A Fertilization Decision Model for Maize, Rice, and Soybean Based on Machine Learning and Swarm Intelligent Search Algorithms

Abstract

Background: The application of base fertilizer is significant for reducing agricultural costs, non-point source pollution, and increasing crop production. However, the existing fertilization decision methods require many field observations and have high prices for popularization and application. Methods: This study proposes an innovative model integrating machine learning (ML) and swarm intelligence search algorithms to overcome the above issues. Based on historical data for maize, rice, and soybean crops, ML algorithms including random forest (RF), extreme random tree (ERT), and extreme gradient boosting (XGBoost) were evaluated for predicting crop yield. Coupled with the cuckoo search algorithm (CSA), the prime fertilization decision model (FDM) was established to discover the optimal fertilization strategy. Result: For all three crops, the yield simulation accuracy of the ERT model was the highest, with an R2 and RRMSE of 0.749, 0.775, and 0.744, and 0.086, 0.051, and 0.078, respectively. Considering soil nutrient and fertilization characteristics as the determinants of yield and optimizing fertilization strategies, the proposed model can increase the average yield of maize, rice, and soybean in the study area by 23.9%, 13.3%, and 20.3%, respectively. Conclusions: The coupling model of ERT and the CSA constructed in this study can be used for the intelligent and rapid decision-making of the base fertilizer application for crops considered in the present study.

Simulating Maize Response to Split-Nitrogen Fertilization Using Easy-to-Collect Local Features

“…Random forest and XGBoost thus showed "substantial" performance for maize yield prediction [69]. In previous studies, the accuracy of XGBoost to predict grain yield was found to be R 2 = 0.66-0.81 and RMSE = 0.92-1.68 Mg ha −1 [37,61,71]. In comparison, the accuracy (R 2 ) of decision-tree models applied to healthcare and genomics ranged from 80 to 88% [72,73].…”

“…The prediction accuracy of EONR is generally lower than grain yield prediction. Previous studies reported R 2 values of 0.36-0.86 and RMSE of 33-57 kg N ha −1 [18,37,71]. Indeed, the selected functions and their accuracies vary widely among sites [13] potentially injecting bias and errors [21].…”

A reasoning method for rice fertilization strategy based on spatiotemporal knowledge graph

“…These models incorporate complex parameters that are difficult to model, lack generalizability, and cannot be dynamically updated (Liu et al, 2019). Mathematical models generally take soil nutrient test results or remote sensing images as data sources and establish fertilizer effect equations based on statistical regression (Wang et al, 2020), machine learning (Gao et al, 2023), and deep learning (Escalante et al, 2019). They tend to be objective but offer poor interpretability and low reuse.…”

Forecasting costs and risk of corn and soybean crops through Monte Carlo simulation

“…According to current projections, it is likely that the high prices of soybeans and corn will continue during the marketing period of the 2022/2023 crops [6]. Given this scenario, analyses, comparisons, and forecasts of production costs in agribusiness have been utilized as tools for decision-makers [7][8][9]. Pitrova et al (2020) [10] and Amorim et al (2020) [11] mentioned that computational simulation has proven to be an appropriate support tool for decision-making on rural properties.…”