Flow Series Generation from Water Depth Data Using Statistical and Machine Learning Models: The Tocache Station Case

Aguirre-Escalante C.; Quispe-Janampa D.; Ruiz-Rengifo L.; Dueñas-Tuesta M.; Díaz-Quintana E.; Reategui-Inga M. · Mathematical Modelling of Engineering Problems (2025) · DOI: 10.18280/mmep.120407

Resumen

The integration of spline models and Gaussian processes in hydrological studies represents an innovative approach, particularly for developing countries where flow data are scarce but water level records are more accessible. This study focuses on the ‘Bridge Tocache’ station in Peru, where a procedure was developed and validated to generate historical flow series using these two models. The spline model uses recorded water levels as input, while the Gaussian process model incorporates both water levels and the month of the year, capturing seasonal patterns. To address missing water level data, a stochastic process based on kernel density estimation was applied, segmenting the year into 24 fortnights. The models performed well, achieving coefficients of determination (R²) above 0.98. Statistical comparisons confirmed no significant differences between the generated and observed flow series in terms of mean, standard deviation, skewness, and kurtosis. The adaptability of these models lies in their independence from geographic-specific assumptions, making them generalizable to other hydrometric stations and regions with limited data availability. This methodology offers a cost-effective alternative for improving hydrological monitoring, reducing reliance on expensive equipment. Its broader applicability extends to sustainable water resource management, enhancing planning and decision-making in regions with limited monitoring capabilities.

Instituciones: Universidad Nacional Agraria de la Selva; Universidad Nacional Intercultural de la Selva Central Juan Santos Atahualpa