Practical AI on farms
The combination of artificial intelligence (AI), machine learning and data is opening the door to a new wave of decision-making tools for farmers. One of the major advancements lies in the collection and analysis of data at an individual farm or paddock level.
Goanna Ag CEO, Jay Jalota says they have developed AI solutions for three key areas in cotton production – planting, irrigation and spraying. Based in Goondiwindi, Goanna Ag has been providing on-farm data services for 20 years.
“AI is essentially the ‘brain’ of this technology – it can analyse and reflect on information, and adapt or learn from new information as it is added,” he says. “This adaptation is called machine learning, where historical and new data is provided to the AI for it to analyse, look for patterns and make conclusions without human bias.”
The third component is the data itself. For the AI to make valuable and accurate predictions, it needs large and relevant data sets.
“It takes time to collect detailed data sets across a variety of seasonal conditions,” says Jay. “As the data sets expand and include information from a variety of real-world scenarios, the ‘smarter’ the AI becomes. The more site-specific the data is, the more accurate the predictions generated will be, allowing for more informed decisions.”
Making FastStart more site-specific
Growers have been using the FastStart guidelines for more than 10 years to choose the optimal planting time for their area. For optimal cotton crop establishment and vigorous early growth, the soil temperature at 10 cm depth needs to be above 16–18°C at 8 am (AEST), and forecast to rise for the week after planting.
“There’s no problem measuring the soil temperature, but there are a number of unknowns related to the seven day temperature forecast,” says Jay. “With support from CSD and Syngenta, we have created a unique, site-specific soil temperature forecasting tool that can predict soil temperature for a rolling seven-day period with 90 per cent accuracy.”
Using over 200 million historical data points from the 100 plus weather stations in the FastStart network, the Goanna Ag team trained the AI to recognise patterns and nuances in the relationship between soil temperature and the surrounding environment (air temperature, humidity, solar radiation etc) at each weather station site. This is the first step away from a ‘one-size fits all’ prediction and moving toward a site-specific prediction model.
Once the AI training was complete, the team began ‘feeding’ the AI real-time data from the FastStart weather stations and the Bureau of Meteorology’s gridded weather forecast. Using a feedback loop they could continuously ‘predict soil temperature, measure soil temperature and repeat’ at each site and in just a few months the AI was predicting soil temperature seven days in advance and within 0.5°C with 90 per cent accuracy. The accuracy of these predictions will improve over time as the AI continues to learn from the new data collected each season.
This AI feature was released in September 2024 on the CSD website, where it was consulted approximately 6000 times during September and October for planting decisions. An analysis of the FastStart planting date compared to the user-reported planting date at 308 GoField sites for the 2023–24 season showed that approximately 15 per cent of fields were planted before optimal conditions were established.
“This can lead to issues with slow crop establishment and potentially needing to replant,” says Jay. “The AI-enhanced FastStart predictions will increase grower confidence that the optimal soil temperature conditions are in place and will result in more fields being planted at the optimal time each season.”
Smarter irrigation scheduling
Goanna Ag’s GoField irrigation scheduling service was launched 20 years ago and is used by around 60 per cent of Australian cotton growers, advising growers on about 20,000 irrigation events each year.
For every irrigation event, GoField considers about 7000 site-specific data points, including weather, soil moisture, plant canopy temperature, satellite images and crop evapotranspiration (ETc) models.
With access to AI and machine learning capability, GoField is analysing data collected during successive seasons on growing day degrees, crop stages, satellite imagery, plant canopy temperatures, soil moisture, weather forecasts, and the like. This, along with client feedback and advances in irrigation and crop science, means that GoField can provide growers with increasingly robust and reliable recommendations and forecasts for each field.
Safe spraying decisions
Spray drift is a serious issue for Australian growers, with various widespread incidents occurring each season. The WAND (Weather And Networked Data) network is a world-leading Australian technology that predicts hazardous inversion events. With over 4200 subscribers across NSW and Queensland, WAND helps spray operators avoid spraying in conditions that could result in long-distance spray drift.
Compared to traditional methods for identifying inversions, the WAND forecasts give spray operators an extra four hours of ‘safe-to-spray’ time per day, on average.
The extensive research behind the technology initially provided a two-hour forecast with an accuracy of about 85 per cent. Since the deployment of WAND, Goanna Ag has used AI and machine learning to extend the forecast period to 24-hours with a similar level of accuracy.
“The 102 WAND towers are continually collecting data, and we used over 35 million data points to train the AI to make 24-hour forecasts, which are of more practical use when planning and arranging the logistics of spray operations,” says Jay.
Find out more: www.goannaag.com.au
Goanna Ag execs Jay Jalota (CEO) (left), founder Tom Dowling (COO) and John Pattinson (CDO).
GA sensor.
GA data sets and WAND forecast.
Are hitchhiker pests an increasing threat to Australian biosecurity? Strategic cover cropping to control cotton root disease
