I think my general advice is to think about how the non-dectects will affect the analyses. If they're not going to matter, don't worry too much about them. Just report the limits and the percent below the limits. If they will affect the analyses, choose better / more-appropriate analyses.

How you need to handle non-detects may depend upon how many you have.

I often site the following US EPA document, page 4-42. If there aren't too many non-detects, it recommends just replacing them with half the detection limit and going about your usual analyses. ( http://www.epa.gov/region6/qa/qadevtools/mod4references/secondaryguidance/g9-final.pdf ). This obviously isn't a statistically savvy approach --- and you will find more savvy approaches online --- but in my experience is fine if there aren't too many non-detects (and if the non-detects aren't particularly concentrated in one treatment !!! ).

As you have more non-detects, you may need to use robust techniques ---- in the EPA document exemplified by trimmed means, etc., but in your case, with robust analyses, that might include the R packages you mentioned, or like the WRS2 package, or a variety of other approaches, as you think best.

If you have many non-detects (> 50% according to EPA) , you may need to collapse your data into counts of detect / non-detect and go with tests of counts (e.g. chi-square et al.).

* * *

Your question about the varying calculated limits is an interesting one. I think I would retain the original reported mg/L value and do the soil calculation. (That is, I wouldn't mess with the half-the-detection-limit stuff at this step). I would keep track of the the calculated detection limits for the soil and report them. That is, something like "The nominal detection limit for Ca was 0.01 mg/L, which corresponded to a detection limit of 0.25 to 0.32 meq/100g. Non-detects accounted for 12% of samples for Ca."

I don't think I've seen any source dealing with this question directly. But in my experience, reviewers, etc., don't get too worked up about non-detects as long as there aren't too many and they're handled reasonably.

The question is confusing. Which is the case, detection limit or reporting limit?

The example uses concentration and volume. The reporting requirements are meq/100-g. What is the relationship to volume? There is no problem in the example. Sample 1 is below a concentration of 1. 0.9 is below the reporting concentration.

If there is a detection limit problem and not a reporting limit problem, are the results reported as non-detects?

Hello!

Usually I would say, calculate your concentrations using the LOD as a value for data under the LOD. In this case you will conclude that these measurements are below this concentration, which is fair to say.

Now as far as I understand your soil samples are meant to be "replicates" of a certain terrain or whatever. So basically you have "missing data" in your statistical samples. You can handle that in many ways (like using the LOD calculated value as a replacement data, maybe extrapolating the data based on thecalibration curve), all of the methods have biases.

Keep in mind, that it is also very informative to address, how many missing samples do you have in a sample compared to other samples. I would say this requires a nonparametric, maybe Poisson approach. This analysis would complement any other on "valid" data, if you ignore to use the