One of the greatest threats to the future of Australia’s Great Barrier Reef is farm runoff of fertilizers and pesticides that damage the health of the underwater ecosystem. This typically happens because farmers are overwatering their crops. Anticipating rainfall and adjusting irrigation standards accordingly can drastically reduce runoff, save farmers precious resources, and increase crop yield.
But forecasting rainfall, especially on a hyperlocal scale, remains a challenge. Eric Wang, an Internet of Things (IoT) researcher at James Cook University (JCU) in Cairns, Australia, and Ph.D. student Neethu Madhukumar have been developing a hybrid system that pulls data from a number of sources to accurately determine how much farmers should irrigate on a given day.
First, a probability-based network processes data from Australia’s Bureau of Meteorology (BOM), including historical data from the past three years and daily forecast data from local stations and satellites. It also pulls from sensors and monitors set up on partipating farms that measure soil moisture, humidity, temperature, and other factors.
The results from the probability-based network are then pitted against a deep-learning neural network that reprocesses all of the information to provide a final forecast. The system then goes the extra mile to create a simulation that can provide recommendations to the farmer on how much to water based on the crop’s growth rate, which in this case is sugarcane, for the best possible outcome. This semi-autonomous system saves farmers from intense calculations and ultimately results in higher crop yields, according to the researchers.
All of this number crunching requires major supercomputing power and it’s all about finding the sweet spot between what the skies will do and what the crop needs.
“The previous day’s forecast is not always the most reliable,” Wang explained. “Sometimes the best information comes out of the forecast three or even four days prior. It really depends on the time of the year and what the weather pattern looks like. Refining and training the neural network to respond to these subtle changes is a real challenge.”
Wang and his team have been up against a number of challenges in getting this system to perform at the level that a farmer can put confidence into it and invest in it. First, they simply do not have enough data, notably cleaned high-quality data. The BOM only provides this level of quality data to the public going three years back. So, the team is mostly working with raw unprocessed data.
“The real-time data that is provided is not processed,” Wang said. “So there are a lot of errors, such as duplicate events, and we have to parse out those anomalies.”
The team will have to work with the BOM to access older data, which will take more time and resources. Creating a system that is hyperlocal is another challenge. Australia, like the U.S., has several climate zones, and how the system works in each zone varies. Furthermore, what happens across a zone or region can also vary greatly and a single forecast does not give a fully accurate picture of what a farmer may experience day-to-day.
“We are just starting to build a kind of framework,” Wang said. “The more data coming in, the better the system will be.”
The next stage in development will be creating a feedback loop for the system. By installing cameras on the farms, and increasing the number of sensors, they can further inform the system so that it can measure how the crop is actually performing, rather than just generating a simulation. They also want to make the system fully autonomous so that it can control and manage the irrigation, making it a hands-off operation for farmers.
There are about 10 farmers currently using the hybrid system, and Wang said they have seen higher productivity and greater savings on labor and electricity. These early adopters are helping the team develop the system so it is scalable for farms large and small.
Wang and his team are hoping to form partnerships and increase their funding so that they can continue to make the system cost-effective and accessible to farmers, and ultimately protect the environment.
Comments