Hull rot, primarily caused by Rhizopus species in Australia, remains a significant challenge for almond production, particularly during the early hull split stage. The disease reduces nut yield through infected nuts and results in twig dieback and death of fruiting spurs, ultimately impacting production in subsequent seasons. Traditional hull rot severity assessments rely on labour-intensive visual observations, limiting the scale and replication of research trials.
Recent advancements in proximal imaging technology and artificial intelligence (AI) offer new possibilities for automating hull rot detection and assessment. The Green Atlas Cartographer integrates high-resolution terrestrial imaging and AI-assisted mapping, enabling efficient single-tree-level mapping of orchard parameters, including flower and fruit counts and tree structure, across entire orchards.
A trial was initiated in the 2024 almond season at the Mildura SmartFarm, Victoria, Australia, to evaluate the feasibility of using this technology for hull rot severity assessment. The industry standard planting comprises of 1,350 trees over 4.5 hectares, replicated with three almond scions (Nonpareil, Carmel, Monterey) grafted onto three rootstocks (Nemaguard, RP40, Krymsk 86). Trees are planted at a density of 300 trees/ha with 5.0 m spacing within rows and 6.67 m between rows and subjected to three nitrogen application rates (0.5×, 1×, 2× industry standard).
In January 2024, 100 high-resolution images of hull rot-affected trees were captured using the Green Atlas Cartographer mounted on an electric all-terrain vehicle. These images were manually annotated to identify visible hull rot symptoms, and the labelled dataset was used to train an AI-based detection model. This preliminary model was tested on orchard maps generated in February 2024. Testing was also completed on a second orchard block which features a replicated high-density planting (1, 2, or 4 m within-row spacing with 5 m between rows) and includes three scions (A12, Carina, Vela) grafted onto four rootstocks (Garnem, Nemaguard, RP20, RP40) across 6.5 hectares.
Heatmaps generated from the geo-referenced imaging data identified areas indicative of hull rot strikes. In the standard industry planting, elevated hull rot incidence was detected in Nonpareil rows receiving high nitrogen applications, correlating with visual observations. In the high-density planting, the heatmaps revealed higher hull rot incidence in the cultivar Carina compared to A12 and Vela, consistent with manual assessments.
This study demonstrates the potential of proximal imaging combined with AI for efficient, large-scale hull rot detection and mapping. The technology could enable precision orchard management by localising interventions to high-pressure zones, significantly reducing chemical inputs and labour requirements while enhancing research capacity for disease management.