Data_ImpactofBiasNonstationarity.zip

<b>Introduction</b><br><br>This dataset is an expanded dataset of the results in Van de Velde et al. (2020) and Van de Velde et al. (to be submitted).<br>It contains the results for the set of indices described in Van de Velde et al. (to be submitted) of bias adjustment by a combination of occurrence-bias-adjusting methods (all described in Van de Velde et al. (2020)) and intensity-bias-adjusting methods (partly described in Van de Velde et al. (2020), all described in Van de Velde et al. (to be submitted).<br><br><b>Methods</b><br><br>The methods used for occurrence-bias-adjustment were:- Direct method (intensity-bias adjustment without occurrence-bias-adjustment);- Thresholding;- Stochastic Singularity Removal (SSR);- Triangular Distribution Adjustment (TDA).<br>The methods used for intensity-bias-adjustment were:- Quantile Delta Mapping (QDM);- modified Quantile Delta Mapping (mQDM);- Multivariate Bias Adjustment (MBCn);- Multivariate Recursive Quantile Nesting Bias Correction (MRQNBC);- dynamical Optimal Transport Correction.<br><br>Of these methods, the combinations of all occurrence-bias-adjusting methods and QDM and MCBn (as examples of univariate and multivariate methods) were discussed in Van de Velde et al. (2020). As thresholding seemed to be the most robust method, this method was withheld for use in Van de Velde et al. (to be submitted). The combinations of thresholding and all intensity-bias-adjusting methods were thus discussed in Van de Velde et al. (to be submitted).<br><br>Though this was a sensible choice w.r.t. readability, the combinations that were not discussed in any of those two papers could still reveal minor patterns or could be interesting for comparison with other results.<br><br>All methods except SSR, TDA and dOTC were considered deterministic and thus the combinations of these methods were only computed once. The combinations of SSR and TDA with a deterministic bias-adjustment method were computed 20 times. The combinations of SSR or TDA and dOTC were computed 100 times, 10 times for the occurrence-bias-adjusting step x 10 times for the intensity bias-adjusting step.<br><br>For all indices, the values were first calculated for each repetition and then averaged.<br><br><b>Data</b><b><br></b>The bias adjustment is applied on the EURO-CORDEX (Jacob et al., 2014) model RCA4 (Strandberg et al., 2015), with MPI-ESM-LR GCM (Popke et al., 2013) boundary conditions. RCA4 is used as it is one of the few RCMs with potential evaporation as an output variable.<br><br>For the observations, a dataset made available by the Belgian Royal Meteorological Institute is used in this study. This dataset comprises 117 years (1901-2017) of daily precipitation amount, daily average temperature and daily potential evaporation. climate change conditions. The time series used for adjustment were as follows: 1970-1989 was chosen as the ‘historical’ or calibration time period and 1998-2017 was chosen as the ‘future’ or validation time period. For 2006-2017, model data forced with RCP 4.5 was used. Only the model data for the grid cell containing the Uccle observatory was used and compared.<br><br>The local observations cannot be shared with third parties and thus were not included here. As such, including the fully adjusted time series seemed irrelevant (as there is no local data to compare them with), yet these can be delivered upon request. Instead, this dataset focuses on the post-processed results, to enable comparison between the different methods.<br><b>Post-processing</b><b><br></b>As discussed in Van de Velde et al. (2020) and Van de Velde et al. (to be submitted), we also calculated the 'Residual bias relative to the observations' (RB_O) and 'Residual bias relative to the model bias' (RB_MB). These two metrics allowed us to compare the different methods thoroughly and to display where the bias adjustment has, or has no, added value. For more information on these metrics, we refer to the aforementioned papers.<i><br></i><b>Content</b><br><br>This dataset contains 5 .csv-files, each containing the results of an intensity-bias-adjusting method with all occurrence-bias-adjusting methods. Each of these files is constructed as follows:- Columns 1, 9 and 15 contain the names of the indices used.- Column 2 contains the observed values for each index.- Column 3 contains the raw climate model data for each index.- Columns 4-7 contain bias-adjusted values for each index- Columns 10-13 contain the RB_Ovalues for each index. - Columns 16-19 contain the RB_MBvalues for each index. <br><br><b>Code</b><br><br>The code used for bias adjustment and for the evaluations is available on Zenodo/Github via https://doi.org/10.5281/zenodo.4247518<br><br><b>References<br></b><br>Jacob, D.; Petersen, J.; Eggert, B.; Alias, A.; Christensen, O. B.; Bouwer, L. M.; Braun, A.; Colette, A.; Déqué, M.; Georgievski, G.; Georgopoulou, E.; Gobiet, A.; Menut, L.; Nikulin, G.; Haensler, A.; Hempelmann, N.; Jones, C.; Keuler, K.; Kovats, S.; Kröner, N.; Kotlarski, S.; Kriegsmann, A.; Martin, E.; van Meijgaard, E.; Moseley, C.; Pfeifer, S.; Preuschmann, S.; Radermacher, C.; Radtke, K.; Rechid, D.; Rounsevell, M.; Samuelsson, P.; Somot, S.; Soussana, J.-F.; Teichmann, C.; Valentini, R.; Vautard, R.; Weber, B. &amp; Yiou, P.: EURO-CORDEX: new high-resolution climate change projections for European impact research, <i>Regional Environmental Change</i>, 2014, 14, 563-578<br><br>Popke, D.; Stevens, B. &amp; Voigt, A.: Climate and climate change in a radiative-convective equilibrium version of ECHAM6, <i>Journal of Advances in Modeling Earth Systems</i>, <i>Wiley Online Library</i>, 2013, 5, 1-14<br>Strandberg, G.; Bärring, L.; Hansson, U.; Jansson, C.; Jones, C.; Kjellström, E.; Kupiainen, M.; Nikulin, G.; Samuelsson, P. &amp; Ullerstig, A.: CORDEX scenarios for Europe from the Rossby Centre regional climate model RCA4, <i>SMHI</i>, <i>SMHI</i>, 2015 <br><br>