Observation and prediction are different jobs
Monitoring tells us what the climate system has been doing. Prediction asks what it may do next. Satellites mostly do the first, but their data is deeply woven into the second. Climate models are computer programs that represent the atmosphere, oceans, ice and land using physical laws. They need observations to be built, tested and sometimes started in a realistic state.
It helps to separate two timescales. Weather forecasts look days ahead and depend on precise starting conditions. Climate projections look at decades and describe statistics, such as average conditions and the likelihood of extremes, under different assumptions about future emissions. A climate model does not try to say what the weather will be on a given day in 2060.
How satellite data enters the process
Data assimilation. In weather and seasonal forecasting, millions of satellite measurements are blended into models to estimate the current state of the atmosphere and ocean. The same approach is used to build reanalyses, which are consistent historical reconstructions of past conditions. Researchers use these to study how the climate has behaved.
Evaluation. Modelers compare simulated results with observed values such as temperature patterns, sea ice extent, cloud cover and sea level. If a model consistently misses something observed from space, such as how much sunlight clouds reflect, developers look for flaws in how that process is represented.
Constraining processes. Some of the largest uncertainties in projections involve clouds, aerosols and how the climate responds to added greenhouse gases. Satellite observations of these features help narrow the range of plausible model behavior. Researchers also use observed relationships, such as how the system responded to past volcanic eruptions or natural swings, as tests for models.
Inputs. Models need information about the land surface: vegetation, snow, soil moisture, and ice extent. Satellite products supply much of this.
What better data could change
More accurate, more frequent and more complete observations could help in several ways:
- Narrowing uncertainty in how strongly the climate responds to greenhouse gases, though this is a hard scientific problem and data alone may not settle it.
- Improving the representation of clouds and precipitation in models.
- Better tracking of the carbon cycle, which would help determine how much carbon land and oceans absorb.
- Improved short-term and seasonal forecasts, which are useful for farming, water planning and energy.
- Detecting changes in the system sooner, which helps check whether projections are on track.
The role of computing and machine learning
Computers that run higher-resolution models can resolve more detail, such as storms and local terrain. Machine learning techniques are being explored for speeding up parts of models, for filling gaps in observational records and for spotting patterns. These tools show promise, but they depend on data quality, and their predictions need careful testing, especially for conditions outside those seen in the past.
Uncertainty is part of the picture
Projections come with ranges. Some uncertainty comes from the models, some from natural variability, and some from unknown future human choices. Better observations can shrink the first kind, but they cannot remove the others. Scientists typically express results as likelihoods or ranges instead of single numbers, and a reader should be wary of claims that sound more precise than that.
Looking ahead
The most important improvement may be continuity: maintaining high-quality records long enough that models can be tested against changes that unfold slowly. As observations lengthen and models improve, projections are likely to be refined, though surprises remain possible.