DigiFarm

The benefit of digitising Australian agriculture was estimated at $20.3bn annually.  Agriculture was also said to lag other industries in digital adoption (see https://www.crdc.com.au/precision-to-decision).  Landcare administered grants for ‘smart farming partnerships’ in an attempt to evaluate technologies for farm readiness and to improve adoption. From this, the University of Sydney’s Digifarm at Narrabri was born.

Digifarm is a collection of subprojects that combines research with the Landcare mission of farmer extension.  By utlising both extension and research, we are looking at whether technologies are suitable now, and also whether they are likely to remain suitable in the future.

The University of Sydney has been in Narrabri for 60 years operating a distinguished wheat breeding program.  It has a 2600ha property used for cropping and beef production.  It also runs around 40 000 crop trial plots per year and contains remnant native forest.

In the last year we implemented AgWorld for our cropping tasks, and Agriwebb for our livestock.  We also use PCT-Ag to evaluate crop spatial information, and Cibo Labs for pasture spatial information. 

Precision agriculture for crops has been around for 20 years or more, though is still not widely adopted.  Mapping nutrient requirements and ‘yield’ of a pasture (the animal) is much more difficult.  Recently, satellite products have started estimating biomass and cover at a broad scale.  We use Cibo Labs, another Australian company that focuses on dryland to arid pastures. 

We are also looking at improving the estimates by measuring at ground level, using a variety of technologies; cattle GPS eartags, vegetation estimates via infrared camera (NDVI), sonar, lidar, and hyperspectral sensors.

Farms comprise more than 50% of the Australian land surface, and a much higher proportion of specific habitats.  Farmers are therefore custodians of much of Australia’s wildlife. We have a grid of 20 camera, 20 audio, and 12 bat-echo traps across the farm, to quantify how wildlife are using the different farm landscapes.  Additionally, we have conducted human surveys of birds, flora, and invertebrates.  We have installed smart traps for mirids, a pest of cotton, which evaluate their numbers on a daily basis.

In the future, it is predicted that landowners will be able to buy a sensor at an affordable price, place it somewhere on their land, and receive regular updates of the number and diversity of birds and bats simply by email.  Making this a reality requires the training of AI algorithms to recognise calls and extract meaningful information

Soil moisture models for dryland cropping can improve predictions of future yield, and thereby help farmers to spend the right amount of money on inputs (seed, fertiliser) to optimise profits.  Good models combine big data from the Bureau of Meteorology and satellites, to local data from on-farm soil moisture probes and rain gauges.  Digifarm has installed a large number of soil moisture probes, with the intention of (1) improving predictive models, and (2) evaluating the right number of soil moisture probes needed in dryland systems.  Hence we have evaluated sensors from different suppliers, found some of the practical issues of maintaining them on an active farm, and are developing simulations that will find their way into yield prediction applications.  We’re also playing around with different ways to present live data.

We have been using drone, satellite, and ground data to improve the APSIM plant growth model (https://www.apsim.info/apsim-model) of wheat under different levels of nitrogen

We will soon (before end of March) have a robot operating over crop and pasture plots.  The robot is a generic platform for trying out different ideas for automating data collection and some actions.  Initially we’ll be using it to operate a sprayer that is equipped with AI to recognize weeds, and with sensors to evaluate biomass and plant quality. (see https://agerris.com/the-digital-farmhand)

We are using moisture probes, soil analysis, and drone surveys, to study cover crops, and are beginning a study of natural sequence farming in our creek pastures.  We also have trials of native grains, with the aim of finding a suitable balance between production and natural ecosystems.

Precision weed management is developing rapidly, with emerging AI algorithms and matching hardware.  Commercial products are available for green-on-brown (green weeds against bare soil) but green-on-green (weeds within crops) is harder.  The process involves identifying a weed in real-time as the tractor (or field robot) passes, and performing an action.  The most common action is spraying a herbicide, but we are experimenting with lasers, and mechanical options are also available.

Soil testing usually means sending something to a lab, but we are experimenting with simple tests that farmers can do in the field to quickly assess a soil.  The SLAKES app and the teabag index have been conducted across our farm and compared to prior maps of soil type.

We have several subprojects using cattle weighing systems, including in-yard walkover weighers, and in-field Optiweighs.  The latter is a young Australian company which has produced what is estimated to be the world’s first in-field cattle weighing system.  Cattle are attracted to the weigher by a lick block, and are not stressed, whereas they need persuasion to use a walkover weigher.  Producer gets a daily estimate of the herd weight, and can better plan pasture utilisation and time to sell.

We are just starting a subproject in which collars will be placed on feral pigs and tracked via satellite for approximately 2 years.  This is a community-level project, since we expect pigs will travel within a 10 to 50 km range.  The project is being managed and joint-funded by the Local Land Services.