Temporal sensitivity of the wavelength calibration of a photodiode array spectrometer

Subtle differences in the relationship between wavelength and pixel on photodiode array spectrometers contribute to difficulties in transferring calibrations from one instrument to another and may even introduce errors on a single instrument over time. To quantify the level of drift that might be expected in photodiode instruments, we calibrated the wavelength scale of two Zeiss MMS-1 photodiode spectrometers weekly over a 12-month period. We found no evidence of drift in the wavelength calibration. The wavelength calibration was consistent within 0.03 nm over at least 150 days and better than 0.1 nm over the year. To provide context for the wavelength accuracy, we applied small perturbations to wavelength in two partial least squares (PLS) models. We found that wavelength perturbations introduced a linear increase in bias of about 7%/nm (for example, a 1-nm perturbation shifted fruit dry matter prediction from 14% to 21%) in a kiwifruit dry-matter model and about 3.6 °C/nm in an Intralipid temperature model. By including small wavelength perturbations in the training sets, we were able to reduce this error to less than 1.7%/nm and 0.2 °C/nm in the dry-matter and temperature models, respectively. These results suggest that the wavelength scale of photodiode instruments can be very stable. However, in light of the high sensitivity of the PLS models we examined, we recommend testing and, where possible, mitigating the sensitivity of PLS models to small wavelength shifts.