Weather predictions and climate projections involve the use of climate models. The non-linear and chaotic nature of the climate system creates some natural limits to predictability.
This means that daily weather forecasts have decreasing skill beyond a few days, multi-week weather forecasts are experimental at this stage, and there are inevitable limits to the skill and scope of seasonal (multi-month) forecasts and multi-year climate predictions.
Research is ongoing into the area of decadal prediction, and the latest IPCC assessment report devotes a chapter to this topic (Chapter 11 ). There is evidence that some ‘skill’ is possible for predictions for up to ten years of average temperature and to some extent for average rainfall. However, natural variability in the climate system due to internal processes (e.g. the El Nino Southern Oscillation and the Pacific Decadal Oscillation) and external drivers (e.g. 11-year sunspot cycle, volcanic eruptions) limit predictability.
Multi-decadal climate projections are also affected by this natural climate variability and by human-induced changes in external factors such as greenhouse gases, aerosols and land-use. Projections require plausible estimates of the (i) change in radiative forcing associated with these factors, (ii) an estimate of the response of the climate system, as well as (iii) an estimate of the time evolution of internally generated climate variability.
Estimates of the change in radiative forcing are based on scenarios of future emissions and atmospheric concentrations of greenhouse gases and aerosols. There are four scenarios called Representative Concentration Pathways (RCPs), each of which is similar up to about 2030 but diverge after that. This means that the climatic effect of each scenario is similar for projections to 2030. Projections beyond the middle of the century show a greater sensitivity to the emission scenario: by 2090, the climate response under the lower RCPs is much weaker than under higher RCPs. However, for some variables, such as regional rainfall, the sensitivity to the emission scenario may be smaller than the range of uncertainty due to results from different climate models.
Climate projections have internal variability, but it is not designed to replicate the observed sequence of climate variability. Outputs from acceptable models typically have variability statistics for the historical period that are similar to that observed, except for the sequencing of events. Similarly, projections show the effect of internal variability and external climate forcings, but are not designed to predict the actual ups and downs of climate variability at scales from weather (days) to multi-decadal variability. This includes periods of accelerated and reduced global warming, regional droughts or floods, the phase of the El Niño Southern Oscillation and other aspects of climate variability. For projections to 2030, the climate variability may be similar or larger than the forced change in many cases.