Eliminating external interferences in handheld NIR spectroscopy for grape quality assessment

One indicator for determining the perfect time for harvesting grapes to achieve the best quality wine is the sugar and acid content. Traditional methods are costly and time-consuming, whereas handheld NIR spectroscopy combined with chemometric methods are fast and non-destructive. To use this, a calibration process is required to extract the quality parameter from the spectrum. This involves first collecting spectral data and then determining the corresponding reference data, which is often done in a controlled laboratory environment. In practical application, external parameters from an uncontrolled environment, such as different light sources or temperatures, disturb the spectral signal, reducing the performance of the model and even making it unusable in some cases. Several methods have proven effective for other fruits in eliminating these influences and making the models robust. This study gives an overview about these methods and shows the potential of the External Parameter Orthogonalization (EPO) method in the use case of viticulture. To achieve this, a data set with measurement pairs collected both in the laboratory and in the vineyard is used. Partial Least Squares Regression (PLSR) is used as a baseline model.