Integrated vineyard precision control system pilot

Abstract

This project provides the foundational work to road-test key components of a digital vineyard guidance system that is designed to facilitate costs-of-production savings for Riverland growers. Outcomes include: (i) an audit of underpinning technologies that would support the digital guidance system; (ii) a long-range wide area network (LoRaWAN) base station at the Loxton Research Centre; (iii) a ‘proof-of-concept’ ground-based image retrieval system to provide visual imagery of vineyard development over time; and (iv) an open-source dashboard that can provide decision-relevant information to growers. Case studies are used to demonstrate potential grower value in terms of labour and operating cost savings.

Summary

This pilot project represents a collaboration between engineers, scientists and economists from the University of Adelaide, in partnership wine grape growers in the South Australian Riverland spanning a six month period in the first half of 2019. The long-term ambition of the collaboration is to create an operational digital system that collates a variety of information that collectively can help increase information transferability, transparency and ease-of-access, support on-farm decision making, and create a return on investment both to Riverland growers through improving gross margins and profitability. This report documents the outcomes of ‘stage 1’ of this process, which has been designed to provide foundational proof-of-concept work to road-test all the key components of a comprehensive digital vineyard guidance system, thereby paving the way for developing operational systems in subsequent stages. 

The project leverages substantial and accelerating advances both in Australia and internationally in a range of agricultural technologies, spanning sensing, connectivity, data analytics, automation and prediction. Taken as a whole, these technologies have the potential transform vineyard processes by enabling near real-time tracking and/or future prediction of vineyard decisions (e.g. irrigation, spraying, nutrition, canopy management), resource utilisation (e.g. labour, machinery, water, energy, nutrition and other chemicals) and vineyard performance (growth, yield and other measures of vine development). The vision is that by digitally linking ‘actions’ undertaken by growers on the vineyard with ‘outcomes’ (both physical outcomes such as yield and quality measures, and financial outcomes such as gross margins and profitability), it becomes possible to develop predictive analytics and advisory services to optimise vineyard decision making. 

In addition to this high-level vision, the work in the pilot project responded to the following design criteria identified by Riverland growers in the early project stages: 

• The need for systems to provide guidance and advice (e.g. when to irrigate or spray), rather than simply displaying data; 
• The need for ‘producer led’ innovation models to develop operational field-tested systems that can be rapidly adopted by growers, rather than focusing on technologies that are ‘stuck at the bench’ or that do not move beyond proof-of-concept or pilot phase; 
• The need to focus on cost-of-production issues to achieve ‘fit-for-purpose’ grape quality as a means of increasing gross margins and profitability, and the associated need to provide an economic lens over the technology development to ensure grower benefit; 
• The need for open systems that support interoperability between different sensing systems, algorithms and visualisation solutions (e.g. dashboards), rather than multiple (often proprietary) systems that are not interoperable and can create technology ‘lock-in’ issues for growers; and 
• The need to account for the unique characteristics of the South Australian Riverland, including being a large-area bulk growing region with very large-scale operations. 

To this end, the specific outcomes of the ‘stage 1’ pilot project are summarised as follows: 

• An assessment was conducted of the ‘technology readiness level’ of a range of technology solutions that can potentially be incorporated into a digital viticulture guidance system, including a review of sensor systems, connectivity solutions, algorithms and modelling solutions. The assessment found a high level of maturity of in-situ sensors, but less maturity in other data acquisition systems (e.g. vision data, financial data, other grower records). Commercial adoption of algorithms that ‘value add’ on data remains low in the viticulture industry, and integrated solutions that translate data streams to advice and decision recommendations are limited. 
• A review of space-borne data acquisition streams showed a high level of technological maturity in terms of spatial, temporal and spectral resolution, yet only a small subset of potential use-cases have thus far been commercialised within a viticultural context. Potential applications of space-borne data streams in the pilot successfully demonstrated the detection of spatial anomalies in vigour using a range of satellite products with different resolutions, temporal frequency, record lengths and pricing structures. 
• A proof-of-principle mobile ground-based image retrieval system was developed and demonstrated to work in-field for a range of realistic operating conditions (i.e. realistic tractor speed, vibrations and light conditions), with high potential for cost-effective ‘incidental’ data capture. The system is currently tractor-mounted and uses a multi-camera system to provide visual spectrum imagery (i.e. high resolution photos) and photogrammetric information that provides information on canopy size, volume and density. 
• Long-range wide area network (LoRa WAN) technology was selected as the demonstration communications technology due to its low cost, long range and increasing breadth of compatible sensor options. A LoRa base station was installed and operationalised at the Loxton Research Centre, with demonstrated range up to 18 km (directionally dependent, based on line-of-sight). The technology was demonstrated using set of meteorology, soil moisture and plant sensors at the Sherwood Vineyard, located 3.5 km from the Loxton research centre. 
• Computer vision technology was applied to detect bunches within a canopy, and segment key canopy elements such as bunches, canopy, trunk, and green shoots. An assessment of potential future applications of machine learning to enable monitoring of canopy indicators, flowers, berries, bunches, diseases, weeds and water shoots is provided. 
• Conceptual approaches to numerically modelling physical and biological processes in the context of vineyards were identified, focusing on models that: (i) estimate and predict vine development; (ii) predict grape quality; (iii) estimate and predict yield; (iv) estimate and predict disease risk; and (v) simulate on-farm operations. The review included both machine learning and mechanistic (biophysical) modelling approaches. Pathways to develop systems that provide guidance and advice by building on existing research platforms were reviewed. 
• An open-source data storage and visualisation (i.e. dashboard) system was developed that presents real-time data feeds to growers, including ‘internet of things’ data, satellite data, ground-based imagery from the tractor-mounted camera system, historical water pricing information and management records (e.g. spray records). 
• Preliminary estimates of grower costs highlighted significant year-to-year operating cost variations across vineyards, with the largest incurred costs being hired labour, followed by contracts, interest payments, water, and repairs and maintenance. A scenario-based assessment using a small (<10 ha), medium (11-80 ha) and large farm (>80 ha) highlighted areas of pre-existing technology investment, and identified potential areas for digital technologies to save water, electricity, fuel and/or labour costs, as well as the potential to increase overall farm yield. 
• Preliminary benefit-cost estimates found considerable quantitative economic benefits from digital agricultural solutions that targeted decision making, with a benefit-cost ratio conservatively estimated to be 3.38, indicating that for each dollar invested, the grower would receive a return of $3.38. 

It is recommended that future project phases continue to build the open-source digital viticulture platform, with the following elements in mind: 

• A visualisation platform that provides a ‘one-stop-shop’ for all key data streams relevant for grower decisions, with a focus on ease-of-access and display; 
• A benchmarking app that enables growers to compare key on farm attributes (e.g. resource utilisation per ha, yield per ha, etc.) with anonymised ‘similar’ vineyards; 
• A series of prediction services focusing on yield, objective quality measures and disease risk; and 
• A series of advisory services, focusing on infrastructure management (e.g. adequacy of irrigation infrastructure design, and any infrastructure malfunction such as leaks and blockages), irrigation requirements, water market investments, canopy management, nutrient management and spray management.