Owing to the significant spatial and temporal variability of weather patterns, prediction often becomes a tricky affair. The ways in which factors determining the weather over a particular time period—wind, temperature, relative humidity and pressure—behave, bring in an uncertainty in determining which conditions are most likely to have a greater influence on the weather. Arriving at a conclusive (deterministic) prediction is therefore a challenge.
However, it is also important to arrive at a strong probability, if not complete accuracy, in forecasts. This becomes pertinent in the case of the Indian South-West monsoons, the variability of which has long-term impacts on agricultural yield, economy, water resources and power generation. Prediction also becomes important for mitigation of natural disasters like floods or cyclones.
Around the world, forecasting is currently done using high resolution ensemble prediction systems. The National Centre of Environmental Prediction (NCEP) in the United States and the European Centre for Medium Range Weather Forecasting (ECMRWF), Reading, UK have relied on these systems for years. Earlier, India mostly used deterministic, low resolution models to make forecasts. In the past six years, major improvements were made. Beginning with National Monsoon Mission of 2012, the Ministry of Earth Sciences (MoES), Government of India, sought to achieve high resolution prediction capability in different space and time scales. Accordingly, the models used by the NCEP were identified and attempts were made to establish an improved operational prediction system for monsoon in short, medium, extended and long range time scales.
The first developments began with the adoption of the Climate Forecast System (CFS) used by the NCEP. The CFS is a coupled ocean-atmosphere global modelling system and getting initialised with data from ocean, atmosphere and land, it provides long range forecasting (a season ahead). But since there were other forecast applications at shorter space and time scale, therefore the Indian Institute of Tropical Meteorology (IITM), Pune, the Indian Meteorological Department (IMD) and the National Centre for Medium Range Weather Forecasting (NCMRWF) made further improvements to it. The CFS version 2 (T382) provides a horizontal resolution of ~38 km which is highest in the category of seasonal forecast model globally and used presently for operational seasonal monsoon forecast by IMD. Forecast for a period of 15-20 days (~2 to 3 weeks) was made possible using a combined global forecast system (GFS)/CFS 2 system, which proved useful for agricultural, town planning, prediction of extreme rainfall events and heat waves over the country.
The next step, taken by IITM and NCMRWF was the development of an ensemble based high resolution model which further increased the resolution to (~27 km). Subsequently, IITM made an unprecedented improvement in increasing the resolution of their probabilistic system by further increasing the resolution to (~12 km). Another short range ensemble prediction system at 12 km resolution based on UK modelling system, has been established at NCMRWF . These systems, now being used in India, utilise the global model, where observations are gathered and data from across the globe (land and ocean) from various platforms—satellites, radiosondes, surface meteorological observations, are all assimilated
Speaking with G’nY, Parthsarthi Mukhopadhyay, senior scientist at IITM, Pune notes, “Weather conditions are chaotic in nature and subject to continual changes and this as such brings up the chance of errors in forecasting. With the increase in time of forecast, the errors in forecasts also increase. Due to this, a deterministic approach —meaning a single model to use for weather forecast will have possibility of more errors. To avoid this, scientists have developed ensemble technique where many members of a model are being used with slightly different (perturbed) initial condition to generate a set of forecasts instead of a single value (deterministic) forecast and from the set of forecasts, probability of most likely forecast could be generated with a percentage. Such approach is proven to produce forecast with reduced error and inculcate
The forecast output at high (~12 km) resolution therefore has better skill as compared to the previous coarser resolution of (~27 km). As Mukhopadhyay explains, it is similar to how a higher resolution camera brings out better details in a picture: “More details on the parameters affecting weather behaviour and its variability can be found when such forecast model is used. Therefore it becomes easier to find the weather variability over specific area and region with greater details.” Figure 1 shows a recent event of heavy rainfall over coastal Karnataka and the corresponding probability.
Previously, the highest resolution was being used by the ECMWF (~16 km). In fact none of the weather forecast organisations around the globe generate ensemble prediction information on such a scale. The (~12 km) resolution is now being used by the IMD operationally for issuing 10 days forecast over the country and attempts are being made to develop the block level forecast for agricultural purpose.
The development of high resolution short range ensemble weather prediction system has provided promise and opportunity for a suite of other applications of weather model output in the society. The Forest Survey of India, located in Dehradun, through utilisation of the 12 km weather forecast attempts to monitor/forecast the possibility of forest fires. Further, in the backdrop of an increased emphasis on generation of green energy, especially wind and solar by the Indian government, this 12 km resolution model forecast provides the necessary data for prediction. The variety of applications are shown in Figure 2.
The development of 12 km resolution weather forecast and its probabilistic prediction for 10 days has brought a significant change in the realm of numerical weather prediction capability in the country and also enhanced the possibility of a suite of application to the society which otherwise would not have been possible. A timely step of MoES, in procuring two state of the art high power computing (HPC) systems—Pratyush at IITM, Pune and Mihir at NCMRWF, Noida have played a key role in accomplishing the ensemble prediction system at the highest resolution available globally.