Bias correction

Climate models are simplified representations of complex real world processes. An obvious example of this simplification is representation of the landscape as an array of flat grid cells of uniform altitude, temperature, humidity, snow depth, etc, even though we know full well that there are variations within each grid cell. These simplifications have consequences on the realism of model simulations. Models might not represent correctly, or at all, important climate phenomena such as the El-Nino/La Nina episodes, the rapid melt of Arctic sea ice with raising air temperature or the breakup of Antartica’s ice sheets. Another consequence is that when comparing observed climate conditions with their corresponding simulated variables, systematic differences appear. One model might overestimate tropical night-time temperature, while another might underestimate monsoon precipitation over East Asia.

These systematic biases usually have to be corrected for before using future climate simulations to assess the impacts of climate change. Indeed, if one wants to model crop productivity around 2050 with a model that is systematically too dry over the historical period, future projections likely won’t make sense. So-called bias correction methods are used to post-process climate simulations such that their statistics over the historical period match those of observations. The hypothesis is that the bias over the future period is likely the same as the bias over the historical period. This is a fairly strong hypothesis that has been shown to be false [], but until someone finds a better approach, it’s the best we can do.

Algorithms

class flyingpigeon.processes.KDDM_BC_Process[source]

kddm_bc Kernel Density Distribution Mapping Bias Correction (v1.0)

Bias correction method using Kernel Density Distribution Mapping (KDDM).

Parameters:
  • obs (application/x-netcdf, application/x-tar, application/zip) – netCDF files storing the data with the desired statistics, typically observations.
  • ref (application/x-netcdf, application/x-tar, application/zip) – netCDF files storing the reference data whose statistics are going to be mapped to the obs, typically simulations over the observed period.
  • fut (application/x-netcdf, application/x-tar, application/zip) – netCDF files storing the data to be downscaled using the mapping defined between ref and obs.
Returns:

  • output_netcdf_ref (application/x-netcdf) – Bias-corrected ref dataset.
  • output_netcdf_fut (application/x-netcdf) – Bias-corrected fut dataset.
  • output_log (text/plain) – Collected logs during process run.

References

McGinnis S., Nychka D., Mearns L.O. (2015) A New Distribution Mapping Technique for Climate Model Bias Correction. In: Lakshmanan V., Gilleland E., McGovern A., Tingley M. (eds) Machine Learning and Data Mining Approaches to Climate Science. Springer, Cham, DOI: https://doi.org/10.1007/978-3-319-17220-0_9