Are the Projections of Future Climate Change Reliable in the IPCC Reports?

As we know, the projections of future climate change including impacts and strategies in the IPCC Assessment Reports were based on global climate models with scenarios on various human activities. Global climate model simulations provide key inputs for climate change assessments. In this study, the main objective is to analyze if the projections of future climate change by global climate models are reliable. Several workshops have been held on this issue, such as the IPCC expert meeting on assessing and combining multi-model climate projections in January of 2010 (presided by the co-chairs of the IPCC WGI and WGII AR5), and the workshop of the combined global climate model group held by NCAR in June of 2010. Many modeling experts and users took part in those workshops. These series of workshops and conferences on climate models investigated several key issues on advances in determining the reliability of future climate change projections. The description and analysis of the workshop results are intended as guidance for future climate scientists and for impact and adaptation studies by using the results from model inter-comparison projects on global and regional climate change.

1. Has the time of applying the “one model, one vote” approach ended?

Since the IPCC First Assessment Report, increasingly more global climate models are used in the IPCC model inter-comparisons and future climate change projections. The future climate changes projected by those climate models have obvious uncertainties, especially at regional scales. Therefore, based on both synoptic prediction and the evaluation of replicated climate change for the 20th century by multiple models, the projections from multi-model ensembles have a certain capability to overcome some weaknesses of single climate model projections. The multi-model ensembles in the first four IPCC Assessment Reports employed the averaged values of all models. This means that the “one model, one vote (it is a visual description of un-weighted ensembles by the model groups)” approach was used, which did not consider the model’s quality or “uncertainty”. Despite this fact, models differ in terms of horizontal and vertical resolutions, physical, chemical and biological processes and parameterizations, as well as natural and anthropogenic forcing included. In particular, some models were developed from other models, wherefore, they are not independent.

Now, the upcoming IPCC Fifth Assessment Report (AR5) is being compiled. It is expected that more climate models will be incorporated and linked for the inter-comparison of models and climate change projections than in the AR4 (23 models). Among them, model groups will provide various projections of several model versions (e.g., the climate change projections of eight NCAR model versions). Based on the above mentioned reasons, most model experts suggested that the climate change projections estimated by the “one model, one vote” approach (i.e., multimodel mean-un-weighted) should be ceased, and that the multi-model weighted projections and probability climate predictions are integrated in the AR5. But, as some experts argue that the results by the multimodel un-weighted means were not worse than the multi-model weighted means, according to the IPCC AR4 multi-model simulations of climate change for the 20th century, the arguments should be studied and discussed within the AR5.

2. How to measure a climate model’s quality or “goodness”? How to rank numerous models?

The key questions on multiple model weighted ensembles are “how to measure a climate model’s quality or ‘goodness’”, and “how to rank the numerous models”. To solve these problems, a metric for model performance relevant for climate prediction has to be determined. Summarizing several workshops, model quality metric and index were determined as follows: (1) in general, a fine resolution and process related approaches to the global climate system are required; (2) models simulated significant climate variables (e.g., climate mean state distribution and patterns, time variations, and extreme events) and major processes (e.g., ENSO, monsoon, decadal and inter-decadal variability, and thermohaline circulation), they were more closer to the observations (the observed characteristics in the 20th century or the key features in the paleoclimatic period); (3) testing the models accuracy on climate hindcasts and predictions for time-scales of 10–20 years; (4) judging the model’s independence, i.e., similarity of model dynamic structures, the physical, chemical and biological process parameterizations; (5) a list of scientific references and papers published for each model, especially in high quality journals.

Among those metrics, both (2) and (3) are the main issues. The detailed, operable and idiographic metrics and indexes should be constituted for intercomparisons. At the same time, while the assumption that a model is good for the past is needed, it should also continue to be good in future climate projections. The relationship between the weighted ensembles and the numbers of different climate models should be investigated. Various difficulties exist on the ranking of the different models. Such difficulties depend mainly on the climatic variables selected and the scale of both time and space, as well as how to synthesis those metrics.

3. How to project future climate change by multi-model ensembles?

The central issue of the IPCC Assessment Reports is to predict and project the global and regional climate change for the next 10–20 years in the 21st century. The findings are linked to climate impacts and strategies on international and national levels. The aforementioned workshops recommended: (1) multimodel weighted ensembles should replace “one model, one vote” approach in the AR5 based on the ranking and selection of models; (2) encouraging experts to investigate the theory, methodology and skills of multimodel weighted ensembles: in particular, the main concerns are the number of model ensembles for optimal schemes, the measuring of these optimal schemes and mathematic schemes and more; and (3) improve the provision of the results of projections of future climate changes. At least, the best projection results, probability distribution, ranges of reliability, scattering extents of predictions and projections by the multiple models, and the quantity ranges of uncertainties should be given.

Taking this into account, it is expected that more reliable projections of future climate changes will be provided by the IPCC AR5, while scientists should pay more attention to multi-model ensembles.