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

Addressing Uncertainties in Projected IDF Relationships under Climate Change

Kumar Puran Tripathy1 and Pradeep Mujumdar2
Kumar Puran Tripathy and Pradeep Mujumdar
  • 1Department of Civil Engineering, Indian Institute of Science, Bangalore, India (
  • 2Department of Civil Engineering, Indian Institute of Science, Bangalore, India (

The Intensity-Duration-Frequency (IDF) relationships are commonly used in urban hydrologic designs. A major source of uncertainty arises due to small samples of data and use of multiple GCMs, in developing the IDF for future periods. A major objective of this study is to address uncertainties in the IDF relationships for future periods, under climate change. The study proposes a Bayesian method for addressing the parameter uncertainty in the Generalized Extreme Value (GEV) distribution for the Annual Maximum Series (AMS). Uncertainties due to the use of multiple GCMs are addressed through the Reliable Ensemble Averaging (REA) method. The posterior distributions of the three parameters of GEV distribution are obtained using Markov Chain Monte Carlo (MCMC) method. Twenty-three CMIP5 GCMs with four RCPs are considered for studying the effect of climate change on the IDF relationship for the case study of Bangalore, India. Change Factor Method (CFM) is used for spatially downscaling the projected time series of precipitation and scale-invariance theory is used for temporal disaggregation. Results are compared across different CFM schemes considering multiple bin sizes. Uncertainties in design intensities are quantified through probabilistic IDF relationships.

How to cite: Tripathy, K. P. and Mujumdar, P.: Addressing Uncertainties in Projected IDF Relationships under Climate Change, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-2966,, 2020


Display file

Comments on the display

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

displays version 1 – uploaded on 27 Apr 2020
  • CC1: Question about the observations and the model, Oscar E. Jurado, 04 May 2020

    I was wondering if you could clarify the role that the observations had on your final model. I understand the AMS data was somehow incorporated with the GCM monthly data, but I don't understand exactly how, or in what way the observations influence the final results.

    Thank you!

    • AC1: Introduction, KUMAR PURAN TRIPATHY, 04 May 2020

      I am Kumar Puran Tripathy, Masters student in Civil Engineering at Indian Institute of Science Bangalore. I am working for my dissertation project under Prof. Pradeep Mujumdar in the topic -- Addressing the Uncertainties in Projected IDF relationships under climate change. I am thankful to all the EGU organizers and community for selecting my Project for the display. 


    • AC2: Answer, KUMAR PURAN TRIPATHY, 04 May 2020

      Thank you for your question. 

      I am not sure which observation you are asking about! 

      Let me discuss all. 

      I have shown that the parameteric uncertainties are bigger than that of the Model uncertainties. Actually, this has a major role when assessing the correctness of the model or your model efficacy. Suppose you don't have any idea which uncertainties are dominatung in your model, it gives an idea what to take (on your model). 


      Additive change factors have lesser future quantile values than that of Multiplicative. This observation on surface looks superficial, yet it has its own advantage. In case you are building a model that predicts the best quantile for the future. First, you incorporate all the model information that you can absorb. This is a take away from here. Actually I have considered both. I found that the multiplicative schemes work well in my location. However, it is contingent upon your geography and it's climate. 

    • AC3: Answer, KUMAR PURAN TRIPATHY, 04 May 2020

      I am sorry if I have interpretated wrongly earlier. 

      I have taken a AMS data for 33 years (1969-2001) for all the 7 time durations (sucha as 15-minute, 30-min and so on upto 24 hour). Basically ny data is a 33 by 7 matrix. And every column is  fitted with GEV distribution ( it has the maximum negative loglikelihood. Also by definition of GEV distribution the data exactly fits). If your question is how the parametric uncertainty is incorporated in the final model, then actually the posterior distribution that I have got ( you can view my presentation) is a measure of uncertainty here. It incorporates both prior and likelihood function upadated using Bayes rule. Different statistics are obtained from the posterior distribution, which take over the uncertainty part. 


      Similarly for model uncertainty, I have used Reliability Ensemble averaging. There are 15 different models and 4 RCPs for different cases. Thus, can you say which one is actually the best?  Therefore the model uncertainty arises. 


      However, in the final model we have not explicitly found out how much uncertainty is due to parameters or Due to GCMs, which is although challenging, yet we have found a similar response -- the model uncertainty is found to be lesser than that of paramter uncertainty. The exact amount is esoteric and intricating. 


      I hope it clears and I correctly exaplin your doubt. If still there are some doubts, please feel free to email me at 


      Thank you