No, a bias correction is not a MUST, particularly not after STATISTICAL Downscaling. This inherently corrects for the biases. This is one of the main differences compared to Dynamical Downscaling (DDS). After DDS, you should check first how big the biases are.

Dear Dr. Patrick Laux,

 Thank you for your answer. I will happy also if you suggest me some literatures related to this, to support the idea.I want to quote on my paper.

Dear Dr. Kenneth M Towe,

Thank you for your response and attachment. I will look at your attachment. To make more  clear my question, it is about statistical downscaling from GCMs. When we downscale from general circulation models, specifically using Statistical Downscaling Model, some authors add the following formula on their result

For temperature Td=Ts-(Tc-Co)

Td—the de-biased (corrected) daily temperature data for the

future periods (e.g., 2011–2040 and 2041–2070).

TS—the daily temperature data generated by SDSM for the

future period  TC —the long-term mean monthly values of simulated temperature for the control

To—the long-term mean monthly values of observed temperature for the control

And for precipitation

Pdeb=Pscen∗(Pobs/Pcont)Pdeb=Pscen∗(Pobs/Pcont)

Where, Pdeb is de-biased (corrected) daily precipitation for future period, Pscen is SDSM generated daily time series precipitation for future period, Pcont is the long-term average of monthly precipitation for the control period simulated by SDSM, and Pobs is the long-term average monthly observed value of precipitation(Salman et al 2007). But Many researchers do not apply this bias correction formula just they take the default bias correction value on the SDSM Software.

Without applying this formula, I got a good result on daily temperature and precipitation, but when I apply this formula on the downscaled result to correct it ,the result after bias correction is not realistic compared to other studies, that’s why I ask whether this formula is an optional or a must in statistical downscaling from GCMs

Thank you.

Dear Gatnet,I guess it is written in Wilby and Wigley (1997), but I am not sure.I am interested in the literature you refer to, because I really see no need for a bias correction after a SDS (assuming that the SDS is from large scale (GCM grid cells) to  point scale (data from observation stations)). As a reviewer I would critically question the reason for that;-).

Dear Kenneth,

of course, you are right.

First, you have to identify the biases (systematic derivations of model results from observations), which is not that difficult. If you are not coming from the application side (e.g. a climate impact modeler) but from the climate modelling side, you should also try to find the reasons for the model biases.

For regional climate modellers, the reason is usually the scale mismatch between the observations and the RCM grid cells, inadequate parametrization schemes, errors in the observations, etc.

The scale mismatch cannot be the point in SDS, that is what I tried to point out in the previous communication, however, inadequate transfer functions or predictor variables could be the reason. This makes it difficult to justify the need for additional post-processing bias correction. In this case, the goal should be to find improved transfer functions or predictors for the SDS.

But you need some knowledge about the biases first, that was my main assumption.

Cheers, Patrick 

Dear Kenneth,

I think this discussion is not about biases of measured variables. It is about differences between modeled variables (from climate models) and observed data (station data). The reasons for these biases from climate models (compared to observations) are mentioned above.

For the figures you attached I have no clue. I guess that they illustrate the deviations of the current climate compared to an earlier period. I would not label this as bias, it is rather variability. But maybe I misinterpret the figures wrong.

Hope this answers your question! 

Dear Kenneth,

if it is not differences between different periods, it might be differences between different sources?!

But, I am confused: Looking at your Iceland example, it is written monthly changes 2015 minus 1931-1940 averages. This means you look at DIFFERENCES of a recent year to a 10-yr period in the past. This is not a bias in the sense of a systematical deviation from a model to an observation, right?

There are many sources of observations, station-based as well as gridded interpolated products. For Germany, for instance, I am using data from the German Weather service. For other regions, I am using data from the countries weather services.

KMNI (climate explorer) provides or provided(I did not check for a long time) data for many countries worldwide.

There are plenty of other sources such as TRMM (derived global product from remote sensing), APHRODITE for Asia, etc. The list is endless ...

Cheers, Patrick

bias correction is a vital tool during a statistical downscaling, especially when processing re-analysis type of climate data.