I am using the quantile delta method for bias correction of future rainfall data (NA-CORDEX).
As my observed data are point (station) wise CSV files. I have extracted for corresponding locations. All the annual average value for both historical, future and observed rainfall seems fine which comes ~1000-1500 mm. Boxplot for unique rainfall values in historical, future model, and observed data are attached. Here, on wet days, the minimum value for observed data is 2.54 mm.
I found bias-corrected rainfall annual data is 20 times larger than the raw data ( Which I have checked with other approached too like mixed distribution having gamma distribution, Gamma Quantile Mapping, and found nearly the same result). This seems an unrealistic result.
Does anyone get such results or I am doing it in the wrong way?
Thanks
Hi TBS:
I have been there. Please check out Article Linking Global Climate Models to an Integrated Hydrologic Mo...
Conference Paper Uncertainty Transfer in Modeling Layers: From GCM to downsca...
Therein, we highlight problems related to zeroes (0) in the quantile delta method and how the GCM drizzle problem is inherited and causing big problems in the the estimation of the quantile distribution of both native GCM and downscaled precipitation products. I hope this references help in any way.
Dear Takhellambam,
I am not sure if I understood correctly your question. Was does "unique" means, do you mean "daily". Moreover, I am wondering if you introduced a wet day threshold for the model output? If you sort out all "zeros" or very small values (drizzle effect of RCM output), the boxes should look different. However, values are still too low compared to the observations, but that's why you should bias correct :-).
Please keep me updated.
Cheers, Patrick
Thank you so much, Patrick Laux for your reply. Here, unique rainfall values refer to the distinct rainfall amount which recorded in hourly interval. Such large bias-corrected rainfall value comes from observed data. I am assuming that the large unique values of observed data led to the higher quantile value correspond to the percentile of projected rainfall. In addition, the number of rainy/wet days (rain>0) in the climate model is more than observed rainfall. The raw climate models and observed rainfall data have approximately similar annual average values due to large hourly data with fewer rainy days in observed rainfall and a smaller quantity of rainfall with higher rainy days in climate models.
Okay, this means that you have to adapt the threshold for rainy days (could e.g. be 0.5 mm or 1.0 mm that it corresponds roughly to the frequency of the observations). After that, you can apply the QM and the values of the modelled wet days should be increased and reach those of the observations.
Thank you for the suggestion. This may help in getting a realistic value. Could you please share in detail about choosing the threshold value like links or approach?
- Badges
- Science topic