The monsoon in India is responsible for meeting the freshwater requirements of more than 1.3 billion people living in the country. The moist southwest monsoon winds bring plenty of rainfall from June to September. India’s GDP is closely linked with the monsoon as agricultural production is largely dependent on the quantum of rainfall (Gadgil and Kumar, 2006). Moreover, the monsoon renews water in dams which is used in the drier seasons for domestic consumption, irrigation and construction. Therefore, getting an advance estimate of rainfall and the availability of water in dams is of utmost importance for its sustainable management. Timely estimation on different timescales—from a few days to the complete season—forms the basis of monsoon forecast science.
Monsoon forecast and its challenges
Although, monsoon forecasts commenced in India more than a hundred years ago (Blanford, 1884), it still poses a challenge for meteorologists. The most important reason behind the problem is the complexity of monsoon flow and its evolution which are influenced by a number of factors. The problem partly arises from the non-linearity in defining the system. However, there are certain aspects of the monsoon that are associated with inherent predictability. These can be used to provide an outlook for future evolution with a degree of success (Charney and Shukla, 1981; Goswami and Xavier, 2003; Sahai et al., 2017). In the last 100 years there has been progress in terms of developing sophisticated forecast techniques and models and collecting observational data.
With advanced computational capability, monsoon forecast has evolved in generation and dissemination. India has now shifted to a dynamical model forecast based on the Navier-Stokes equation. These forecasts are generated by solving a complex set of differential equations using numerical methods on high-performance computers. Since these forecasts consist of differential and integral equations, they require certain initial and boundary conditions which are based on observations. These atmospheric observations are received at operational centres from across the world through the Global Telecommunication System (GTS), World Meteorological Organization (WMO) via India Meteorological Department (IMD). All these observations are then processed, checked for quality and then, assimilated based on mathematical techniques to generate the initial condition (or boundary condition). This form of dynamical forecast based on boundary conditions is in contrast with the statistical method which ascertains predictors of rainfall through the frequentist approach. The dynamical forecast is based on a sophisticated and expensive computation that takes place round-the-clock. The manifold increase in the computing capability under the Ministry of Earth Sciences (MoES)of the Indian Government has made dynamical forecasting viable.
Typically, the monsoon forecast can be classified into four categories on the basis of a modern approach. These are:
- Short-range forecast (validity for 1-3 days)
- Medium-range forecast (validity for 4-10 days);
- Extended-range forecast, (validity for 10 days to a maximum of 4 weeks) and;
- Long-range forecast (mostly done for seasonal predictions).
Each of these forecasts has its own specific application that can benefit several stakeholders. However, this article dwells on various facets of the extended range forecast.
The perspective of extended range forecast
The extended range forecast (Sahai et al., 2019) aims to provide a rainfall outlook from 10 days to a maximum of 4 weeks. This forecast is crucial for agricultural and hydrological stakeholders. Extended range forecast project at the Indian Institute of Tropical Meteorology (IITM) was envisioned more than a decade ago to improve forecast beyond weather scale which is a challenging gap area in research and operational forecast domain. WMO recommended the Global Framework for Climate Service (GFCS) to generate forecasts for assisting stakeholders for both, short and long-term. Improvement in forecast delivery based on the extended range scale is essential to meet the exigencies of such stakeholders. A fully operational extended range forecast has been in place since July 2016.
Extended range forecast model
Several efforts have been undertaken in the past decade at IITM to refine the extended range forecast, of which the multi-model ensemble (MME) prediction system is an example. For generating the MME, the latest version of National Centre for Environmental Prediction (NCEP) coupled forecast system (CFS) model has been used (Saha et al., 2014). CFS components provide a realistic estimate based on initial conditions that have been prepared from coupled data assimilation system and the forecast is now operational. Extended range forecast is provided every week for the following four weeks. Forecasts are provided for various small regions of India (sub-regions) which are mostly homogeneous in terms of rainfall distribution. Like seasonal forecasts, extended range forecasts can be both statistical and dynamical. Dynamically extended range forecast is currently being generated by IMD. The CFS and global forecast system (GFS) models are solved on high performance computing systems at T126 (~110 km) and T382 (~38 km) global grid intervals which signify the horizontal spatial resolution of the model and is referred to as CFST126 and CFST382 and GFST126 and GFST382 (Fig. 1).
The extended range forecast provides an outlook of the fluctuations within the monsoon—onset, active/break cycles, withdrawal, tropical intra-seasonal oscillation such as Monsoon Intra-Seasonal Oscillations (MISO) and Madden Julian Oscillation (MJO), northeast monsoon—and also extreme weather conditions such as heat and cold-waves, heavy rainfall events and cyclogenesis. Extended range forecast has several uses especially, in the field of agriculture, hydrology, health, etc. For instance, receiving prior information regarding availability of rainwater and optimal time to sow crops is always useful. In this regard, weather forecasts for 2-3 weeks become paramount for planning. Similarly, such extended range forecasts about rainfall in a river catchment area would prove useful for dam managers. It can help them decide the amount of water to be held or released in order to circumvent the risks associated with floods or water shortage.
Indian Institute of Tropical Meteorology and extended range forecast
Any forecasting system has to be operationally stable and reliable before it can be deployed. In order to improve the extended range forecast products of IITM, the skills of the model in capturing various events needed to be verified (Fig. 2 a,b,c). The figure shows IMD subdivisions based on the forecasted data for the years 2001-2014. The skills for the monsoon forecast for the first three pentads have been given in P1, P2 and P3. Each pentad represents rainfall averaged for 5 days. Thus, forecast in P1 means rainfall averaged for a lead time of 1-5 days and P2 indicates forecast averaged for a lead time of 6-10 days and so on. The skill is measured as a correlation coefficient for each subdivision. Closer the values are to 1, the greater is the fidelity to the prediction. As can be seen from figure 2, the monsoon forecast from 2001-14 worked out to be significantly reliable for up to three pentads (P3). Correlation between model forecast and observation is ~0.4 in P3 reaching up to 0.7 in P1. (Sahai et al., 2019).
The current generation extended range forecast provides a reliable outlook of rainfall for 2-3 weeks in advance. The operational forecast in this scale has several uses that can be optimally used to assist agricultural and hydrological sectors.