A University of Guelph research project has introduced a data-driven multi-layer perceptron model that significantly enhances the accuracy of land suitability prediction for various crops in Canada. The model uses advanced machine learning techniques to predict the land suitability of multiple crops, including barley, peas, spring wheat, canola, oats, and soy.
This innovative approach was developed by a team of Food from Thought-funded researchers, including Drs. Khurram Nadeem and Ayesha Ali, statistics professors in the College of Engineering and Physical Sciences and Aman Bhullar, a PhD candidate in the Department of Mathematics and Statistics.
Traditionally, land suitability models for agriculture in Canada have relied on single-crop inventories and expert opinion. However, these approaches are subjective and limited in precision, as they depend on the judgment of experts and often fail to incorporate objective data. To overcome these challenges, the research team utilized a data-driven approach, leveraging the power of machine learning algorithms.
“We took data about crop yields from 2013 to 2020 at a larger scale and made it more specific to individual farms by using information about the land, climate, and soil from Google Earth Engine to make our predictions more accurate,” said Dr. Nadeem.
He explained that the main idea of the research was to create a model that can understand how different crops relate to each other.
“We used a function that identifies different crops, so the model could learn about multiple crops at once and consider their specific needs,” he explained.
This approach made the predictions more precise, reducing the difference between predicted and actual crop yields by up to 2.82 times compared to older models that focused on only one crop.
The study revealed exciting findings about crop suitability in various regions of Canada. Barley, oats, and mixed grains were found to be more tolerant to soil-climate-landscape variations, indicating their suitability for cultivation in numerous areas across the country. On the other hand, non-grain crops displayed higher sensitivity to environmental factors.
Dr. Nadeem believes the implications of this research are significant for Canada’s agricultural sector.
“The multi-crop model developed through this project offers a more precise and data-driven approach to assessing land suitability for crop cultivation, especially in northern regions.”
He said the model could be integrated into cost-benefit analyses and facilitate better decision-making regarding using these lands for agricultural purposes.
While expert opinion-based and single-crop models have dominated land suitability prediction in the past, this research demonstrates the potential of data-driven approaches and multi-crop models. By harnessing the power of machine learning and considering the interconnectedness of different crops, the research team has paved the way for more advanced and accurate land suitability assessments in Canada and beyond.
The research project provides valuable insights and tools to support sustainable agriculture practices, enabling farmers and policymakers to make informed decisions based on reliable and accurate predictions of land suitability for various crops.
Read the full research study here: https://www.nature.com/articles/s41598-023-33840-6