SBIR Phase I: Development of a multi-output machine learning modeling framework for a hybridized perennial cover crop for specialty crop systems

点击翻译按钮获取中文摘要

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project comes through the development of a prototype software modeling technology and associated monitoring methodology that will be used to transform the value proposition of farm-based practices that improve soil carbon sequestration. The solution enables verification of high-quality carbon offsets, provides feedback regarding important agronomic metrics, and can be applied on a large and geographically expansive scale. Specifically, this prototype is tailored to a novel species of hybridized grass planted as a sustainable, carbon-sequestering cover crop in specialty cropping systems. The solution's agronomic value proposition will enable farming operations to play a part in climate change mitigation while also improving profitability. This innovation will apply to millions of acres of specialty crop farms throughout the country, where cover cropping is currently underutilized despite environmental benefits. The solution will encourage a higher rate of adoption of the practice, lowering strain on farming budgets and natural resources. Additionally, the technology prototype will be applicable to other sustainable, land-based, carbon removal practices, extending its value-added potential to a wide variety of farming practices and circumstances.This SBIR Phase I project seeks to develop a machine-learning based, multi-output modeling software system to quantifying soil carbon sequestration associated with specified farming practices and other important agronomic metrics. The farming practice modeled by this prototype is a hybridized, perennial, cool-season grass planted in specialty cropping systems as a cover crop. Cover cropping is paired with the practice of no-till to enable on-farm soil carbon sequestration, which is quantified by the software modeling prototype for third party verification. Inputs for the models come from a combination of remote and in situ monitored data, such as soil and biomass sample analysis, drone-captured multispectral imagery, and satellite imagery. The quantity and type of model inputs is determined by several factors, including the scalability of monitoring costs, their effect on the accuracy of the models, and the requirements of third-party protocols for verifying soil carbon models. Project tasks include the calibration of the monitoring methodology for new target variables, the addition of plant water status and plant nitrogen status models to a soil carbon model, and the delivery of a functional and scalable prototype capable of generating accurate predictive models of environmental changes attributed to the aforementioned cover crop.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.