scholarly journals Cross-validation of bias-corrected climate simulations is misleading

2018 ◽  
Vol 22 (9) ◽  
pp. 4867-4873 ◽  
Author(s):  
Douglas Maraun ◽  
Martin Widmann
Keyword(s):  
Climate Change ◽  
Bias Correction ◽  
Cross Validation ◽  
Climate Model ◽  
False Negative ◽  
Internal Variability ◽  
Validation Period ◽  
Climate Simulations ◽  
Free Running ◽  
The Difference

Abstract. We demonstrate both analytically and with a modelling example that cross-validation of free-running bias-corrected climate change simulations against observations is misleading. The underlying reasoning is as follows: a cross-validation can have in principle two outcomes. A negative (in the sense of not rejecting a null hypothesis), if the residual bias in the validation period after bias correction vanishes; and a positive, if the residual bias in the validation period after bias correction is large. It can be shown analytically that the residual bias depends solely on the difference between the simulated and observed change between calibration and validation periods. This change, however, depends mainly on the realizations of internal variability in the observations and climate model. As a consequence, the outcome of a cross-validation is also dominated by internal variability, and does not allow for any conclusion about the sensibility of a bias correction. In particular, a sensible bias correction may be rejected (false positive) and a non-sensible bias correction may be accepted (false negative). We therefore propose to avoid cross-validation when evaluating bias correction of free-running bias-corrected climate change simulations against observations. Instead, one should evaluate non-calibrated temporal, spatial and process-based aspects.

scholarly journals Cross-validation of bias-corrected climate simulations is misleading

2018 ◽  
Author(s):  
Douglas Maraun ◽  
Martin Widmann
Keyword(s):  
Climate Change ◽  
Bias Correction ◽  
Cross Validation ◽  
Climate Model ◽  
False Negative ◽  
Internal Variability ◽  
Validation Period ◽  
Climate Simulations ◽  
Free Running ◽  
The Difference

Abstract. We demonstrate both analytically and with a modelling example that cross-validation of free running bias-corrected climate change simulations against observations is misleading. The underlying reasoning is as follows: a cross-validation can have in principle two outcomes. A negative (in the sense of not rejecting a Null hypothesis), if the residual bias in the validation period after bias correction vanishes; and a positive, if the residual bias in the validation period after bias correction is large. It can be shown analytically that the residual bias depends solely on the difference between the simulated and observed change between calibration and validation period. These changes, however, depend mainly on the realisations of internal variability in the observations and climate model. As a consequence, also the outcome of a cross-validation is dominated by internal variability, and does not allow for any conclusion about the sensibility of a bias correction. In particular, a sensible bias correction may be rejected (false positive) and a non-sensible bias correction may be accepted (false negative). We therefore propose to avoid cross-validation when evaluating bias correction of free running bias-corrected climate change simulations against observations. Instead, one should evaluate temporal, spatial and process-based aspects.

scholarly journals Ensemble bias correction of climate simulations: preserving internal variability

2021 ◽  
Vol 11 (1) ◽  
Author(s):  
Pradeebane Vaittinada Ayar ◽  
Mathieu Vrac ◽  
Alain Mailhot
Keyword(s):  
Climate Change ◽  
Bias Correction ◽  
Regional Climate ◽  
Climate Change Impacts ◽  
Internal Variability ◽  
Ensemble Member ◽  
Climate Change Signal ◽  
Changing Climate ◽  
Uncertainty Model ◽  
Climate Simulations

AbstractClimate simulations often need to be adjusted (i.e., corrected) before any climate change impacts studies. However usual bias correction approaches do not differentiate the bias from the different uncertainties of the climate simulations: scenario uncertainty, model uncertainty and internal variability. In particular, in the case of a multi-run ensemble of simulations (i.e., multiple runs of one model), correcting, as usual, each member separately, would mix up the model biases with its internal variability. In this study, two ensemble bias correction approaches preserving the internal variability of the initial ensemble are proposed. These “Ensemble bias correction” (EnsBC) approaches are assessed and compared to the approach where each ensemble member is corrected separately, using precipitation and temperature series at two locations in North America from a multi-member regional climate ensemble. The preservation of the internal variability is assessed in terms of monthly mean and hourly quantiles. Besides, the preservation of the internal variability in a changing climate is evaluated. Results show that, contrary to the usual approach, the proposed ensemble bias correction approaches adequately preserve the internal variability even in changing climate. Moreover, the climate change signal given by the original ensemble is also conserved by both approaches.

scholarly journals Multivariate bias corrections of climate simulations: which benefits for which losses?

2020 ◽  
Vol 11 (2) ◽  
pp. 537-562 ◽  
Author(s):  
Bastien François ◽  
Mathieu Vrac ◽  
Alex J. Cannon ◽  
Yoann Robin ◽  
Denis Allard
Keyword(s):  
Bias Correction ◽  
Cross Validation ◽  
Climate Models ◽  
Climate Model ◽  
Temporal Structure ◽  
Climate Projections ◽  
Validation Method ◽  
Climate Simulations ◽  
Temporal Structures ◽  
Temporal Dependencies

Abstract. Climate models are the major tools to study the climate system and its evolutions in the future. However, climate simulations often present statistical biases and have to be corrected against observations before being used in impact assessments. Several bias correction (BC) methods have therefore been developed in the literature over the last 2 decades, in order to adjust simulations according to historical records and obtain climate projections with appropriate statistical attributes. Most of the existing and popular BC methods are univariate, i.e., correcting one physical variable and one location at a time and, thus, can fail to reconstruct inter-variable, spatial or temporal dependencies of the observations. These remaining biases in the correction can then affect the subsequent analyses. This has led to further research on multivariate aspects for statistical postprocessing BC methods. Recently, some multivariate bias correction (MBC) methods have been proposed, with different approaches to restore multidimensional dependencies. However, these methods are not yet fully apprehended by researchers and practitioners due to differences in their applicability and assumptions, therefore leading potentially to different results. This study is intended to intercompare four existing MBCs to provide end users with aid in choosing such methods for their applications. For evaluation and illustration purposes, these methods are applied to correct simulation outputs from one climate model through a cross-validation method, which allows for the assessment of inter-variable, spatial and temporal criteria. Then, a second cross-validation method is performed for assessing the ability of the MBC methods to account for the multidimensional evolutions of the climate model. Additionally, two reference datasets are used to assess the influence of their spatial resolution on (M)BC results. Most of the methods reasonably correct inter-variable and intersite correlations. However, none of them adjust correctly the temporal structure as they generate bias-corrected data with usually weak temporal dependencies compared to observations. Major differences are found concerning the applicability and stability of the methods in high-dimensional contexts and in their capability to reproduce the multidimensional changes in the model. Based on these conclusions, perspectives for MBC developments are suggested, such as methods to adjust not only multivariate correlations but also temporal structures and allowing multidimensional evolutions of the model to be accounted for in the correction.

scholarly journals A Bayesian posterior predictive framework for weighting ensemble regional climate models

2017 ◽  
Author(s):  
Yanan Fan ◽  
Roman Olson ◽  
Jason P. Evans
Keyword(s):  
Climate Change ◽  
Cross Validation ◽  
Statistical Approach ◽  
Climate Models ◽  
Climate Model ◽  
Regional Climate ◽  
Regional Climate Models ◽  
Internal Variability ◽  
Model Bias ◽  
Computing Model

Abstract. We present a novel Bayesian statistical approach to computing model weights in climate change We present a novel Bayesian statistical approach to computing model weights in climate change projection ensembles. The weight of each climate model is obtained by weighting the current day observed data under the posterior distribution admitted under competing climate models. We use a linear model to describe the model output and observations. The approach accounts for uncertainty in model bias, trend and internal variability, as well as including error in the observations used. Our framework is general, requires very little problem specific input, and works well with default priors. We carry out cross-validation checks that confirm that the method produces the correct coverage.

Bias correction of extreme values of high resolution climate simulations for risk analysis.

2021 ◽  
Author(s):  
luis Augusto sanabria ◽  
Xuerong Qin ◽  
Jin Li ◽  
Robert Peter Cechet
Keyword(s):  
Climate Change ◽  
Bias Correction ◽  
Extreme Values ◽  
The Body ◽  
Future Climate ◽  
Return Periods ◽  
Climate Conditions ◽  
Mitigation And Adaptation ◽  
Climate Simulations ◽  
The Impact

Abstract Most climatic models show that climate change affects natural perils' frequency and severity. Quantifying the impact of future climate conditions on natural hazard is essential for mitigation and adaptation planning. One crucial factor to consider when using climate simulations projections is the inherent systematic differences (bias) of the modelled data compared with observations. This bias can originate from the modelling process, the techniques used for downscaling of results, and the ensembles' intrinsic variability. Analysis of climate simulations has shown that the biases associated with these data types can be significant. Hence, it is often necessary to correct the bias before the data can be reliably used for further analysis. Natural perils are often associated with extreme climatic conditions. Analysing trends in the tail end of distributions are already complicated because noise is much more prominent than that in the mean climate. The bias of the simulations can introduce significant errors in practical applications. In this paper, we present a methodology for bias correction of climate simulated data. The technique corrects the bias in both the body and the tail of the distribution (extreme values). As an illustration, maps of the 50 and 100-year Return Period of climate simulated Forest Fire Danger Index (FFDI) in Australia are presented and compared against the corresponding observation-based maps. The results show that the algorithm can substantially improve the calculation of simulation-based Return Periods. Forthcoming work will focus on the impact of climate change on these Return Periods considering future climate conditions.

scholarly journals 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

2018 ◽  
Vol 22 (6) ◽  
pp. 3175-3196 ◽  
Author(s):  
Mathieu Vrac
Keyword(s):  
Bias Correction ◽  
Climate Model ◽  
Regional Climate ◽  
Grid Cell ◽  
Dependence Structure ◽  
Historical Period ◽  
Temporal Properties ◽  
Temperature And Precipitation ◽  
Climate Simulations ◽  
Site Dependence

Abstract. Climate simulations often suffer from statistical biases with respect to observations or reanalyses. It is therefore common to correct (or adjust) those simulations before using them as inputs into impact models. However, most bias correction (BC) methods are univariate and so do not account for the statistical dependences linking the different locations and/or physical variables of interest. In addition, they are often deterministic, and stochasticity is frequently needed to investigate climate uncertainty and to add constrained randomness to climate simulations that do not possess a realistic variability. This study presents a multivariate method of rank resampling for distributions and dependences (R2D2) bias correction allowing one to adjust not only the univariate distributions but also their inter-variable and inter-site dependence structures. Moreover, the proposed R2D2 method provides some stochasticity since it can generate as many multivariate corrected outputs as the number of statistical dimensions (i.e., number of grid cell  ×  number of climate variables) of the simulations to be corrected. It is based on an assumption of stability in time of the dependence structure – making it possible to deal with a high number of statistical dimensions – that lets the climate model drive the temporal properties and their changes in time. R2D2 is applied on temperature and precipitation reanalysis time series with respect to high-resolution reference data over the southeast of France (1506 grid cell). Bivariate, 1506-dimensional and 3012-dimensional versions of R2D2 are tested over a historical period and compared to a univariate BC. How the different BC methods behave in a climate change context is also illustrated with an application to regional climate simulations over the 2071–2100 period. The results indicate that the 1d-BC basically reproduces the climate model multivariate properties, 2d-R2D2 is only satisfying in the inter-variable context, 1506d-R2D2 strongly improves inter-site properties and 3012d-R2D2 is able to account for both. Applications of the proposed R2D2 method to various climate datasets are relevant for many impact studies. The perspectives of improvements are numerous, such as introducing stochasticity in the dependence itself, questioning its stability assumption, and accounting for temporal properties adjustment while including more physics in the adjustment procedures.

Climate Model Biases and Modification of the Climate Change Signal by Intensity-Dependent Bias Correction

2018 ◽  
Vol 31 (16) ◽  
pp. 6591-6610 ◽  
Author(s):  
Martin Aleksandrov Ivanov ◽  
Jürg Luterbacher ◽  
Sven Kotlarski
Keyword(s):  
Climate Change ◽  
Bias Correction ◽  
Climate Models ◽  
Climate Model ◽  
Daily Precipitation ◽  
Analytical Description ◽  
Analytical Theory ◽  
Climate Change Signal ◽  
Future Research ◽  
Model Biases

Climate change impact research and risk assessment require accurate estimates of the climate change signal (CCS). Raw climate model data include systematic biases that affect the CCS of high-impact variables such as daily precipitation and wind speed. This paper presents a novel, general, and extensible analytical theory of the effect of these biases on the CCS of the distribution mean and quantiles. The theory reveals that misrepresented model intensities and probability of nonzero (positive) events have the potential to distort raw model CCS estimates. We test the analytical description in a challenging application of bias correction and downscaling to daily precipitation over alpine terrain, where the output of 15 regional climate models (RCMs) is reduced to local weather stations. The theoretically predicted CCS modification well approximates the modification by the bias correction method, even for the station–RCM combinations with the largest absolute modifications. These results demonstrate that the CCS modification by bias correction is a direct consequence of removing model biases. Therefore, provided that application of intensity-dependent bias correction is scientifically appropriate, the CCS modification should be a desirable effect. The analytical theory can be used as a tool to 1) detect model biases with high potential to distort the CCS and 2) efficiently generate novel, improved CCS datasets. The latter are highly relevant for the development of appropriate climate change adaptation, mitigation, and resilience strategies. Future research needs to focus on developing process-based bias corrections that depend on simulated intensities rather than preserving the raw model CCS.

scholarly journals Impact of internal variability on climate change for the upcoming decades: analysis of the CanESM2-LE and CESM-LE large ensembles

2019 ◽  
Vol 156 (3) ◽  
pp. 299-314 ◽  
Author(s):  
Gabriel Rondeau-Genesse ◽  
Marco Braun
Keyword(s):  
Climate Change ◽  
Climate Models ◽  
Climate Model ◽  
Ghg Emissions ◽  
Internal Variability ◽  
Mean Annual Temperature ◽  
Large Ensemble ◽  
Climate Change Effects ◽  
Large Ensembles ◽  
The Mean

Abstract The pace of climate change can have a direct impact on the efforts required to adapt. For short timescales, however, this pace can be masked by internal variability (IV). Over a few decades, this can cause climate change effects to exceed what would be expected from the greenhouse gas (GHG) emissions alone or, to the contrary, cause slowdowns or even hiatuses. This phenomenon is difficult to explore using ensembles such as CMIP5, which are composed of multiple climate models and thus combine both IV and inter-model differences. This study instead uses CanESM2-LE and CESM-LE, two state-of-the-art large ensembles (LE) that comprise multiple realizations from a single climate model and a single GHG emission scenario, to quantify the relationship between IV and climate change over the next decades in Canada and the USA. The mean annual temperature and the 3-day maximum and minimum temperatures are assessed. Results indicate that under the RCP8.5, temperatures within most of the individual large ensemble members will increase in a roughly linear manner between 2021 and 2060. However, members of the large ensembles in which a slowdown of warming is found during the 2021–2040 period are two to five times more likely to experience a period of very fast warming in the following decades. The opposite scenario, where the changes expected by 2050 would occur early because of IV, remains fairly uncommon for the mean annual temperature, but occurs in 5 to 15% of the large ensemble members for the temperature extremes.

scholarly journals Selection of Bias Correction Methods to Assess the Impact of Climate Change on Flood Frequency Curves

Water ◽  
2019 ◽  
Vol 11 (11) ◽  
pp. 2266 ◽  
Author(s):  
Enrique Soriano ◽  
Luis Mediero ◽  
Carlos Garijo
Keyword(s):  
Climate Change ◽  
Bias Correction ◽  
Climate Models ◽  
Climate Model ◽  
Control Period ◽  
Flood Frequency ◽  
Climate Projections ◽  
The Future ◽  
Flood Quantiles ◽  
The Impact

Climate projections provided by EURO-CORDEX predict changes in annual maximum series of daily rainfall in the future in some areas of Spain because of climate change. Precipitation and temperature projections supplied by climate models do not usually fit exactly the statistical properties of the observed time series in the control period. Bias correction methods are used to reduce such errors. This paper seeks to find the most adequate bias correction techniques for temperature and precipitation projections that minimizes the errors between observations and climate model simulations in the control period. Errors in flood quantiles are considered to identify the best bias correction techniques, as flood quantiles are used for hydraulic infrastructure design and safety assessment. In addition, this study aims to understand how the expected changes in precipitation extremes and temperature will affect the catchment response in flood events in the future. Hydrological modelling is required to characterize rainfall-runoff processes adequately in a changing climate, in order to estimate flood changes expected in the future. Four catchments located in the central-western part of Spain have been selected as case studies. The HBV hydrological model has been calibrated in the four catchments by using the observed precipitation, temperature and streamflow data available on a daily scale. Rainfall has been identified as the most significant input to the model, in terms of its influence on flood response. The quantile mapping polynomial correction has been found to be the best bias correction method for precipitation. A general reduction in flood quantiles is expected in the future, smoothing the increases identified in precipitation quantiles by the reduction of soil moisture content in catchments, due to the expected increase in temperature and decrease in mean annual precipitations.

Sign in / Sign up

Export Citation Format

Share Document