Online NIRS analysis for the routine assessment of the nitrate content in spinach plants in the processing industry using linear and non-linear methods

This study aimed to assess the robustness of the NIRS models developed following different strategies for the routine prediction of nitrate content in spinach plants using an online FT-NIR spectrophotometer. To achieve this, 516 spinach plants from different cultivars, harvest dates, orchards and seasons, were used. Two strategies were followed to make up the calibration and validation sets; the first included in the calibration set those samples belonging to the 2018 and 2019 harvesting seasons, while the second also included in this set part of the population of the 2020 harvesting season. Modified partial least squares quantitative models were initially developed and externally validated. In view of the results and to obtain significant improvements, a non-linear regression technique (the LOCAL algorithm) was applied. The models developed using the non-linear regression technique and considering the greatest possible variability in the training set (samples belonging to 2018, 2019 and 2020 harvesting seasons) reported the best prediction results (R-p(2) = 0.60; SEP = 758 mg/kg), which enabled to classify the product in the main categories or classes established by the official regulations, according to its commercial destination.

Automated summary

This study aimed to assess the robustness of NIRS models for predicting nitrate content in spinach plants, using different strategies. Results showed that utilizing non-linear regression technique with consideration of maximum sample variability achieved the best prediction performance.