AI4AgSys for MI Farmers

Our overarching goal is to formulate solution-oriented strategies and outcomes, helping landowners and managers create inspirational actions that meet diverse objectives in a changing climate and society.This team from four colleges of MSU aims at co-development of an AI-based forecasting system for 4 major types of agricultural production systems (AI4AgSys) in Michigan: corn, soybean, wheat, and potatoes. AI4AgSys is a large-scale, pre-trained deep learning forecasting system that will be trained with massive, diverse datasets, using self-supervised learning and multi-task learning techniques, where the system learns by predicting parts of the data from other parts, thereby reducing the need for extensive labeled data. Our specific objectives are:

Obj. 1: Co-design, co-assessment, and co-applications of AI4AgSys through four engagement workshops

Obj. 2: Constructing AI4AgSys that has two components: a mobile app and AI Foundation module;

Obj. 3: Developing rich geospatial time series databases supporting AI4AgSys; and

Obj. 4: Conducting field campaigns to collect needed data for AI4AgSys parametrization and validation at 30 farmlands in the southern part of Michigan’s Lower Peninsula.

We propose to use the eddy covariance (EC) method as our primary tool (see Chen et al. 2008) in making intensive, continuous measurements of NEP, water loss through evapotranspiration (ET) and energy balance at the six KBS-GLBRC “Scale-Up Fields”: switchgrass, restored prairie and continuous corn fields (two replicates of each system). A combination of dynamic chambers and microclimatic and biometric measurements will also be used to collect various needed data on soil C, resource use (water, light and N), carbon gain (photosynthesis) and loss (respiration) and changes in major biophysical variables as well as heat energy to achieve our study objectives.

We hypothesize that significant differences exist in ecosystem production, biophysical regulations and belowground carbon allocation among the three biofuel production systems. These differences are clearly reflected at multiple temporal scales.

The data and results produced from this study will be posted on an open Webpage in order to best promote their uses (e.g., modeling research of the Thrust 4 toward the missions of the DOE-BRC.

Objective 1:
Examine the magnitudes and dynamics of net ecosystem production (NEP) and soil C at the six “Scale-Up Fields” that have been identified at the Kellogg Biological Station (KBS).

Objective 2:
Investigate the biophysical regulations and resource use limitations for maximizing NEP and soil C at these sites.

Acknowledgements

Funded by the DOE

foxyform