Monsoon rainfall in India – How well do reanalyses compare to real-world data?

NERSC researcher Tarkeshwar Singh (CDP group) recently published an article with colleagues from India. They investigated the performance of three high-resolution atmospheric reanalyses over the Indian region by comparing the rainfall data with focus on the extreme events.


Monsoon rainfall in India – reasons and implications

The Indian Summer Monsoon is an annual natural phenomenon over India that affects millions of people directly and indirectly. It accounts for over 80% of the rainfall over the Indian region each year, just during the months June to September. The distribution of monsoon rain depends on both the land relief and the wind direction the rain-laden clouds are affected by. But changes in extreme rainfall events are related to changes in large-scale circulation patterns over the tropical regions, where the moisture is coming from which rains down during monsoon season.

When and where heavy rainfall occurs during this season has implications for society in more ways than one: Agriculture depends on rain, but too much rain results in flooding that can rob hundreds of thousands of people of their homes and livelihood. Understanding how the amount of rainfall during the monsoon may change in the future is crucial, for example to allow for the mitigation of flood risk in certain regions.

Atmospheric reanalyses of rainfall

Investigating the climate on fine scales has become more and more important in the past decades, and one tool sticks out in the world of climate prediction: reanalysis datasets. They are created by combining real-world data, such as for example atmospheric data (air temperature, moisture content, amount of rainfall, wind direction, etc.) and satellite data with computer models to create an extremely detailed look back in time. Atmospheric datasets never cover every point in a landscape, so to produce the best possible and most complete look back in time – a reanalysis – mathematical formulas are used to fill in the gaps. Such a reanalysis can span various time scales; common are several decades, dependent on the data coverage. With such a reanalysis one can look at any point in time (that is covered by it) and see how the weather most likely was back then in an area of interest.

For climate investigations, climate models can use reanalysis data; a best possible look back in time helps to get a clearer picture of what the future will bring. These reanalyses can either be produced by including data from the entire globe for a certain period, or they can only include data from one area, such as for example South Asia.

Three atmospheric reanalyses covering India: IMDAA, NGFS, and ERA5

In recent years, three high-resolution atmospheric reanalyses were produced that cover South Asia and use satellite data and conventional observations. Singh and his colleagues set out to study how good these three datasets are compared to real-world atmospheric data from the Indian subcontinent.

  • IMDAA (Indian Monsoon Data Assimilation and Analysis) is a high-resolution (~12km) regional reanalysis exclusively for the South Asia region, produced by Met Office (MO), U.K, National Centre for Medium Range Weather Forecasting (NCMRWF), India and the India Meteorological Department (IMD).
  • NGFS (NCMRWF Global Forecast System model) on the other hand is a global high-resolution (~25km) reanalysis, produced by the NCMRWF in India.
  • ERA5 (ECMWF Reanalysis version 5) is also a global high-resolution (~30km) reanalysis, similar to NGFS. ERA5 is produced by the European Centre for Medium-Range Weather Forecasts in Europe.

IMDAA is the first high-resolution atmospheric reanalysis for the Indian region, and it was produced to better understand past Indian summer monsoons to better predict future monsoon trends. All three reanalyses can be used to investigate the monsoon rainfalls for the past decades, and Singh and his colleagues chose to study the period 1999-2018.

The real-world atmospheric data they chose as comparison stems from the India Meteorological Department – it is the measured rainfall from across India from 1999-2018. The dataset had to be adjusted to be comparable to the reanalyses, information on that process can be found in the publication.

How well do the three different reanalyses perform compared to the real-world data?

Probability Distribution for the rainfall intensity of IMD, IMDAA, NGFS and ERA5 reanalysis data respectively. Similar curve for light rain category is shown in the inset. Singh et al. 2021Probability Distribution for the rainfall intensity of IMD, IMDAA, NGFS and ERA5 reanalysis data respectively. Similar curve for light rain category is shown in the inset. Singh et al. 2021Singh and his colleagues divided the Indian landscape into six homogeneous monsoon rainfall regions and they checked how well each reanalysis matches the real-life dataset for the different seasons (winter, pre-monsoon, monsoon, post-monsoon; and annual) in the different regions.

Overall, all three reanalyses compare well to the observed rainfall over the 20-year period. Some notable findings are:

  • IMDAA and NGFS performs best when looking at extreme monsoon rainfall but ERA5 is underestimating the observed extreme rainfall intensity.
  • NGFS performs best when looking at light rainfall, and the other two reanalyses also perform well.

Singh and his colleagues investigated the three reanalyses in great detail, and if you are interested in reading the full description of their results, you can read their article online.

First author Tarkeshwar Singh on the relevance of this study: “This work provides useful insights for the climate modeling community to establish appropriate benchmarks for performing model evaluation using newly-developed high resolution atmospheric reanalysis datasets over the Indian region to study Indian monsoon features critically.”



Singh, T, Saha, U, Prasad, VS, & M Das Gupta. Assessment of newly-developed high resolution reanalyses (IMDAA, NGFS and ERA5) against rainfall observations for Indian region. Atmospheric Research 2021. 259, 105679, 

Add comment

Login to post comments