New Bias-correction Method
We have published an article in Geoscientific Model Development which is part of the work I started when I was working at KUP. The purpose of the paper is two-fold. First, we describe two high-resolution climate simulations (2 km horizontal resolution) for the Alpine region in the period 1979–2005 and assess their performance over Switzerland. Second, a new process-informed bias-correction technique for precipitation is introduced and applied to these simulations. This article is rather technical (therefore the choice of journal), but it becomes necessary as we are using this data for other studies.
Simulations
The simulations were carried out at CSCS and we were very ambitious, as they span the entire Alpine region with an unprecedented resolution of 2 km (see domains below). One simulation is driven by the ERA-Interim reanalysis, and can be seen as a simulation that tries to mimic the most recent climate. The other is driven by an Earth System Model, the Community Earth System Model (CESM), and is the natural choice when the aim of the study is climate change. The first part of the article validates both simulations against an observational gridded product, with an emphasis on the evaluation of daily precipitation (note that the evaluation of the model performance for this domain in terms of the skill for surface wind is presented elsewhere). We find that the simulation nested to ERA-Interim is able to reproduce the main precipitation regimes, while the one driven by CESM presents severe seasonal biases
Bias correction
The aforementioned biases call for the application of bias corrections techniques. We opted for Quantile Mapping, which is a non-parametric approach that corrects the full distribution function as a whole. The idea is simple: the raw (to be corrected) and reference (corrector) Probability Density Functions (PDF) are obtained, and then values in the biased set are mapped “onto” the reference PDF, following the following rule (and graphically illustrated below):
\[ X_{\mbox{corr}}(t,x,y) = \left( \mbox{ECDF}_{\mbox{obs}}(t,x,y) \right)^{-1} \left( \mbox{ECDF}_{\mbox{model}}(t,x,y) \right) \]
However we apply a twist, which is motivated by the fact the precipitation regimes are very complex over this region. We apply the corrections after having found the PDFs for various independent regions, which are defined based on objective criteria with clustering methods (Ward algorithm). A pretty neat approach that surely we are using often in the future.
Tags: paper publications
Categories: Dissemination Science
subscribe via atom