Experimental and Machine Learning Hybrid Strategy for Optimizing Nitrogen Release from Chitosan-Functionalized Biogenic Silica Nanoparticle-Coated Urea Fertilizer

Abstract

The current world’s population increased three times than it was in the mid of the 20th century. The dramatic rise in population has imposed a tremendous load on agricultural industry to meet rising demand for food. The application of conventional fertilizers is one of the most common and effective strategies for raising crop yields. These fertilizers release essential plant nutrients, both organic and inorganic, into soil to promote the growth of crops. Nitrogenous fertilizers are utilized to address nitrogen deficiencies. Urea is one of the most commonly used nitrogenous fertilizers that contains 46% nitrogen. However, it has been observed that on average, more than 70% of this urea fertilizer is wasted in the localized region of crop fields, which is related to environmental contamination and longterm economic losses. Slow-release urea fertilizer was created in such a manner that fulfills plant needs according to the requirement for growth. Coating urea fertilizer with appropriate materials decreases its water solubility and slows its release in the soil. This research study will focus on the usage of chitosan functionalized silica nanoparticles for coating of urea granules. Urea granules were coated using fluidized bed coater. Different formulations were made, and their release rates were calculated in water using UV-Visible Spectroscopy. Scanning Electron Microscopy (SEM) was used to check the surface structure of synthesized nanoparticles and coated granules. Fourier Transform Infrared Spectroscopy, Xray diffraction and crushing strength were employed to identify the nature of chemical bonds, the structural parameters and shelf life of the prepared samples. The nitrogen release from coated urea is a complex phenomenon. In the present study, machine learning models were employed to optimize the release of nitrogen from coated urea fertilizers. Data gathered from the literature was used to train four machine learning models i.e. Decision Trees, Gaussian Process Regression, Ensembled Learning Trees and Support Vector Machines. Particle Swarm Optimization and Genetic Algorithms were also combined with the machine learning models. The results suggest that Gaussian Process Regression combined with GPR is favored for optimizing the nitrogen release (R2 ~ 0.9766 and RMSE ~ 0.1215). In addition, a Graphical User Interface had been developed using the optimized GPR to facilitate the calculation of Release Time.