EGU General Assembly 2020
© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

Current Risk of Extreme Monsoon Rainfall over India using Large Ensemble Simulations

Shipra Jain1,2, Adam A Scaife3,4, Nick Dunstone3, Doug Smith3, Saroj K Mishra2, and Ruth Doherty1
Shipra Jain et al.
  • 1School of Geosciences, University of Edinburgh, Edinburgh EH9 3FF, United Kingdom of Great Britain and Northern Ireland (
  • 2Centre for Atmospheric Sciences, Indian Institute of Technology Delhi, Hauz Khas, Delhi, India
  • 3Met Office Hadley Centre, Fitz Roy Road, Exeter, Devon EX1 3PB, United Kingdom
  • 4College of Engineering, Mathematics and Physical Sciences, University of Exeter, Exeter, Devon, UK

India suffers from severe social-economic losses due to floods and droughts during boreal summer (June-September) and therefore there is a growing interest in the current risk of extreme monsoon rainfall. In this analysis, we estimate the risk of flood, drought and unprecedented (outside the range of present observational record) rainfall over India using UNprecedented Simulated Extremes using ENsembles (UNSEEN) method. The UNSEEN is a statistical framework under which the risk of unprecedented rainfall extremes can be estimated using a large ensemble of initialized climate simulations to sample a broad range of internal variability. This is the first application of the method to the hindcasts from multiple coupled atmosphere-ocean models. Under this method, we first test individual models against the observed rainfall record over India and select models that are statistically indistinguishable from observations. The risk of floods, droughts and unprecedented rainfall is then estimated using a large ensemble of summer precipitation simulated by the selected set of models. We note that in present climate the risk of drought is higher than the flood, with droughts being more frequent and intense than the floods. This asymmetry in rainfall extremes is found to be partly due to the asymmetry in El-Nino Southern Oscillation (ENSO) phase, with El Nino reaching higher magnitude more frequently than La Nina. The current risk of record breaking drought (>23% deficit w.r.t climatological mean) is 1.6% whereas the risk for record-breaking flood (>16% excess) is 2.6%. There is even a risk of 30% rainfall deficit that could occur around once in two centuries, which is not yet seen in observations and would have a catastrophic influence on India.

How to cite: Jain, S., Scaife, A. A., Dunstone, N., Smith, D., Mishra, S. K., and Doherty, R.: Current Risk of Extreme Monsoon Rainfall over India using Large Ensemble Simulations, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-7674,, 2020.


Display file

Comments on the display

AC: Author Comment | CC: Community Comment | Report abuse

displays version 3 – uploaded on 29 Apr 2020, no comments
displays version 2 – uploaded on 26 Apr 2020
PDF version
  • CC1: Comment on EGU2020-7674, Laura Suarez-Gutierrez, 28 Apr 2020

    Hi Shipra,

    Thanks for the new upload, it's working perfectly now. Very interesting!



displays version 1 – uploaded on 10 Apr 2020
  • CC1: Quick question on bias correction, Flavio Lehner, 15 Apr 2020

    [Thanks, Shipra, for uploading a presentation to our session already. I'm using this to test/illustrate the commenting function, but also because I have an actual science question.]

    I think this is an interesting and innovative approach to using the NMME. In your workflow, "Step 1" (slide 9) is bias correction and "Step 3" (slide 10) is selecting models that match observations. I was wondering whether you can briefly explain the bias correction step. My first impression was that bias correction would to some degree *enforce* a match between model and observations, but it depends what was done exactly. So I was wondering what differences remain between model and observation after bias correction (step 1) that allow you to further select models based on whether they match observations (step 3)? It seems your bias correction is just fixing the absolute mean bias, leaving higher order moments untouched to be used as criterion for step 3, correct?



    • AC1: Reply to CC1, Shipra Jain, 22 Apr 2020

      Hi Flavio,

      Thanks for the interesting question. You are right- we calculated the difference b/w model ensemble mean and observations and then removed that value from each ensemble member. We did this for each model. Hope that answers your question. 


  • CC2: Comment on EGU2020-7674, Laura Suarez-Gutierrez, 22 Apr 2020

    Hi Shipra,

    I just downloaded your presentation and the visualization somehow did not work with my office provider (text was missing or misplaced)




    • CC3: Reply to CC2, Laura Suarez-Gutierrez, 22 Apr 2020

      Just wanted to add, maybe you could try uploading a .pdf version?

      • AC2: Reply to CC3, Shipra Jain, 22 Apr 2020

        Hi Laura,

        I am sorry that it didn't work and thanks for the useful suggestion. I will add the pdf version.

        Best wishes,