Uni- and multivariate bias-adjustment on a 1 km grid over Norway

The national climate assessment report series ‘Klima i Norge – 2100’ by the Norwegian Centre for Climate Services provides information on past and future changes in meteorological parameters and derived indices, hydrological components and effects on natural hazards. As a consequence, one of the key-user of national weather and climate data in downstream applications, namely hydrologists, are directly involved in the process of framing the report, including the selection of simulations and the application of bias-adjustment methods.

For the upcoming update of the report (expected to be published in 2024), a set of nine variables from an ensemble of regional climate model (RCM) projections will be bias-adjusted on a 1x1 km grid covering the complete Norwegian mainland. To this end, different methods have been implemented, including empirical quantile mapping, which has already been used in the former reports, quantile delta mapping and multivariate bias-adjustment. Applying the methods to a set of RCMs yields a variety of datasets. These datasets, by design of the methods, reproduce the reference data with various accuracy, for instance in terms of spatial, temporal and inter-variable consistencies. Furthermore, the bias-adjusted climate projections differ on how they inherit specific climate change signals from the RCMs. Compared to the raw data, they may or may not preserve monthly or seasonal trends, changes in the quantiles or in the dependency structures. Since it is not always clear whether all the changes in the RCMs are physically plausible and relevant, selecting a single method may not be appropriate. And even if it’s clear, the corresponding bias-adjustment method may have other undesirable shortcomings.

In this presentation we will show examples of how the different methods can affect the projected climate change signals in various aspects, and thus are adding a level of uncertainty. We will further emphasise the point of investigating these uncertainties in climate change assessments when bias-adjustment is involved. By feeding the data into a downstream application, the projected changes may differ substantially.