Predicting total upland sediment yield using regression and machine learning models for improved land management and water conservation

• Published in: Journal of hydroinformatics Vol. 26; no. 8; pp. 2045 - 2058
• Main Authors: S. M., Shravan Kumar, Pandey, Manish, Umamahesh, N. V.
• Format: Journal Article
• Language: English
• Published: London IWA Publishing 01.08.2024
• Subjects: Accuracy; Agricultural production; Artificial intelligence; Artificial neural networks; Cellular automata; Climate change; Correlation coefficient; Correlation coefficients; Data analysis; Datasets; Environment models; Environmental modeling; Hydrology; Land conservation; Land management; Land use planning; Learning algorithms; Machine learning; Methods; Neural networks; Performance evaluation; Performance prediction; Principal components analysis; Regression analysis; Regression models; Sediment; Sediment transport; Sediment yield; Soil analysis; Soil erosion; Soil management; Topography; Transport phenomena; Variables; Water conservation; Water quality; Watershed management
• ISSN: 1464-7141; 1465-1734
• DOI: 10.2166/hydro.2024.159

In this study, various regression models were utilized to predict total sediment yield in tons, while their performance was evaluated for accuracy and reliability. The dataset utilized contains numerous predictors that have been standardized and processed through principal component analysis to improve model performance. Models evaluated here include linear regression, normalized linear regression, Principal Component Analysis (PCA), Pearson Correlation Coefficient (PCC) with generalized ridge regression, kernel ridge regression, multivariate regression, lasso regression approaches such as artificial neural network Cellular Automata-Artificial Neural Network (CA-ANN or ANN), and more. Results suggest that the ANN model achieved the lowest mean squared error (MSE), 113.641; this suggests superior predictive capability compared to other models. Although environmental data were complex and relationships complex, an ANN model showed less error, followed closely by CA-ANN with an MSE of 124.83. Traditional models such as linear or lasso regression revealed larger errors with negative squared values that indicated poor fits to data. This analysis highlights the effectiveness of advanced machine learning in environmental modeling, emphasizing the importance of selecting models suited to data and specific phenomena, aiding environmental planners in predicting and managing soil erosion and sediment transport.