High resolution remote sensing, GIS and Google Earth for avocado fruit quality mapping and tree number auditing

Abstract

Satellite imagery, Geographical Information Systems (GIS) and Google Earth were demonstrated as effective tools for tree auditing and for defining the spatial variability of tree condition across two commercial 'Hass' Avocado blocks in Childers, Australia. Using a classified normalised difference vegetation index (NDVI) map derived from a WorldView2 image captured on April 19th 2013, and then repeated May 29th 2014, three replicate trees within high, mid and low growth regions were selected from each block. The annual harvest of these trees was conducted on May 21st 2013 and May 7th 2014, along with measures of leaf fluorescence and SPAD. Fruit quality measures were also conducted post-harvest. Spectral information corresponding to each of the sampled tree canopies was extracted and used to derive a number of pigment and structural based vegetation indices. These indices were correlated against the measured fruit and tree parameters, resulting in strong correlations with average fruit weight (g) (R=0.86∗ ∗, P=0.0001, RMSE=14.7 g); % commercial yield (R=0.71∗ ∗, P= <0.0001, RMSE=13.2%); Fm/Fv (R= -0.67∗ ∗, P= <0.0001) and SPAD (R=0.67∗ ∗, P= <0.0001). Using algorithms produced from the linear relationships between the measured parameters and the derived vegetation indices, maps representing the spatial and temporal distribution of each of the parameters were developed. This information supports improved tree management and harvest segregation based on fruit size. For the auditing of tree numbers, a more exact method for calculating orchard area and tree density was examined that increased the prediction accuracy of one orchard from 90.4 to 97.3%. The development of a spatial database that incorporated individual trees within an orchard was also investigated. At the regional level an accurate inventory of trees supports improved handling, marketing and forward selling decisions; whilst at the orchard level, would offer more efficient data recording and enhanced 'paddock to plate' traceability.