Proximal sensing of soil properties for nitrogen use efficiency improvement

Agricultural production systems must adapt to meet the demands of a rapidly growing population while also being sustainable. The adoption of improved crop varieties, pesticides, and synthetic fertilizers during the Green Revolution facilitated agricultural production, but the rate of productivity increase has slowed. Precision agriculture, which utilizes information technology, sensors, and control systems, is a promising strategy for improving agricultural efficiency, productivity, and sustainability.Characterization of the spatial variability of relatively fixed soil properties that have a strong influence on NUE is needed to make informed management decisions at low cost and with potentially high spatial and temporal resolution, therefore sensor development for better NUE of winter wheat is needed. Proximal sensing, specifically optical sensing potentially provides a route to obtaining high spatial and temporal resolution data that can be interpreted to understand the underlying control of NUE across a field. This research focused on determining whether a simple, universal optical sensor can be developed to map areas of different NUE potential, i.e., areas where the relatively stable soil properties are likely to be, on average, associated with high, medium or low NUE. This thesis focused on evaluating the potential to use proximal optical sensing to collect data related to texture and organic matter and whether these properties can be used to define inherent limitation to NUE. An NUE growth trial was conducted to assess the impact of soil texture and organic matter on NUE. The results showed that while soil texture and organic matter had a significant impact on dry matter accumulation, it was the addition of organic matter that increased NUE in loam soil. The loam soil had inherently higher NUE compared to the sandy loam and clay soils used for the experiment. The second study used bottom-up and top-down approaches to identify diagnostic wavebands for predicting texture and organic matter. The results revealed that the NIR region, especially 1400, 1900, and 2200 nm, are the three dominant wavebands for predicting clay and organic matter, which were consistent in both bottom-up and top-down approaches. Clay and organic matter contents have high predictability of around 0.690 – 0.890. Sand and silt contents have lower model accuracy with about 0.600 R2 for sand content. The results however show that these can be useful for sensor development mainly focusing on clay and organic matter content. The last part investigated whether spectral data could be used to classify soils to grow winter wheat in terms of inherent limitations to NUE. The results show that the identified wavebands can separate the three textural classes used for the NUE pot experiment with >90% accuracy and identify clusters in field soil samples. These wavebands can also detect organic matter content in the soil however the models did not perform as well with >80% accuracy. The nitrogen and fertilizer treatments were also not distinguished by the spectra. The linear SVM had the best classification results for both soil texture and organic matter. In conclusion, the development of a simple, universal optical sensor for mapping areas of different NUE potential in winter wheat production systems is demonstrated in this thesis. The study shows that NIRS has the potential to provide high spatial and temporal resolution data that can be interpreted to understand the underlying control of NUE across a field. This approach has the potential to reduce adverse environmental impacts and improve agricultural efficiency and productivity. Further research could explore the potential for scaling up this approach for use in commercial farming operations.