Assessment of Soil Fertility Using Induced Fluorescence and Machine Learning.

Techniques such as proximal soil sampling are investigated to increase the sampling density and hence the resolution at which nutrient prescription maps are developed. With the advent of a commercial mobile fluorescence sensor, this study assessed the potential of fluorescence to estimate soil chemical properties and fertilizer recommendations. This experiment was conducted over two years at nine sites on 168 soil samples and used random forest regression to estimate soil properties, fertility classes, and recommended N rates for maize production based on induced fluorescence of air-dried soil samples. Results showed that important soil properties such as soil organic matter, pH, and CEC can be estimated with a correlation of 0.74, 0.75, and 0.75, respectively. When attempting to predict fertility classes, this approach yielded an overall accuracy of 0.54, 0.78, and 0.69 for NO3-N, SOM, and Zn, respectively. The N rate recommendation for maize can be directly estimated by fluorescence readings of the soil with an overall accuracy of 0.78. These results suggest that induced fluorescence is a viable approach for assessing soil fertility. More research is required to transpose these laboratory-acquired soil analysis results to in situ readings successfully.

On-Farm Experimentation to transform global agriculture.

Restructuring farmer-researcher relationships and addressing complexity and uncertainty through joint exploration are at the heart of On-Farm Experimentation (OFE). OFE describes new approaches to agricultural research and innovation that are embedded in real-world farm management, and reflects new demands for decentralized and inclusive research that bridges sources of knowledge and fosters open innovation. Here we propose that OFE research could help to transform agriculture globally. We highlight the role of digitalization, which motivates and enables OFE by dramatically increasing scales and complexity when investigating agricultural challenges.

Joint measurements of NDVI and crop production data-set related to combination of management zones delineation and nitrogen fertilisation levels.

This research compared four nitrogen (N) management strategies (uniform N rate: UR, variable N rate based on crop proximal sensing: VR-PS, variable N rate based on management zones: VR-MZ and variable N rate based on integrating crop sensing and MZ: VR-PSMZ), evaluating their effect on maize grain yield, partial factor productivity (PFPN), and net return above N fertiliser cost (RANC). The study provided a practical tool for choosing the fertilisation strategy that best performs in each agro-environment. These datasets are a supplementary material to the research paper by [3]. Data were collected over seven site-years experiments conducted in North-Eastern Colorado (USA). In dataset 1, for each site-year, data includes geo-referred points where grain yield and Normalised Difference Vegetation Index (NDVI) were measured, each one associated with its respective N rate, management zone (MZ), PFPN, RANC, and N management strategy. In order to group the observations reflecting homogeneous crop vigour, NDVI values were clustered within NDVI classes. In dataset 2, the main soil properties measured in several geo-referred points in each location are provided.

Spatial management strategies for nitrogen in maize production based on soil and crop data.

Nitrogen (N) fertilisation determines maize grain yield (MGY). Precision agriculture (PA) allows matching crop N requirements in both space and time. Two approaches have been suggested for precision N management, i.e. management zones (MZ) delineation and crop remote and proximal sensing (PS). Several studies have demonstrated separately the advantages of these approaches for precision N application. This study evaluated their convenient integration, considering the influence of different PA techniques on MGY, N use efficiency (NUE), and farmer's net return, then providing a practical tool for choosing the fertilisation strategy that best applies in each agro-environment. A multi-site-year experiment was conducted between 2014 and 2016 in Colorado, USA. The trial compared four N management practices: uniform N rate, variable N rate based on MZ (VR-MZ), variable N rate based on PS (VR-PS), and variable N rate based on both PS and MZ (VR-PSMZ), based on their effect on MGY, partial factor productivity (PFPN), and net return above N fertiliser cost (RANC). Maize grain yield and PFPN maximisation conflicted in several situations. Hence, a compromise between obtaining high yield and increasing NUE is needed to enhance the overall sustainability of maize cropping systems. Maximisation of RANC allowed defining the best N fertilisation practice in terms of profitability. The spatial range in MGY is a practical tool for identifying the best N management practice. Uniform N supply was suitable where no spatial pattern was detected. If a high spatial range (>100 m) existed, VR-MZ was the best approach. Conversely, VR-PS performed better when a shorter spatial range (<16 m) was detected, and when maximum variability in crop vigour was observed across the field (range of variation = 0.597) leading to a larger difference in MGY (range of variation = 13.9 Mg ha-1). Results indicated that VR-PSMZ can further improve maize fertilisation for intermediate spatial structures (43 m).

Effects of preprocessing of ultraviolet-induced fluorescence spectra in plant fingerprinting applications.

Preprocessing is an important step in data analysis. Dealing with spectral data, normalization is mandatory in order to compare items collected under various conditions. This paper addresses normalization of frontface fluorescence spectroscopy data where spectra are affected by an unknown multiplicative effect. The usual methods for reducing multiplicative problems are reviewed and a more detailed analysis of the normalization by closure is provided based on data on the fluorescence of plants as a means for plant species fingerprinting. As normalization is essentially the reduction of information, some methods of carrying it out are likely to remove either meaningful or discriminant pieces of information. As a result, it is demonstrated that normalization by closure should be performed using spectral data in a range where the spectra contain no information relevant to the problem at hand. This applies provided that in this range the signal-to-noise ratio is high enough. When the noise level is too high, a compromise should be found between preserving useful information and limiting the amount of noise introduced by the normalization procedure. Even if this study were carried out using fluorescence spectra, the overall process is likely to be applied to other spectral data.