I'm working on the impact of climate change on water resources. how to choose the best ensemble from RCM projected rainfall? what method I should use to compare different RCM and choose the best out of that?
Dear Razi, the first step is to see what models you have available and then examine the quality of their data. A path to the answer could be to examine sensitivities and uncertainties on the final output you are studying.
Some papers you may find helpful:
https://journals.ametsoc.org/view/journals/hydr/19/8/jhm-d-17-0189_1.xml
Article Statistical downscaling of regional climate model output to ...
Article Selecting regional climate scenarios for impact modelling studies
start with looking at simple stat such as mean median std, number of rainy days and number of non rainy days etc. you should choose those RCM which different from each other. this will widen the ensemble and cover the uncertainty.
Dear Razi,
In addition to Angelos Alamanos and Shailesh Kumar Singh it would be good to determine the sensitivity of the hydrological variable(s) of interest (e.g. average streamflow, annual maximum flow) to different rainfall statistics at different temporal scales (e.g. daily, two-daily, etc. rainfall sums, annual maximum values etc.). Subsequently, you can focus on the most important statistics identified in this sensitivity analysis when comparing in-situ rainfall and RCM rainfall for the same historic period. Based on this comparison, you can rank and select RCMs to be included in your ensemble.
If relevant (i.e. depending on the hydrological variable(s) of interest), a similar analysis can be done for other meteorological variables such as temperature.
Good luck,
Martijn
Dear Razi,
that is a good question which puzzled me a lot during the last few years.
There are various approaches on that and it always depends on your expectations ...
1. As decision maker you may want to use various future projections, covering e.g. wet a future, a dry future, or a hot or cold future. There are some papers from Alex Ruane on this ... I have applied it also in my latest publication "To bias correct or not to bias correct ..."
2. You may also want to determine the "best" performing RCM based on an validation of the historical "baseline" simulations. Then you compare the historical simulations with the statistics of observations or re-analysis data.
The question here is if this gives you enough credibility that the future projections are also the most reliable ones? The underlying scenarios to drive the RCMs are extremely uncertain, which means that even you trust a certain RCM more than another one, this does not urgently lead to more reliable simulations for the future period.
I would therefore prefer the first procedure of considering RCMs which cover different states (cold/wet, hot/wet, cold dry, cold wet, normal) to generate an ensemble, but if you think you can identify the best performing (or if you are only interested in simulations for the past), there is a subsetting algorithm from the group of Samaniego (main author is Stefan Thober).
Dear Razi,
In addition to the detailed response of Patrick Laux, I would like to mention that recently I have published a paper on GCM(s) selection. Please have a look at that paper on the following link:
Article Evaluation of CMIP5 Models and Ensemble Climate Projections ...
I hope this will help you.
With best regards,
Firdos
Dear Razi,
In stead of choosing best RCM ensemble (unless you have confidence in these models based on previous research or a simple evaluation in the baseline period), using as more RCMs as possible is always welcome.
Best,
Zhao
Thank you all for your valuable suggestions. It will surely help me to select the best possible RCM.
Great answers so far. However I will differ and state that there is no such thing as the 'best RCM' or 'GCM' when climate projections are concerned. We can say this when comparing historical RCM simulations to observations/re-analyses.
Advice: use as many models and realizations as possible. Try ensemble selection methods to extract the most releavnt information from a large ensemble.
- Badges
- Science topic
- Similar topics
- Geoscience
- Asia
- India