Spatial Analysis to Quantify Numerical Model Bias and Dependence

Jun, Mikyoung; Knutti, Reto; Nychka, Douglas

doi:10.1198/016214507000001265

Cited by 199 publications

(189 citation statements)

References 1 publication

Supporting

Mentioning

179

Contrasting

Order By: Relevance

“…One explanation is that several institutions have contributed a set of two or three climate models, have shared parts of code, and input datasets and expertise of those developing the GCMs, so the resulting parts of the model bias may be similar in some or all models (Jun et al 2008;Knutti et al 2010). Therefore, determining underlying reasons for high/low SSs with the GCM model characteristics that are responsible for the biases may not be possible.…”

Section: Discussionmentioning

confidence: 99%

Performance evaluation of AR4 Climate Models in simulating daily precipitation over the Indian region using skill scores

Anandhi

Nanjundiah

2014

Theor Appl Climatol

View full text Add to dashboard Cite

The ability of Coupled General Circulation Models (CGCMs) participating in the Intergovernmental Panel for Climate Change's fourth assessment report (IPCC AR4) for the 20th century climate (20C3M scenario) to simulate the daily precipitation over the Indian region is explored. The skill is evaluated on a 2.5°×2.5°grid square compared with the Indian Meteorological Department's (IMD) gridded dataset, and every GCM is ranked for each of these grids based on its skill score. Skill scores (SSs) are estimated from the probability density functions (PDFs) obtained from observed IMD datasets and GCM simulations. The methodology takes into account (high) extreme precipitation events simulated by GCMs. The results are analyzed and presented for three categories and six zones. The three categories are the monsoon season (JJASO -June to October), non-monsoon season (JFMAMND -January to May, November, December) and for the entire year ("Annual"). The six precipitation zones are peninsular, west central, northwest, northeast, central northeast India, and the hilly region. Sensitivity analysis was performed for three spatial scales, 2.5°grid square, zones, and all of India, in the three categories. The models were ranked based on the SS. The category JFMAMND had a higher SS than the JJASO category. The northwest zone had higher SSs, whereas the peninsular and hilly regions had lower SS. No single GCM can be identified as the best for all categories and zones. Some models consistently outperformed the model ensemble, and one model had particularly poor performance. Results show that most models underestimated the daily precipitation rates in the 0-1 mm/day range and overestimated it in the 1-15 mm/day range.

show abstract

Section: Discussionmentioning

confidence: 99%

Performance evaluation of AR4 Climate Models in simulating daily precipitation over the Indian region using skill scores

Anandhi

Nanjundiah

2014

Theor Appl Climatol

View full text Add to dashboard Cite

show abstract

“…In our analysis, high correlations are found in some variable metrics, but the physical reasoning has not been uncovered. Previous studies have pointed out that the CMIP3 models are not mutually independent and their e¤ective number is only between five and ten (Jun et al 2008a(Jun et al , 2008bPennell and Reichler 2010). The estimation of the e¤ective model number by using PCA is equivalent to that of the number of e¤ective metrics or measures of inter-model similarity, since the numbers of nonzero eigenvalues of inter-model and inter-variable covariance matrices of C are identical.…”

Section: Discussionmentioning

confidence: 99%

“…Even if model biases were distributed randomly, the number of available models would be unlikely su‰cient for their perfect cancellation (e.g., Knutti et al 2010). In fact, the e¤ective number (or degrees of freedom: DOFs) of the CMIP3 models has been estimated to be only between five and ten (Jun et al 2008a(Jun et al , 2008bKnutti et al 2010;Pennell and Reichler 2010). In other words, the amount of information provided as an ensemble of those models may be less than what would be expected under the assumption that all the models were mutually independent1.…”

Section: Introductionmentioning

confidence: 99%

Relationship of the Reproducibility of Multiple Variables among Global Climate Models

Nishii

Miyasaka

Nakamura

et al. 2012

Journal of the Meteorological Society of Japan

View full text Add to dashboard Cite

Numerous e¤orts have been made for evaluating the performance of global climate models with such expectation that those models with higher reproducibility of the current climate should provide more reliable projections of climate changes into the future. Attempts have been made to define a single general metric through which the overall performance of a global climate model can be assessed. On the basis of general metrics defined through several techniques of multivariate analysis, the present study compares global climate models from a viewpoint of their reproducibility of climatological-mean fields of multiple variables. The analyses indicate that a reproducibility of a particular variable is not necessarily independent of that of others, which may bring redundant information into a general metric. The model reproducibility in upper and mid-tropospheric temperature and lower-tropospheric humidity, for example, tends to be anti-correlated with that in upper and mid-tropospheric humidity. It is argued that attention has to be paid to this kind of trade-o¤ relationships among some variables and resultant redundancy in synthesizing multiple metrics. A possibility is suggested that an arbitrary selection of variables can yield some redundant information of variables. The redundancy is, however, found to exert no serious influence on the quality of a general metric as long as it is based on the su‰cient number of variables. In our attempt to evaluate the climate models by introducing general performance metrics with reduced redundancy of variables, the overall model ranking is found rather insensitive to the specific definition of the metric.

show abstract

“…The use of a global criterion and consequently of a single predictor for all domain D does not acknowledge that some predictors are locally better than others, but it does allow a reconstructed map that retains more physical interpretability to be chosen. A similar issue arises in climate science when we need to determine what is the "best" global or regional circulation model in a multi-model ensemble, and how to combine information to gain better predictions (see for example Jun et al 2008b, a;Knutti 2010;Tebaldi et al 2004Tebaldi et al , 2005. In that context, the field output, be it any physical variable, represents the solution of systems of partial differential equations (PDEs): it therefore has an important physical meaning.…”

Section: Interpretability Of Spatial Predictors Comparisonsmentioning

confidence: 99%