I'm trying to build a model to asses different physiological parameters (Dry weight, yield, Nitrogen concentration, one model for each) from a field based on reflectance from different wavelengths i.e. asses one parameter based on other parameters.
The data is from an old experiment I didn't conduct. In the original experiment they looked at the effect of fertilizing with different levels of nitrogen on yield but they measured other physiological parameters and measured spectral data. I'm ignoring the different treatments since I'm only interested in the end result but this becomes relevant later.
I built a MLR and used a leave-one-out cross validation to asses the model. In some models I got a higher R2 in the CV than in the full sample model (only calibration, no testing). I think that this happens because the point I'm leaving out was actually already used, it is identical to another point that was used for calibration.
I need to find a way to calculate the maximum distance between two values for them to be identical with some confidence. I plan on testing other types of regressions for this, maybe it affects the method.
For the reflectance data I think this distance should be the measuring error of the instrument, is this true?
For the physiological measurements I don't have any statistics, only a final number. I will take into account the measuring error (weight, volume etc.) of the instrument but should I (can I) take into account other errors?
I have a biology degree level of statistics, pointing me in the right direction with the name of the field or test I should explore will be appreciated.