ABSTRACT Sustainable agricultural productivity relies on soil nutrient management (SNM), although traditional methods of periodic soil testing and applying uniform amounts of fertilizer usually do not account for spatial and temporal variability. The latest breakthroughs in the technologies of artificial intelligence (AI) and Internet of Things (IoT) enabled data‐driven solutions for soil nutrient monitoring and decision‐making. This paper summarizes the results of 82 peer‐reviewed papers on the topic to critically analyse the use of AI and IoT in SNM, focusing on the characteristics of datasets, modelling, sensing methods, as well as constraints related to practical implementation. The literature review demonstrates that AI‐IoT systems have the potential to enable better nutrient management by optimizing the timing and application of fertilizers due to changing soil and environmental factors. Nevertheless, one key conclusion is that the performance of the system is not limited by the sophistication of the algorithms, but more by the quality of data, sensing realism and the extent of its coverage. Machine‐learning (ML) models tend to have more reliable and portable behaviour compared to deep‐learning (DL) models with less well‐organized but heterogeneous and proxy‐based datasets, which are characteristic of modern applications. Although the technologies of IoT allow high‐frequency observations, the majority of systems are based on the indirect indicators of the nutrient condition and the low levels of sensor density, which reduces the implementation of the technology in the field. The environmental benefits are reported mainly on the basis of decreased losses of nutrients, and the effects of climate are not quantified adequately. In general, the synthesis points at the necessity to prioritize data‐focused design, scalable sensing policies and decision relevance to support AI‐IoT‐based SNM.
Sabharwal et al. (Tue,) studied this question.