Prediction of core browning and fruit firmness in stored 'Braeburn' apples: Multivariate time series classification

Physiological storage disorders affect many commercial apple cultivars and cause economic loss and increase food waste. In general, robust prediction methods for disorders such as 'Braeburn' internal browning, specifically core browning as it is most prevalent in Southwest Germany, do not exist. A good understanding of the causal factors is still lacking, but symptom development varies markedly between orchards. This work describes a combined modeling and machine learning approach to link pre-harvest environmental, tree and management factors with post-harvest storage outcomes to predict internal browning and fruit firmness. Nondestructive spectral time series data were collected from 'Braeburn' apples over seven seasons from 2016 to 2023 from orchards located in Southwest Germany. These data were used as reference measurements to construct a weather-based model to produce multivariate time series for changes in four fruit quality parameters: chlorophyll, anthocyanins, soluble solids and dry matter content during pre-harvest growth and development on the tree. The multivariate times-series were then used as input into a dynamic time warping (DTW)-based k-nearest neighbor (kNN) classifier. The data set contains 1729 objects and the classifier shows reliable good cross validated prediction results for internal browning and firmness, respectively. After DTW and dimension reduction mapping, 2D cluster analysis shows each season as being clearly separated in multidimensional space. However, single season learning cannot be generalized to other seasons. The multidimensional distance between browning clusters is related to the average rate of change in the spectral scanning data during the 50 d pre-harvest period. Fruit lots with high browning show a higher mean cumulative change.