Advances and challenges in climate modeling

In spite of the chaotic nature of the atmosphere and involvement of complex nonlinear dynamics, forecasting climate fluctuations over different timescales is feasible due to the interaction between the atmosphere and the slowly varying underlying surfaces. This review provides insights into climate predictions across subseasonal to decadal timescales and into making projections of future climate change. Different sources of uncertainty in climate predictions are discussed, including internal variability uncertainty, which is large for short-term predictions of up to a decade or two, model uncertainty for predictions at all timescales, and scenario uncertainty for climate change projections at the end of this century. Climate models have been significantly improved in recent decades, mostly through improved parameterization of unresolved processes and enhancement of the spatial resolution, while ensemble forecasting has also been developed to capture strong predictable signals. Future research should aim to reduce uncertainty in climate predictions, for example, through the application of high-resolution climate models. However, sub-grid-scale features would still be parameterized, underlining the need for further improvements in physical parameterizations to account for sub-grid-scale processes. There is also a need for improvement and extension of the current observing system, which will greatly advance understanding of the key processes and features in the climate system. The advanced observing system in the future will also be beneficial for more accurate representation of the initial state of the components of the climate system in order to obtain more accurate climate predictions. In spite of progress in model development, the spread of projected precipitation by different models under a specific radiative forcing of greenhouse gases is still large at the regional scale. Improving future projections of regional precipitation requires better accounting for internal variability and model uncertainty, which can be partly achieved by improvement and extension of the observing system.