Improved SVM-Based Soil-Moisture-Content Prediction Model for Tea Plantation

• Published in: Plants (Basel) Vol. 12; no. 12; p. 2309
• Main Author: Huang, Ying
• Format: Journal Article
• Language: English
• Published: Switzerland MDPI AG 14.06.2023; MDPI
• Subjects: Accuracy; Agricultural production; Algorithms; Analysis; Artificial intelligence; Biology; China; Crop diseases; Crop production; Crop yield; Crop yields; data collection; Deep learning; Drought; Environmental impact; Farms; Geographical locations; Haliaeetus leucocephalus; humidity; hyper-parameter optimization; Irrigation; Irrigation practices; Irrigation scheduling; Irrigation water; labor; Learning algorithms; Machine learning; Management; Measurement techniques; Medical diagnosis; Model accuracy; Moisture content; Moisture effects; Morphology; Neural networks; Optimization; Performance prediction; Plant diseases; Plant sciences; Plantations; prediction; Prediction models; public ownership; Rain; Rainfall; Regression analysis; Resource management; Resource scheduling; Search algorithms; Soil improvement; Soil moisture; soil moisture prediction; Soil temperature; Soil types; Soil water; soil water content; support vector machine; Support vector machines; Tea; Tea (Plant); Tea industry; temperature; Water; Water consumption; Water content; Water resources management; Water use; water utilization; China; hyper-parameter optimization; support vector machine; soil moisture prediction
• ISSN: 2223-7747; 2223-7747
• DOI: 10.3390/plants12122309

Accurate prediction of soil moisture content in tea plantations plays a crucial role in optimizing irrigation practices and improving crop productivity. Traditional methods for SMC prediction are difficult to implement due to high costs and labor requirements. While machine learning models have been applied, their performance is often limited by the lack of sufficient data. To address the challenges of inaccurate and inefficient soil moisture prediction in tea plantations and enhance predictive performance, an improved support-vector-machine- (SVM) based model was developed to predict the SMC in a tea plantation. The proposed model addresses several limitations of existing approaches by incorporating novel features and enhancing the SVM algorithm’s performance, which was improved with the Bald Eagle Search algorithm (BES) method for hyper-parameter optimization. The study utilized a comprehensive dataset comprising soil moisture measurements and relevant environmental variables collected from a tea plantation. Feature selection techniques were applied to identify the most informative variables, including rainfall, temperature, humidity, and soil type. The selected features were then used to train and optimize the SVM model. The proposed model was applied to prediction of soil water moisture in a tea plantation in Guangxi State-owned Fuhu Overseas Chinese Farm. Experimental results demonstrated the superior performance of the improved SVM model in predicting soil moisture content compared to traditional SVM approaches and other machine-learning algorithms. The model exhibited high accuracy, robustness, and generalization capabilities across different time periods and geographical locations with R2, MSE, and RMSE of 0.9435, 0.0194 and 0.1392, respectively, which helps to enhance the prediction performance, especially when limited real data are available. The proposed SVM-based model offers several advantages for tea plantation management. It provides timely and accurate soil moisture predictions, enabling farmers to make informed decisions regarding irrigation scheduling and water resource management. By optimizing irrigation practices, the model helps enhance tea crop yield, minimize water usage, and reduce environmental impact.