Xclim is a Python library that enables computation of climate indicators over large, heterogeneous data sets. It is built using xarray objects and operations, can seamlessly benefit from the parallelization handling provided by dask, and relies on community conventions for data formatting and metadata attributes. xclim is meant as a tool to facilitate both climate science research and the delivery of operational climate services and products. In addition to climate indicator calculations, xclim also includes utilities for bias correction and statistical adjustment, ensemble analytics, model diagnostics, data quality assurance, and metadata standards compliance.
ESPO-G6-R2 v1.0 is a set of statistically downscaled and bias-adjusted climate simulations based on the Coupled Model Intercomparison Project 6 (CMIP6) models. The dataset is composed of daily timeseries of three variables: daily maximum temperature, daily minimum temperature and daily precipitation. Data are available from 1950 to 2100 over North America. The simulation ensemble is comprised of 14 models driven by two emissions scenarios (SSP2-4.5 and SSP3-7.0). In this paper, we describe the workflow used for the bias-adjustment, which relies on the detrended quantile mapping method and the Regional Deterministic Reforecast System (RDRS) v2.1 reference dataset. Using the framework defined in the VALUE project, we show the improvements made by the bias-adjustment on marginal, temporal and multivariate aspects of the data. We also verify that the bias-adjusted climate data have similar climate change signal to the original climate model simulations. Finally, we provide guidance to users on how to use this dataset.
Bias-adjusted climate simulations are increasingly disseminated through online platforms to support adaptation actions. However, there exists no agreed upon operational framework for producing these ”decision-ready” ensembles and for communicating the related uncertainty. In this paper, we use a systematic approach to assess the uncertainty related to bias-adjusted climate simulations across five dimensions: internal variability, shared socioeconomic pathway, global climate model, observational reference and bias-adjustment method. We calculate the fraction of uncertainty associated with each dimension for precipitation-based, temperature-based and multivariate indicators over Quebec, Canada and focus particularly on three locations: Montréal, Gaspé and Kawawachikamach. The results show that the uncertainty associated with the reference dataset can be very large and in some instances can become the first or second largest source of uncertainty. Using simple examples, we show that the resulting differences could lead to different conclusions with respect to some adaptation solutions or possibly create confusion with users. These results raise questions on the robustness of climate projections distributed through these web platforms and the ethical responsibility of data providers to adequately evaluate and communicate the underlying uncertainty.