ISIMIP3: replaced bias-adjusted climate input data


Posted by Iliusi Vega del Valle on May 17, 2021

The ISIMIP3b bias-adjusted climate input data have been completely replaced. Any new ISIMIP3b simulation should use this new data and can now be started!

It is not yet decided whether impact simulations that were done based on the old input may have to be replaced, and the decision will depend on the magnitude of the effect of the correction. Modellers will be informed of a proposed associated test via the isimip-modellers mailing list. Based on the outcome of such test, sector-specific decisions will be taken in close cooperation with the sector coordinators.

The replacement of the ISIMIP3b climate forcing data is based on updates of (i) the bias adjustment and statistical downscaling method ISIMIP3BASD from v2.4.1 to v2.5.0 and (ii) the observational reference dataset W5E5 from v1.0 to v2.0.

The update of ISIMIP3BASD solves three issues:

  1. It allows for a bias adjustment in running-window mode and this way fixes the discontinuity issue which was the main motivation of this data update. The improvement can be seen in plots of regional mean annual cycles for all climate models and bias-adjusted climate variables (old results in dashed, new results in solid). The averages shown in those plots were calculated over all years of a climate scenario (color) and all land grid cells belonging to a Giorgi region (panel).
  2. The bias-adjustment method for near-surface relative humidity (hurs) was changed to no longer produce unrealistic distributions under climate change for climate models which simulate too frequent supersaturations (hurs > 100 %). Several of the ISIMIP3b GCMs have that problem, particularly in high-latitude winter. The problem and its solution are described in detail in the updated ISIMIP3b bias adjustment fact sheet.
  3. The daily maximum near-surface air temperature (tasmax) is now guaranteed to exceed the daily minimum near-surface air temperature (tasmin) by at least 0.01 K. Before the update, there were some cases of equal values of tasmax and tasmin.

The update of W5E5 has two consequences:

  1. A bug was fixed that affected the computation of surface downwelling shortwave radiation (rsds) of the underlying WFDE5 dataset. This has reduced the annual global land mean rsds by about 2 W m-2, see our global maps of multi-year mean rsds for regional changes.
  2. The method used for the bias adjustment of the monthly mean diurnal temperature range (tasrange = tasmax – tasmin) of WFDE5 was changed from a purely multiplicative method to a mixed multiplicative-additive method, see the product user guide of WFDE5. As a result, the excessively high tasmax values that occurred in some months and land grid cells in W5E5 v1.0 do no longer occur in W5E5 v2.0. Global maps of multi-year maximum tasmax values illustrate the improvement (upper row). They also show how the improved W5E5 tasmax data translate to better bias-adjusted tasmax data (lower rows). Differences between old and new bias-adjusted values are greater in most cases than differences between W5E5 v1.0 and v2.0. This is because tasrange is now bias-adjusted using a Weibull distribution, not a Rice distribution as before since the Weibull distribution turned out to fit the data better in most cases, particularly in the upper tail.

Please note that the update of the ISIMIP3a climate input data is still pending. We are waiting for an update of the GSWP3 data that will fix the discontinuity issue of that dataset.