Multivariate analysis

I have been issued data on chemicals in air samples from around the world, elements parts per million.
Some of the stations have returned data where if the parts are really small <20 they have recorded the result as -20. Which I do not know how to include in my analysis of the data. I was therefore planning to omit the results and use whats left, however,
From my analysis I want to know if I can group the air samples so that when new samples are collected we can accurately predict where in the world latitude longitude the sample was collected I therefore need as much data as possible and omitting the -20s for this element from this station would remove 50% of the sample data.
Has anyone experience of mulitvariate analysis and could offer some advice