Search

Evaluating the response of climate to greenhouse gas forcing is a major objective of the climate community, and the use of large ensemble of simulations is considered as a significant step toward that goal. The present paper thus discusses a new methodology based on neural network to mix ensemble of climate model simulations. Our analysis consists of one simulation of seven Atmosphere–Ocean Global Climate Models, which participated in the IPCC Project and provided at least one simulation for the twentieth century (20c3m) and one simulation for each of three SRES scenarios: A2, A1B and B1. Our statistical method based on neural networks and Bayesian statistics computes a transfer function between models and observations. Such a transfer function was then used to project future conditions and to derive what we would call the optimal ensemble combination for twenty-first century climate change projections. Our approach is therefore based on one statement and one hypothesis. The statement is that an optimal ensemble projection should be built by giving larger weights to models, which have more skill in representing present climate conditions. The hypothesis is that our method based on neural network is actually weighting the models that way. While the statement is actually an open question, which answer may vary according to the region or climate signal under study, our results demonstrate that the neural network approach indeed allows to weighting models according to their skills. As such, our method is an improvement of existing Bayesian methods developed to mix ensembles of simulations. However, the general low skill of climate models in simulating precipitation mean climatology implies that the final projection maps (whatever the method used to compute them) may significantly change in the future as models improve. Therefore, the projection results for late twenty-first century conditions are presented as possible projections based on the “state-of-the-art” of present climate modeling. First, various criteria were computed making it possible to evaluate the models’ skills in simulating late twentieth century precipitation over continental areas as well as their divergence in projecting climate change conditions. Despite the relatively poor skill of most of the climate models in simulating present-day large scale precipitation patterns, we identified two types of models: the climate models with moderate-to-normal (i.e., close to observations) precipitation amplitudes over the Amazonian basin; and the climate models with a low precipitation in that region and too high a precipitation on the equatorial Pacific coast. Under SRES A2 greenhouse gas forcing, the neural network simulates an increase in precipitation over the La Plata basin coherent with the mean model ensemble projection. Over the Amazonian basin, a decrease in precipitation is projected. However, the models strongly diverge, and the neural network was found to give more weight to models, which better simulate present-day climate conditions. In the southern tip of the continent, the models poorly simulate present-day climate. However, they display a fairly good convergence when simulating climate change response with a weak increase south of 45°S and a decrease in Chile between 30 and 45°S. Other scenarios (A1B and B1) strongly resemble the SRES A2 trends but with weaker amplitudes. [ABSTRACT FROM AUTHOR]

Projections for South America of future climate change conditions in mean state and seasonal cycle for temperature during the twenty-first century are discussed. Our analysis includes one simulation of seven Atmospheric-Ocean Global Circulation Models, which participated in the Intergovernmental Panel on Climate Change Project and provided at least one simulation for the twentieth century (20c3m) and one simulation for each of three Special Report on Emissions Scenarios (SRES) A2, A1B, and B1. We developed a statistical method based on neural networks and Bayesian statistics to evaluate the models’ skills in simulating late twentieth century temperature over continental areas. Some criteria [model weight indices (MWIs)] are computed allowing comparing over such large regions how each model captures the temperature large scale structures and contributes to the multi-model combination. As the study demonstrates, the use of neural networks, optimized by Bayesian statistics, leads to two major results. First, the MWIs can be interpreted as optimal weights for a linear combination of the climate models. Second, the comparison between the neural network projection of twenty-first century conditions and a linear combination of such conditions allows the identification of the regions, which will most probably change, according to model biases and model ensemble variance. Model simulations in the southern tip of South America and along the Chilean and Peruvian coasts or in the northern coasts of South America (Venezuela, Guiana) are particularly poor. Overall, our results present an upper bound of potential temperature warming for each scenario. Spatially, in SRES A2, our major findings are that Tropical South America could warm up by about 4°C, while southern South America (SSA) would also undergo a near 2–3°C average warming. Interestingly, this annual mean temperature trend is modulated by the seasonal cycle in a contrasted way according to the regions. In SSA, the amplitude of the seasonal cycle tends to increase, while in northern South America, the amplitude of the seasonal cycle would be reduced leading to much milder winters. We show that all the scenarios have similar patterns and only differ in amplitude. SRES A1B differ from SRES A2 mainly for the late twenty-first century, reaching more or less an 80–90% amplitude compared to SRES A2. SRES B1, however, diverges from the other scenarios as soon as 2025. For the late twenty-first century, SRES B1 displays amplitudes, which are about half those of SRES A2. [ABSTRACT FROM AUTHOR]