| One can conceptually distinguish between the predictability of the first and the second kind. The former arises from the initial, while the latter from the boundary conditions. Examples of the predictability of the first kind are weather prediction and seasonal forecasting. The atmospheric response to anomalous sea surface temperature (SST) or changed orography is an example of the predictability of the second kind. Climate change projections for the 21st century are an example of the mixed problem, as both the initial state and the changing atmospheric composition are important. What can be learned from the long history of numerical weather forecasting is that skill improvement was mostly due to better initialization and to model improvement. Both initialization and model performance are far away from being perfect considering the latest climate change projections published in IPCC AR4. Most models were not at all initialized with an estimate of the current climate state and model biases are relatively large. Current activities aim at better initialization and reducing model biases, so that one can expect much improved projections in IPCC AR5. |
|