Project Number: 5070-12610-005-004-S
Project Type: Non-Assistance Cooperative Agreement
Start Date: Sep 1, 2021
End Date: Aug 31, 2022
The primary objective is to improve upon traditional soil carbon modeling approaches by incorporating measurements from multiple soil sensors using machine learning techniques.
Step 1: Collection of soil samples and soil carbon measurements in conjunction with proximal soil sensor data, including near-infrared reflectance, apparent electrical conductivity, and penetration resistance. Step 2: Using the dataset developed in Step 1, a convolutional neural network (CNN) model will be developed to estimate soil carbon and bulk density. Step 3: Train the CNN model on a random subset of the reflectance data and test it on the remaining data using a high performance computer. A traditional model will be used as a baseline for performance. Step 4: On three fields under varying management, unmanned aerial vehicle (UAV) images will be collected and a digital elevation model (DEM) developed for each field using structure-from-motion algorithms . Near surface and profile soil samples will be collected, as well as the associated proximal soil sensing data at the designated sampling points in these fields. Step 5: The CNN model will be refined by incorporating the terrain attributes to improve the estimates of three-dimensional soil carbon stocks in the three fields.