You might try using the technique of Maximum Likelihood  - too hard to explain that approach here, but if you google on that term you will find a number of good examples and references.  This approach is generally useful when your trying to find the best fitting model among a number of input variables, some of which are already correlated with one another.  

The analysis approach will depend on your objective and the data you have available. If you are interested in understanding the relationship between the ecosystem and 'Drivers' there are a range of multivariate methods some based on parametric and some based on non-parametric - designed to look at underlying drivers. However, if your objective is around prediction and less around understanding might want to explore some of the data mining techniques. It is difficult to answer this question without knowing the objectively clearly or the data structure.

Hi Lisiana, your doubts seem to me to be related to statistical methods and not to the typology used in classifying proximate causes and underlying drivers of deforestation/LUCC, which is clearly explained in several studies by Geist and Lambin (e.g. 2002). As they did this kind of meta-analysis, I think that your best bet will be to dialogue with them or the other scholars that worked in team with them in several other articles and try to define a methodology that can be used from now on by scholars doing a similar research.

best

You may be interested in our article (Gils & Loza 2006) on drivers of deforestation in Bolivia. The most interesting finding may be that land tenure, or rather its absence was the dominant driver.

@Marina: Land tenure was not tested as deforestation driver by (Geist &) Lambin.