Researchers working on smart irrigation systems have developed a way to choose the most accurate weather forecast, from those offered in the week leading up to a given day.
Dr. Eric Wang, an Internet of Things researcher at James Cook University (JCU) in Cairns, works on technology that allows farmers to make data-driven decisions.
“Every farmer would like to have a perfect weather forecast, but accurate forecasts are even more important to those who are embracing technology, and especially the Internet of Things (IoT),” Wang said.
In agriculture, the Internet of Things included smart devices that talked to each other to make recommendations such as when, where and how much to irrigate, Wang said.
“This decision requires a lot of information, such as the needs of a particular crop, the current stage of its development, soil moisture and of course the weather.”
Researchers had looked for ways to go beyond standard weather forecasts, such as the Bureau of Meteorology (BOM) seven-day forecast, to help farmers and their smart systems decide if they need irrigation today, Wang said.
Under the supervision of Wang at JCU and Professor Wei Xiang at La Trobe University, PhD candidate Neethu Madhukumar has created a hybrid system that shows real promise in improving the accuracy of rainfall forecasts.
Madhukumar was learning probability theory before starting her doctoral studies, and said she had more math in weather forecasting than most people understood.
“When weather forecasters say they have consulted models, it involves feeding data from satellites and sensors into mathematical models based on the physics of how air, heat and humidity behave,” she said.
Forecasters also apply the judgment and experience of on-duty experts, so instead of trying to rediscover the wheel, Madhukumar’s goal was to find a way to determine the best forecast of those offered by climate models, in the week that led to the day in question Me
“You could assume that the forecast closest to the day in question would be the most reliable, but that turned out not to be the case,” she said.
“So we looked at ways to teach our artificial neural network to understand the relationships underlying all the data, to choose the best prediction.”
Madhukumar has developed a hybrid climate learning model (HCLM), which learns from a combination of climate model data and the final answer to the question: will it rain tomorrow?
First, a probability-based network estimates multiple predictions for different precipitation models.
Then a deep learning neural network reprocesses predictions to produce a better prediction for the next day.
“This combination of distilling knowledge from climate models and using an in-depth learning network to improve forecasting has not been proven before,” Wei Xiang said.
“Using high-quality processed data from the Bureau of Meteorology, instead of raw observations, has helped HCLM learn better.”
Madhukumar said the neural network examined the relationships between massive amounts of input data, processes them through many layers of the network, and learns from mistakes made in previous predictions.
“The higher the quality of the data entered, the better the network learns.”
“We trained the hybrid system by uploading 123,640 data items, representing two years of BOM forecast and weather data for 10 countries in Australia ‘s six major climatic zones.
“When we then tested our system in the same range of climate zones, the hybrid model surpassed the BOM climate models and three other experimental systems, making fewer errors in its predictions.”
Researchers were inclined to point out that their work would not leave BOM out of business.
“This work relies on their expertise and HCLM builds rainfall forecasts on the numerous forecasts produced by BOM climate models,” Wang said.
“We believe this model is the first to unite climate models, a probability network and a neat learning network.
“Our next task will be to work on the next question that every farmer has – if it rains tomorrow, how likely are we to get it?”
The research is published in the IEEE Internet of Things Journal.