Overfertilization with nitrogen fertilizers has damaged the environment and health of soil; yields are declining, while the population continues to rise. Soil is a complex, living organism which is constantly evolving, physically, chemically and biologically. Standard laboratory testing of soil to determine the levels of nitrogen (mainly NH₄⁺ and NO₃^-) is infrequent as it is expensive and slow and levels of nitrogen vary on short timescales. Current testing practices, therefore, are not useful to guide fertilization. We demonstrate that Point-of-Use (PoU) measurements of NH₄⁺, when combined with soil conductivity, pH, easily accessible weather (in this study, we simulated weather in the laboratory) and timing data (i.e. days passed since fertilization), allow instantaneous prediction of levels of NO₃^- in soil with of R²=0.70 using a machine learning (ML) model (the use of higher-precision laboratory measurements instead of PoU measurements increase R² to 0.87 for the same model). We also show that a long short-term memory recurrent neural network model can be used to predict levels of NH₄⁺ and NO₃^- up to 12 days into the future from a single measurement at day one, with R²_NH4+ = 0.64 and R²_NO3- = 0.70, for unseen weather conditions. To measure NH₄⁺ in soil at the PoU easily and inexpensively, we also developed a new sensor that uses chemically functionalized near ‘zero-cost’ paper-based electrical gas sensors. This new technology can detect the concentration of NH₄⁺ in soil down to 3±1ppm (R²=0.85). Gas-phase sensing provides a robust method of sensing NH₄⁺ due to the reduced complexity of the gas-phase sample. Our machine learning-based approach eliminates the need of using dedicated, expensive sensing instruments to determine the levels of NO₃^- in soil which is difficult to measure reliably with inexpensive technologies; furthermore, crucial nitrogenous soil nutrients can be determined and predicted with enough accuracy to forecast the impact of climate on fertilization planning, and tune timing for crop requirements, reducing overfertilization while improving crop yields.

中文翻译：

---

使用使用点传感器工具包和机器学习模型确定和预测土壤化学

氮肥过量施肥损害了土壤的环境和健康；单产下降，而人口继续增加。土壤是一种复杂的生物，在物理，化学和生物学上不断发展。土壤的标准实验室测试，以确定氮的水平（主要是NH ₄ ⁺和NO ₃ ^- ）是罕见的，因为它是昂贵的和缓慢的和氮的水平上短时标变化。因此，当前的测试方法对指导施肥没有帮助。我们证明了NH ₄ ⁺的使用点（PoU）测量当与土壤电导率，pH值，容易获得天气组合（在本研究中，我们在实验室中模拟天气）和定时数据（即天自受精通过），允许NO水平的瞬时预测₃ ^-在土壤中有R的²使用机器学习（ML）模型，该值= 0.70（对于同一模型，使用高精度实验室测量值而不是PoU测量值，可使R ²增加到0.87）。我们还表明，长短期记忆回归神经网络模型可用于预测NH水平₄ ⁺和NO ₃ ^-长达12天到未来从在一天一个单次测量，其中R ² _{NH 4 +} = 0.64和[R² _NO3- = 0.70，用于不可见的天气情况。为了方便，廉价地在PoU上测量土壤中的NH ₄ ⁺，我们还开发了一种新的传感器，该传感器使用化学功能化的接近“零成本”的纸质气体电子传感器。这项新技术可以检测低至3±1ppm（R ² = 0.85）的土壤中NH ₄ ⁺的浓度。由于减少了气相样品的复杂性，气相传感提供了一种可靠的NH ₄ ⁺传感方法。我们基于机器学习的方法无需使用专用的昂贵传感设备来确定NO ₃的含量^-在难以用廉价技术可靠测量的土壤中；此外，可以以足够的精度确定和预测重要的氮肥土壤养分，以预测气候对施肥计划的影响，并调整作物需求的时机，减少过量施肥的同时提高作物的产量。