In January, the Intergovernmental Panel on Climate Change (IPCC) startled the world with a frightening climate scenario: by 2100, global temperatures could rise by almost 6 °C. Fortunately, this is only a worst-case scenario. Other models predict a temperature rise of only 1.4 °C. Most probably, the rise in the next decade will lie between these extremes. But there is also a small chance that none of the IPCC's scenarios will come close to reality.

So is climate prediction no more than a game of chance? Not quite — but it is closer to one than it need be. The accuracy of any model depends significantly on the quality of the underlying raw data. The problem is, the quality is patchy.

Monitoring and predicting the global climate requires a reliable system of constant observations, rapid data exchange and long-term recording, in standardized formats. That is why, in 1992, the Global Climate Observing System (GCOS) was established. Its aim is to ensure that climate-relevant information is obtained and made available to all potential users. The creation of GCOS was a major advance, but the reporting system has significant deficiencies, and there are large gaps in global and regional coverage, which seriously affect climate assessment and modelling efforts.

Terrestrial climate monitoring is currently based on a network of around 1,000 GCOS observation stations. But the reliability of the data that some of them collect is inadequate, and a disproportionate number of these stations are in rich countries, with sparse coverage in many regions of Africa, South America and Asia. Although no one can say how many climate data are needed for accurate monitoring, clearly the geographical distribution should be as even as possible.

But a GCOS station costs up to $500,000 per year to operate and maintain — too much for poorer countries. And maintenance costs are particularly high in remote polar regions; over the past years, several stations have been closed, for example, in Russia and Canada.

Under the United Nations Framework Convention on Climate Change, all countries are required to set up and run appropriate observation programmes, and to exchange data with other nations and with international organizations. But in practice, many poorer countries spend little on their regional climate observations. The training of technical staff and the maintenance of instruments at their observation stations are often inadequate. Through misreporting, instrumental drifts and biases, unreliable communication infrastructures or political unrest, about half of the world's climate data potential is lost or corrupted each month.

The World Meteorological Organization gives regional training and technical support. But this is not enough. The United Nations should also help poor countries to collect and distribute accurate and consistent data sets.

Sea-based climate observation and ocean monitoring, which is likely to add significantly to our knowledge of what drives atmospheric processes, is only just beginning. The deep oceans in particular are still under-observed. Efforts must continue to implement a more systematic ocean observation system. The Global Ocean Observing System (GOOS), founded with GCOS, has the right approach. GOOS coordinates the use of new technologies, such as meteorological buoys, which measure climate-relevant variables in and over the oceans. But it must be expanded.

A worldwide network of sea-based observation buoys will not come cheaply, and will need strong international coordination. Projects such as the Tropical Ocean Global Atmosphere programme, which ended in 1994, have shown that systematic ocean observation is essential for predicting El Niños or seasonal weather. There is a strong case for heavy investment in ocean meteorology.