Correlation test Please consider this table:

     above  below
A11  35,712 15,864
B    22,723 29,741
B10   3,513  5,427
B3b   5,146  3,780
...

I need to test the existence of a correlation between the two the variables. If they would be "counted individuals" (I also apologize for my basic ignorance ...) I'd use a chi-square or a Fisher test.
But here the values are kg obtained as sum of subsequent weighings.
What should I use ?
Many thanks in advance.
UPDATE
I should have been more informative: the table represents the quantity of certain materials above or below a certain geographical altitude, expressed in kilograms. They cannot be counted because they are fragments of different dimensions. Weight is a more meaningful measure of their presence in the two regions (above/below).
The values are not the result of a sampling but the total weight of each material in the area under study.
 A: The problem with this question is that the probabilistic nature of the data is unclear. What sort of sample is it and how do we expect variations to occur due to sampling?
In an extreme case the sample is entirely the same as a the population and there is no probability occuring. For instance, if someone has the hypothesis that a vase filled with blue and red marbles has equal numbers of marbles of both colours then one can count all marbles and colours in the vase and decide with certainty whether the hypothesis is true or not.
Probability is necessary when we would only sample a fraction of the population or when in some other way our sample is not an exact representation of the numbers that we wish to estimate/infer. In order to make inference and estimate the error of estimates or in order to do hypothesis tests, it is important to know how the observations are susceptible to variation.
In the case of counts we might assume a Poisson, binomial or hypergeometric distribution for the distribution of the observations/measurements/sample, which have a known error/variance (or at least the relationship between the mean and variance is known) and we end up with the Chi-distribution (and Chi test) as an approximation.
In your case you need to find some way to express the error/variation of your measurements in relationship to the parameter/model that you wish to estimate/test. (Or maybe there is no error/variation at all?).
