of rainfall for different lead periods, satellite and radar provides short-term (3 to 12 hours) forecast guidance. In the synoptic and climatology method, rainfall intensity and spatial distribution are used. The final forecast is the consensus arrived at using all of the above.

Gale wind: The forecast of gale winds involve predicting time of commencement, duration, area of occurrence, and magnitude. The methods for prediction of gale winds apart from the four elucidated above include dynamical statistical techniques.

Storm surge: A storm surge depends on the pressure drop at the centre, the radius of maximum wind, point of landfall and interaction with sea waves, astronomical tide, rainfall, river run-off, bathymetry, and coastal geometry. The forecast of storm surge includes time of commencement, duration, area of occurrence, and magnitude of storm surge. The methods for prediction of storm surge and coastal inundation include, IMD nomogram, Indian Institute of Technology, Delhi model, and INCOIS, Hyderabad model.

TC forecasting skill accuracy over NIO

The initiatives taken by IMD in recent years have resulted in improved TC forecasting and warning. The trends in forecast performance during 2003-13 are presented in the following section.

The TC track and intensity forecast issued by IMD between 2003 and 2013 is calculated by the margin of error and skills involved. The average track forecast error was about 124, 202 and 268 km and skill involved were about 36 per cent, 53 per cent and 62 per cent respectively for 24, 48 and 72 hour forecasts over the NIO during 2009-2013. The 24-hour track forecast error decreased at the rate of 7 km per year (Fig.3) and skill by 5 per cent per year. Even so, the error is higher than that of the National Hurricane Centre (NHC), USA by about 50 km in a 72 hour forecast period.

The landfall point forecast error has also reduced significantly in recent years. The 12 and 24-hour landfall point forecast errors have decreased at the rate of about 16 and 34 km per year respectively during 2003-2013. The average landfall point forecast error during 2003-13 is about 75, 98 and 124 km for 24, 48 and 72 hour forecasts respectively.


Notwithstanding significant improvements in TC forecasting in recent times, there remains scope for further improvement. Forecast demonstration projects, aircraft probing, augmentation in observational network, introduction of high resolution regional forecast model, data assimilation, synthetic vortex generation, multi-model ensemble and operationalization of ensemble prediction system products are being attempted as of now.

Leave a Reply

Your email address will not be published. Required fields are marked *