Date of this Version
Harun, H. (2023) A Robotic System For In-Situ Measurement Of Soil Total Carbon And Nitrogen (Master's Thesis). Biological Systems Engineering--Dissertations, Theses, and Student Research. Department of Biological Systems Engineering, University of Nebraska, Lincoln.
Surges in the cost of fertilizer in recent times coupled with the environmental effects of their over-application have driven the need for farmers to optimize the amount of fertilizer they apply on the farm. One of the key steps in determining the right amount of fertilizer to apply in a given field is measuring the amount of nutrients present in the soil. To ascertain nutrient deficiencies, most farmers perform wet chemistry analysis of soil samples which requires a lot of time and is expensive. In this research project, a robotic system was designed and developed that could autonomously move to predetermined GPS waypoints and estimate total carbon (TC) and total nitrogen (TN) content in the soil in-situ using visible and near-infrared reflectance spectroscopy - a faster and cheaper method to determine soil nutrients in real-time.
For the locomotion of the robotic system, a Husky robotic platform by Clearpath Robotics was used. A Gen2 robotic arm by Kinova Robotics was used for the precise positioning of the probe in taking soil spectral measurement. The probe was custom designed and built to be used in conjunction with the robotic arm as an end-effector. Two lightweight and inexpensive spectrometers by OceanInsight, namely, Flame VisNIR and Flame NIR+, were used to capture the spectral signatures of soil. The prediction was done with a spectroscopic calibration model and External Parameter Orthogonalization (EPO) was applied to remove the moisture effect from the soil spectra.
The robotic system was tested at University of Nebraska-Lincoln (UNL) NU-Spidercam phenotyping facility. Two sets of spectra were obtained from the field campaign: in-situ and dry-ground spectra. The dry-ground spectra were used as library scans and Partial Least Square Regression (PLSR) was used for the modeling. The in-situ spectra were randomly divided into EPO calibration and validation sets. Satisfactory results were obtained from the initial prediction on dry-ground validation set, with R2 (coefficient of determination) of 0.77 and RMSE (Root Mean Squared Error) of 0.15% for TC and R2 of 0.64 and RMSE of 171 ppm for TN. There was a reduction in R2 and an increase in RMSE values for both TC and TN when prediction was done directly on the in-situ validation set. For TC, the R2 dropped and RMSE increased to 0.25 and 0.29% respectively, and for TN, the R2 dropped and RMSE increased to 0.19 and 259 ppm respectively. This was primarily due to the presence of moisture in the field samples. The R2 increased to 0.62 and RMSE decreased to 0.2% for TC, and the R2 increased to 0.51 and RMSE decreased to 200 ppm for TN, when EPO was applied on both the in-situ validation and dry-ground sets. These findings highlight the importance of accounting for moisture effects in the prediction of soil properties using the robotic system and demonstrate the potential of the system in enabling soil monitoring and analysis in-situ.
Advisor: Yufeng Ge