Search

Climate projections from global circulation models (GCMs), part of the Coupled Model Intercomparison Project 6 (CMIP6), are often employed to study the impact of future climate on ecosystems. However, especially at regional scales, climate projections display large biases in key forcing variables such as temperature and precipitation. These biases have been identified as a major source of uncertainty in carbon cycle projections, hampering predictive capacity. In this study, we open the proverbial Pandora's box and peer under the lid of strategies to tackle climate model ensemble uncertainty. We employ a dynamic global vegetation model (LPJ-GUESS) and force it with raw output from CMIP6 to assess the uncertainty associated with the choice of climate forcing. We then test different methods to either bias-correct or calculate ensemble averages over the original forcing data to reduce the climate-driven uncertainty in the regional projection of the Australian carbon cycle. We find that all bias correction methods reduce the bias of continental averages of steady-state carbon variables. Bias correction can improve model carbon outputs, but carbon pools are insensitive to the type of bias correction method applied for both individual GCMs and the arithmetic ensemble average across all corrected models. None of the bias correction methods consistently improve the change in simulated carbon over time compared to the target dataset, highlighting the need to account for temporal properties in correction or ensemble-averaging methods. Multivariate bias correction methods tend to reduce the uncertainty more than univariate approaches, although the overall magnitude is similar. Even after correcting the bias in the meteorological forcing dataset, the simulated vegetation distribution presents different patterns when different GCMs are used to drive LPJ-GUESS. Additionally, we found that both the weighted ensemble-averaging and random forest approach reduce the bias in total ecosystem carbon to almost zero, clearly outperforming the arithmetic ensemble-averaging method. The random forest approach also produces the results closest to the target dataset for the change in the total carbon pool, seasonal carbon fluxes, emphasizing that machine learning approaches are promising tools for future studies. This highlights that, where possible, an arithmetic ensemble average should be avoided. However, potential target datasets that would facilitate the application of machine learning approaches, i.e., that cover both the spatial and temporal domain required to derive a robust informed ensemble average, are sparse for ecosystem variables.

The rationale for using multi-model ensembles in climate change projections and impacts research is often based on the expectation that different models constitute independent estimates; therefore, a range of models allows a better characterisation of the uncertainties in the representation of the climate system than a single model. However, it is known that research groups share literature, ideas for representations of processes, parameterisations, evaluation data sets and even sections of model code. Thus, nominally different models might have similar biases because of similarities in the way they represent a subset of processes, or even be near-duplicates of others, weakening the assumption that they constitute independent estimates. If there are near-replicates of some models, then treating all models equally is likely to bias the inferences made using these ensembles. The challenge is to establish the degree to which this might be true for any given application. While this issue is recognised by many in the community, quantifying and accounting for model dependence in anything other than an ad-hoc way is challenging. Here we present a synthesis of the range of disparate attempts to define, quantify and address model dependence in multi-model climate ensembles in a common conceptual framework, and provide guidance on how users can test the efficacy of approaches that move beyond the equally weighted ensemble. In the upcoming Coupled Model Intercomparison Project phase 6 (CMIP6), several new models that are closely related to existing models are anticipated, as well as large ensembles from some models. We argue that quantitatively accounting for dependence in addition to model performance, and thoroughly testing the effectiveness of the approach used will be key to a sound interpretation of the CMIP ensembles in future scientific studies.

End users studying impacts and risks caused by human-induced climate change are often presented with large multi-model ensembles of climate projections whose composition and size are arbitrarily determined. An efficient and versatile method that finds a subset which maintains certain key properties from the full ensemble is needed, but very little work has been done in this area. Therefore, users typically make their own somewhat subjective subset choices and commonly use the equally weighted model mean as a best estimate. However, different climate model simulations cannot necessarily be regarded as independent estimates due to the presence of duplicated code and shared development history. Here, we present an efficient and flexible tool that makes better use of the ensemble as a whole by finding a subset with improved mean performance compared to the multi-model mean while at the same time maintaining the spread and addressing the problem of model interdependence. Out-of-sample skill and reliability are demonstrated using model-as-truth experiments. This approach is illustrated with one set of optimisation criteria but we also highlight the flexibility of cost functions, depending on the focus of different users. The technique is useful for a range of applications that, for example, minimise present-day bias to obtain an accurate ensemble mean, reduce dependence in ensemble spread, maximise future spread, ensure good performance of individual models in an ensemble, reduce the ensemble size while maintaining important ensemble characteristics, or optimise several of these at the same time. As in any calibration exercise, the final ensemble is sensitive to the metric, observational product, and pre-processing steps used.