A nesting model for bias correction of variability at multiple time scales in general circulation model precipitation simulations

Abstract

[1] Climate change impact assessments of water resources systems require simulations of precipitation and evaporation that exhibit distributional and persistence attributes similar to the historical record. Specifically, there is a need to ensure general circulation model (GCM) simulations of rainfall for the current climate exhibit low-frequency variability that is consistent with observed data. Inability to represent low-frequency variability in precipitation and flow leads to biased estimates of the security offered by water resources systems in a warmer climate. This paper presents a method to postprocess GCM precipitation simulations by imparting correct distributional and persistence attributes, resulting in sequences that are representative of observed records across a range of time scales. The proposed approach is named nesting bias correction (NBC), the rationale being to correct distributional and persistence bias from fine to progressively longer time scales. In the results presented here, distributional attributes have been represented by order 1 and 2 moments with persistence represented by lag 1 autocorrelation coefficients at monthly and annual time scales. The NBC method was applied to the Commonwealth Scientific and Industrial Research Organisation (CSIRO) Mk3.5 and MIROC 3.2 hires rainfall simulations for Australia. It was found that the nesting method worked well to correct means, standard deviations, and lag 1 autocorrelations when the biases in the raw GCM outputs were not too large. While the bias correction improves the representation of distributional and persistence attributes at the time scales considered, there is room for representation of longer-term persistence by extending to time scales longer than a year.

Multivariate bias adjustment of high-dimensional climate simulations: the Rank Resampling for Distributions and Dependences (R<sup>2</sup>D<sup>2</sup>) bias correction

“…If this is indirect for most of them (i.e., when accounting only for inter-site or inter-variable structures), some authors tried specifically to tackle the question of the temporal properties adjustment, such as Johnson and Sharma (2012) with a nesting 1d-BC model working across multiple timescales, Mehrotra and Sharma (2015) including inter-site dependence or Mehrotra and Sharma (2016) including multiple meteorological variables. However, no general comparison of the pros and cons of the two approaches has been performed and any BC method for both inter-site, inter-variable and temporal properties will necessarily consist of a trade-off between the temporal modifications brought by the multivariate adjustment and the correction of the temporal aspects, while respecting their changes from one time period to another.…”

HESS Opinions "Should we apply bias correction to global and regional climate model data?"

“…In recent years, BC methods have evolved from timeaveraged corrections of mean precipitation and temperature towards more advanced methods that correct higher distribution moments (Piani et al, 2010), include further variables such as radiation, humidity and wind (Haddeland et al, 2012), allow for time-dependent model biases (Buser et al, 2009;Li et al, 2010) or correct model output hierarchically on several nested time scales Johnson and Sharma, 2012).…”

“…This is well known and has been recognized by many authors, e.g. Wilby et al (2000), Wood et al (2004), Randall et al (2007), Piani et al (2010), Hagemann et al (2011, Rojas et al (2011), Haddeland et al (2012), Johnson and Sharma (2012). To overcome this problem, post-processing of either GCM or RCM output by correcting with and towards observations has become a standard procedure in climate change impact studies (CCIS).…”

“…Johnson and Sharma (2012) compared raw output from a GCM (CSIRO Mk3.5) and RCM (MIROC) with observations: in interior Australia, both models over-predicted annual rainfall by up to 200 %, but underpredicted along the coasts. Rojas et al (2011) found that averaged annual precipitation simulated by the HIRHAM 5 RCM over Europe in the control period 1961-2000 almost doubled the observed measurements.…”

“…In line with this, Johnson and Sharma (2012) list a number of reasons that make BC attractive: ease of application, ability to allow future changes in variability (unlike scaling methods), and flexibility to correct the GCM simulations for the parameters of interest. As another advantage, Li et al (2010) mention the lower computational requirements compared to alternatives based on dynamical models.…”