nom_organisme nom_organisme nom_organisme nom_organisme
Downscaling & Bias correction of climate simulations
Downscaling & Bias correction of climate simulations

Figure: Conceptual view on downscaling. The goal is to generate high-resolution simulations, based on lower-resolution data (coming from reanalyses or global climate simulations)

What is the purpose of Bias Correction and Downscaling?


Global Climate Models (GCMs) are the primary tools to simulate multi-decadal climate dynamics and to generate and understand global climate change projections under different future emission scenarios. However, these models have a coarse spatial resolution (typically a few hundred kilometres) and suffer from substantial systematic biases when compared with observations. Therefore, they are unable to provide actionable information at the regional and local spatial scales required in impact and adaptation studies (e.g., Grouillet al., 2016). Hence, higher resolution simulations – and if possible bias corrected – are required for the most relevant climate variables (Vaittinada Ayar et al., 2015).

Downscaling attempts to resolve the scale discrepancy between climate change scenarios and the resolution required for impact assessment. Two main downscaling approaches have been developed since the early 1990s: Dynamical downscaling (based on Regional Climate Models, RCMs) and Statistical Downscaling Models (SDMs), which are nowadays recognized as complementary in many practical applications.

The ESTIMR team develops, evaluates and applies both approaches.

For the first time, high-resolution (12km), century-scale simulations over Europe using the WRF model have been carried out (Vautard et al., 2013; Jacob et al., 2014), and have been used in a series of subsequent studies for evaluation of the added value of high resolution against observations (e.g., Prein et al, 2016; Casanueva et al., 2015), for assessing climate change and impacts (e.g., Jacob et al., 2017; Jerez et al, 2015; Tobin et  al., 2018).

We have developed various SDMs and bias correction (BC) methods, even combined sometimes (Volosciuk et al., 2017), which improves the downscaling results (Vrac and Vaittinada Ayar, 2016). The developed SDMs mostly belong to the “stochastic weather generators” (SWG) family (e.g., Eden et al, 2014) with some focuses on extremes (Wong et al., 2014) and their spatial modelling (Bechler et al., 2015).

Key efforts have been also put on specific BC developments for precipitation (Vrac et al., 2016) and for multivariate (i.e., multi-sites and variables) approaches, both in downscaling and BC contexts (Vrac and Friederichs, 2015; Vrac, 2018).

Note that the CDF-t bias correction method, developed in ESTIMR, is being implemented to run operationally at the IPSL computing centre. This opens the way to automatic BC of the upcoming CMIP6 simulations.

Those various downscaling and bias correction approaches have been applied in different contexts: for example, to drive hydrological models in the Mediterranean basin (Grouillet et al., 2016); to provide future scenarios of precipitation and water availability over India (Vigaud et al., 2013); to investigate potential consequences of rapid ice-sheet melting on Sahelian population (Defrance et al., 2017); or even downscaling of paleao-climate simulations during last glacial maximum (Burke et al., 2014, 2017).

Moreover, within the StaRMIP project, a guidelines study was performed to evaluate both SDMs and RCMs simulations according to statistical properties required to drive impact models (Vaittinada Ayar et al., 2015). We also participated to the European VALUE intercomparison of SDMs, focusing on various aspects (e.g., cross-validation: Guttierez et al., 2018; extremes: Hertig et al., 2018). Moving from climate to weather, methodological advances concerning correction of weather forecasts have also been done with colleagues from Meteo France (Taillardart et al., 2016, Zamo and Naveau 2016).



References  for Downscaling

References  for Bias Corrections

#161 - Last update : 10/05 2022
Retour en haut