Developments in digital agriculture

New technology offers exciting opportunities to complement traditional practices and optimise crop management. By Anna Mouton.

Digital agriculture refers to the application of new technologies for data collection and analysis to help growers manage seasonal and environmental variability. Stellenbosch University has a Digital Agricultural Research Group within the South African Grape and Wine Research Institute.

“The Institute allows for multidisciplinary research because we can accommodate researchers and students from other backgrounds,” said researcher Talitha Venter. They are not limited to vineyards but have worked on crops ranging from apples to avocados.

Venter presented at the Hortgro Technical Symposium on behalf of Prof. Carlos Poblete-Echeverría, the Digital Agricultural Research Group coordinator. She explained that the Group develops digital tools to enhance agricultural productivity, sustainability, and resilience.

From data to decision

“Traditional agricultural management practices are uniform — we normally apply them across the board, regardless of variability,” said Venter. For example, the entire orchard might receive the same fertiliser without consideration of soil or plant variability.

The problem with traditional practices is that uniform management of a variable system leads to a variable outcome. In the case of pome and stone fruit, the outcome is uneven yields, harvest maturity, and fruit quality within blocks.

Digital agriculture offers the potential to quantify temporal and spatial variability and enable appropriate management. For example, some growers already combine remote sensing and GPS-enabled spreaders to tweak their fertiliser applications. This promotes tree uniformity, saves input costs, and reduces environmental impacts.

Venter described the three steps of digital agriculture solutions: data acquisition, data analysis, and implementation.

Advances in data acquisition have brought non-invasive sensors and platforms that provide higher resolution and accuracy at a lower cost. Sensor technologies include RGB imaging (digital cameras), spectroscopy, multi- and hyperspectral imaging, chlorophyll fluorescence, infrared thermography, electrical resistance and conductivity, and LiDAR (laser imaging detection and ranging).

Sensor platforms can be space-, air- or ground-based. “We’re moving to robotics and integrated sensor platforms where you can use one platform to measure multiple variables,” said Venter.

Data analysis draws on artificial intelligence, which includes machine learning — read more about machine learning in agriculture in Machine Learning in Agriculture.

“The last step — implementing these technologies — is the most important,” said Venter. “Providing a map or an integrated information system allows you to use delineated management practices according to differences in your orchard or vineyard block.”

Ideally, the outputs should be available as mobile or web-based apps that facilitate growers’ adoption. “I think the key is making all this information and these technologies practically applicable for producers,” she added. “But it’s quite tricky.”

Application examples

Venter discussed examples in five fields: water management, canopy assessment, pest and disease monitoring, fruit quality, and yield estimation.

Water management

The Digital Agriculture Research Group has conducted several studies using remote-sensing data to determine water requirements and manage stress in orchards and vineyards. These include estimating evapotranspiration in orchards and vineyards from satellite images and monitoring canopy temperature to predict stem water potential in grapevines.

Earlier this year, they completed a project that deployed drones to measure canopy temperature and map water stress in apple orchards. Pome-fruit grower and viticulturist Karin Clüver conducted this research for her MSc. Read more about her findings and how she continues implementing this technology in The Drone Detective.

Canopy assessment

Differentiating tree or vine canopies from soil or cover crops is one challenge when analysing satellite or drone images. “We can classify images into soil, shadow, and plant classes using methods such as artificial neural networks to work accurately with only the plant data,” said Venter.

The Group applies LiDAR to create three-dimensional reconstructions of canopies and to determine canopy volumes. “We can also look at canopy temperature, use multispectral imaging, and, with GPS-enabled monitoring systems, make canopy-volume maps,” she said.

“Furthermore, we tested a technique that could extract the canopy volume from photos, using the sky as background. We tried it in an apple orchard, taking the photos in quadrants, and there was a good correlation between the calculated leaf area index and the index measured by sampling the leaves from the tree.”

Pest and disease monitoring

Venter highlighted three crop-protection projects: thermal imaging to identify apple scab, RGB photography to identify powdery mildew, and proximal and remote sensing to identify root rot in avocados.

The apple-scab research was part of Philip Rebel’s work on mancozeb deposition, featured in previous Fresh Quarterly articles.

Fruit quality

“For my PhD, we’re looking at non-destructive approaches, like multispectral imaging in the field, and relating that to hyperspectral imaging in the laboratory, and using machine learning to relate those to physical and chemical bunch characteristics,” said Venter.

Yield estimates

The Digital Agriculture Research Group is revisiting the use of photography to identify bunches on vines for yield estimates. Similar technology applies to pome and stone fruit.

“We use the YOLO — you only look once — model, which is supervised training where an expert labels the bunches on a photo to train the computer, and we compare the human and artificial intelligence results to assess the model’s accuracy,” said Venter.

This even works on white grapes and green apples. “We don’t know what the neural processes are for humans identifying a bunch, but we don’t only look at colour — our brains consider several different aspects,” she said.

“The nice thing about artificial intelligence is that there is no limit to how many aspects the computer can use to make a decision. It doesn’t get mental fatigue.”

Looking ahead

“Our perspective is to continue supporting traditional agricultural practices with the latest data capture and analysis technologies,” said Venter.

“We call ourselves the Digital Agriculture Research Group even though we’re housed within the Department of Viticulture and Oenology because we work on other crops, and the principles behind these technologies can be applied to any crop.”

Although Venter is confident about the potential of digital technology to transform agriculture, she stressed that artificial intelligence models must be trained and calibrated before release — systems must be ground-truthed to ensure their reliability and accuracy.

She believes that growers, service providers, and research institutions must collaborate to accelerate the development and adoption of new technologies.

“We want to bring people around the table to solve the common problems of how we build proper models that are well-trained and well-tested,” she concluded.

This article is based on a presentation at the 2024 Hortgro Technical Symposium. Go to the Hortgro YouTube channel to watch Venter and other speakers at this event.