European Journal of Sustainable Agroecosystems

Context: Nitrogen is among the most difficult and important fertilizers in modern agriculture, but traditional fertilizer selection techniques often rely on generalized regional recommendations that do not consider field variability and often result in farmers sustaining losses due to under-application or polluting the environment by over-application of fertilizers. AI-based recommendation systems may be better at making predictions but are often criticized for their opaqueness and making farmers less inclined to trust and adopt them. 
Research Aim: This study states that the Explainable Artificial Intelligence (XAI) framework for local nitrogen recommendation combines the predictive power of AI with transparent decision-making needed for precision agriculture. 
Research Methods: The author collected field information from two growing seasons of supplying nitrogen and obtained data on soil fertility, weather, satellite images, plants, crop growth, nitrogen uptake, and management history. To predict the most suitable nitrogen recommendation, the author created AI systems and fitted them with an AI-based explainability module that provides information about the most substantial factors in the recommendation and derives an explanation for the provided recommendation. 
Findings: This Explainable AI system has achieved higher effectiveness in recommending and lower error rate in prediction in comparison to traditional recommending techniques, whilst analysis of factors for making recommendations has revealed nitrogen concentration in soil, indices of vegetation, and cumulative precipitation as the main variables influencing the results of the recommendations made. Implementation of the recommendations based on XAI has led to improvement in nitrogen efficiency of usage alongside obtaining equal or higher yield and reduction of the total amount of fertilizer used as compared with conventional agricultural practices. 
Significance: The findings prove that explanation can be used in the machine learning-based nutrient management without losses in terms of forecasts, which is a fundamental hurdle for farmers and agronomists to accept AI-supported decision making. Conclusions: Explainable AI is a sound and efficient way of introducing transparency in the area of localization of nitrogen management that will help achieve farmers’ goal of higher profitability from use of the equipment during crop production while simultaneously enhancing the efficiency of nitrogen use and creating benefits for the environment.