William Michael Laprade: The climate is changing, drought is becoming more common and plant stress is increasing. Can we use multispectral imaging to better understand the microbiome of wheat plants in order to determine an appropriate treatment to improve plant resilience while reducing harmful pesticide and chemical fertilizer usage?
Nearly 20% of the worlds consumed calories come from wheat. As such, it is vital that we continue looking into ways to increase wheat plants’ resistance to stresses such as pests, disease, flooding and droughts. It has been found that the microbiome of plants and the soil they grow in play an essential role in providing a protective environment for the plant and increasing it’s resilience to various external stresses. Determining the microbiome health and understanding the role the different microorganisms play in the protective environment of the plant will allow us to find better ways to treat the plants.
Currently, we ensure high plant yield by making use of chemical fertilizers and pesticides during the plants growing season. What if, instead of adding pesticides we could just increase the ratio of a specific microbe in the plant to the same effect? Or instead of using strong chemical fertilizers we just increase the amount of a specific group of microbes? By treating the plants in this manner we could reduce plant stress and increase yield by simply altering the ratios of the microbes that already exist within the microbiome, resulting in higher yield and less pesticide and/or chemical fertilizer usage.
Agriculture scientists use multispectral imaging to determine plant health remotely via common measures such as the Normalized Difference Vegentation Index (NDVI) which is commonly used as a metric of plant health or the Normalized Difference Water Index (NDWI) which measures the water content of the plants. It is therefore suggested that we can use high spectral resolution images to provide some insight into the microbiome of the plant.
The aim of this project is to develop methods for analyzing and extracting information from these highdimensional images that correlates to the metabolomic and microbiomic information of the leaves. In order to efficiently extract as much information as possible from the images, we will investigate new and existing machine learning methods for dimensionality reduction, including techniques such as principle component analysis (PCA), autoencoders and generative adversarial networks (GAN). This extracted information will then be used downstream in a larger model for determining factors such as yield and overall plant health.