Errors in climate model daily precipitation and temperature output: time invariance and implications for bias correction

Abstract

Abstract. When correcting for biases in general circulation model (GCM) output, for example when statistically downscaling for regional and local impacts studies, a common assumption is that the GCM biases can be characterized by comparing model simulations and observations for a historical period. We demonstrate some complications in this assumption, with GCM biases varying between mean and extreme values and for different sets of historical years. Daily precipitation and maximum and minimum temperature from late 20th century simulations by four GCMs over the United States were compared to gridded observations. Using random years from the historical record we select a "base" set and a 10 yr independent "projected" set. We compare differences in biases between these sets at median and extreme percentiles. On average a base set with as few as 4 randomly-selected years is often adequate to characterize the biases in daily GCM precipitation and temperature, at both median and extreme values; 12 yr provided higher confidence that bias correction would be successful. This suggests that some of the GCM bias is time invariant. When characterizing bias with a set of consecutive years, the set must be long enough to accommodate regional low frequency variability, since the bias also exhibits this variability. Newer climate models included in the Intergovernmental Panel on Climate Change fifth assessment will allow extending this study for a longer observational period and to finer scales.

Scaled distribution mapping: a bias correction method that preserves raw climate model projected changes

“…Independent of downscaling, bias correction with QM is well known to alter the raw modeled climate change signal (Hagemann et al, 2011;Themeßl et al, 2012;Brekke et al, 2013;Maurer et al, 2013;Pierce et al, 2013;Maurer and Pierce, 2014). This alteration of the raw modeled climate change signal can be attributed to the stationarity assumption, which implies that the error correction values established in a calibration period can be applied to any time period within or outside the calibration time period.…”

Technical Note: The impact of spatial scale in bias correction of climate model output for hydrologic impact studies

“…Figure 3 demonstrates that, as will always be the case due to natural variability, the biases between climate model output (or reanalyses) and observations will be different for different time periods. It is also evident, for the precipitation statistic depicted, that the difference in bias between the two periods is much smaller than the bias itself, explaining why bias correction generally does improve skill, especially given the role of topography in precipitation formation and the lack of detailed topographic representation in the large-scale reanalysis data (e.g., Maurer et al, 2013). Comparing the change in bias between the two periods at different spatial scales (each row of the right column), Fig.…”

“…The probability transformations in quantile mapping are incapable of correcting for GCM biases in low-frequency variability, and autoregressive and spectral transformations have been developed to accommodate these biases where important (Mehrotra and Sharma, 2012;Pierce et al, 2015). While we recognize the deficiencies in quantile mapping, as discussed for statistical bias correction in general by Ehret et al (2012), and there is the promise of recent advances in bias correction, it remains that quantile mapping is widely used and generally effective at removing biases (Gudmundsson et al, 2012), even in the presence of some non-stationarity (Lafon et al, 2012;Maurer et al, 2013;Teutschbein and Seibert, 2013). Our aim in this study is not to advocate for a specific downscaling method, but to understand a specific aspect of this widely used method.…”

“…Quantile mapping is effective at removing some climate model biases, is relatively simple to apply, and has been incorporated into many statistical downscaling schemes used for local and regional impacts analysis (Li et al, 2010;Maraun et al, 2010;Panofsky and Brier, 1968;Piani et al, 2010;Themeßl et al, 2011). While quantile-mapping bias correction does inherently assume that the biases exhibited by a climate model remain constant in future projections, there is some indication that this is not an unreasonable assumption (Maraun, 2012;Maurer et al, 2013), especially where biases are driven by persistent climate model characteristics, such as inadequate representation of topography. Other discrepancies between historic climate model simulations and observations, especially due to internal natural variability (for example, El Niño events simulated by a freely evolving GCM not coinciding with observations), are not necessarily model biases (Eden et al, 2012), but are corrected nonetheless by quantile mapping, which is blind to the source of the bias.…”

Statistical bias correction for climate change impact on the basin scale precipitation in Sri Lanka, Philippines, Japan and Tunisia

“…Average GCM bias is statistically the same between two sets of years (Maurer et al, 2013), and we assumed constant model biases in the projected period. To ensure the adequacy of the stationary bias assumption over decadal timescales and boost reliability in the future projection, we applied the bias correction method to a different set of years.…”