They use so-called "proxy data". You measure something else that has been accumulated and preserved over time and that somehow correlates. You have to create a model, fit it so it correlates with recent measures and you are done.
Level of oxygen for example can be estimated from layers of ice. Mann used tree slice(s) for temperature estimation.
There's a lot of problems with proxy data and their models and there is a lot of space for scientific misdemeanor. Some journals recently started to require a full disclosure of all data and models on which they base their claims, which - surprisingly - isn't always the case.
As more data is gathered over time some models need to be adjusted. How much they need to be adjusted reflects quality of the model.
Level of oxygen for example can be estimated from layers of ice. Mann used tree slice(s) for temperature estimation.
There's a lot of problems with proxy data and their models and there is a lot of space for scientific misdemeanor. Some journals recently started to require a full disclosure of all data and models on which they base their claims, which - surprisingly - isn't always the case.
As more data is gathered over time some models need to be adjusted. How much they need to be adjusted reflects quality of the model.