Point-of-use sensors and machine learning enable low-cost determination of soil nitrogen

Abstract

Overfertilization with nitrogen fertilizers has damaged the environment and health of soil, but standard laboratory testing of soil to determine the levels of nitrogen (mainly NH₄⁺ and NO₃⁻) is not performed regularly. Here we demonstrate that point-of-use measurements of NH₄⁺, combined with soil conductivity, pH, easily accessible weather and timing data, allow instantaneous prediction of levels of NO₃⁻ in soil (R² = 0.70) using a machine learning model. A long short-term memory recurrent neural network model can also be used to predict levels of NH₄⁺ and NO₃⁻ up to 12 days into the future from a single measurement at day one, with $R^2_{{{\mathrm{NH}}_4}^+} = 0.60$ and $R^2_{{{\mathrm{NO}}_3}^-} = 0.70$, for unseen weather conditions. Our machine-learning-based approach eliminates the need for dedicated instruments to determine the levels of NO₃⁻ in soil. Nitrogenous soil nutrients can be determined and predicted with enough accuracy to forecast the impact of climate on fertilization planning and to tune timing for crop requirements, reducing overfertilization while improving crop yields.