Prediction of kiwifruit orchard characteristics from satellite images

Abstract

The dry matter content of New Zealand kiwifruit is currently measured using destructive testing of a 90-fruit sample. Dry matter content varies from 14 to 20% with a standard deviation of 1.05% in the test measurement. This work investigates the use of multispectral data from satellite images of kiwifruit orchards to predict the dry matter of both green and gold kiwifruit. A novel method is developed that reduces the four-dimensional satellite data to three-dimensional unit color vectors and these show a strong linear relationship with measured dry matter. Regression on the data from a set of 20 ‘training’ orchards yielded predictions of dry matter with a standard deviation of 0.76%. When the resulting model was applied to nine test orchards the standard deviation of predicted dry matter was 0.73%. The prediction of dry matter was accurate even when applied to images taken when the fruit was too immature for the standard dry matter measurement test. Therefore, satellite image data may provide a more accurate and non-destructive alternative to the standard 90-fruit test method. It can also provide a way to visualize the variation in dry matter content of the fruit in an orchard; showing regions that would benefit from remedial action and defining areas where fruit harvesting is optimal. Several vegetation indices, derived from satellite image data, are reported in the literature. The strength of the linear correlation between dry matter and twelve common vegetation indices was tested but was found to be much weaker than the correlation developed in this research.