If you are running a gradient analysis, then the change in solvent properties (RI) and temperature over time can cause noticeable baseline drift during the run. This drift up or down relative to the starting baseline can cause a number of quantification problems with the analysis reporting software (as flat baselines are more easily and accurately integrated than sloped ones).
Using the concept of Method # 2 described above, many HPLC manufacturers’s added a feature known as a the ‘Reference Wavelength’ to their systems. This feature allowed a chromatographer to input a second wavelength value (and bandwidth) into the method which would be used to subtract out raw data from the primary wavelength during the analysis. This subtraction occurs in real time, on your raw data gathered from the detector and the resulting data reported to the user is in fact the result of the subtraction. You will never know what the original data looked like before the reference wavelength was subtracted (it has been destroyed). Only the manipulated (subtracted) result is provided. If any sample peak(s) or impurities appeared in the region (depending on the wavelength and bandwidth specified) where you selected a reference wavelength, then the resulting data would have been subtracted from your actual sample and you would never know it happened or have any record of it! This brings up a serious validation issue as you are modifying the original data with no way of knowing (or documenting) how you have changed it. It is for this reason alone that we teach chromatographers to always turn this feature 'OFF' by default. If they want to make use of the feature, then we suggest that they simultaneously collect data from a second, separate wavelength channel such that the two raw data streams are preserved for validation purposes (Method # 2). IOW: Turn 'OFF' the Reference Wavelength feature and record all of the signal data. The separate signals can be compared, subtracted or manipulated as needed for integration and reporting purposes, but the original signal data is left unchanged and secure. This allows you to monitor for contamination, impurities, problems or changes during the run. It also allows others to verify your method for accuracy.
Observational Notes: A common reason we observe this problem in the lab is that I am called in to diagnose a strange problem where the area of a known sample peak changes in an unexpected way. That "way" often includes going NEGATIVE, increasing in mass or decreasing in mass. *This is due to the reference wavelength feature being turned 'ON' and another peak absorbing in the region selected which contributes to the final signal. If the data collected (area) for the 'reference peak' is larger than the sample peak the resulting chromatogram will show a negative peak (this tends to be noticed!), whereas if the reference peak is smaller than the sample peak, the resulting signal decreases, which may or may not be noticed (incorrectly interpreted as a lower concentration sample). You can see the obvious danger posed by this situation. Companies can be put in a situation where all of their past data is found to be invalid and product recalls can result from this finding. The cause was lack of training in the use of the software and/or HPLC system.
How to Solve The Problem: The reason we see this feature cause so many problems in laboratories appears to be due to the fact that the Reference Wavelength feature is being turned 'ON' by default in the software for most DAD/ PDA modules (The real default value for "Reference Wavelength" should always be: 'OFF'). To make matters worse, the default values for the wavelength and bandwidths often supplied by the manufacturers are actually used by most chromatographers (what are the odds that the random values placed in the system are even relevant to your analysis? Why would you use them?). We suggest using a ‘canned’ method template in most laboratories which includes a new default value for this feature... 'OFF' for all analysis methods. Most importantly of all, please obtain formal training in the use of a specialty detector such as a diode array before using one for sample analysis.
Note: The bandwidth chosen for each wavelength is also very important and if chosen poorly, can result in adding noise to your signal, reducing it or even enhancing it. Please refer to this article for more info: http://hplctips.blogspot.com/2011/09/uv-vis-hplc-detector-signal-bandwidth.html