Rainfall time series disaggregation in mountainous regions using hybrid wavelet-artificial intelligence methods.

Area of Science:

• Hydrology and Water Resources Engineering
• Environmental Science
• Data Science and Machine Learning

Background:

• Mountainous regions exhibit high spatial and temporal variability in rainfall, posing challenges for hydro-environmental studies.
• Lack of fine-scale rainfall data due to administrative and economic constraints necessitates rainfall time series disaggregation.
• Accurate rainfall simulation at various scales is crucial for effective water resource management in these areas.

Purpose of the Study:

• To develop and evaluate hybrid models for disaggregating monthly rainfall time series into daily series for mountainous regions.
• To compare the performance of proposed hybrid models (WLSSVM, WANN) against traditional methods (LSSVM, ANN, MLR).
• To enhance the accuracy of rainfall time series analysis for hydro-environmental applications in data-scarce mountainous areas.

Main Methods:

• Application of wavelet transform for decomposing rainfall time series data.
• Development of hybrid models combining wavelet decomposition with Least Square Support Vector Machine (LSSVM) and Artificial Neural Network (ANN).
• Utilizing mutual information and correlation coefficient criteria for selecting optimal sub-series as input for disaggregation models.

Main Results:

• The proposed hybrid models, particularly the Wavelet-Artificial Neural Network (WANN), significantly improved rainfall disaggregation accuracy.
• WANN model showed efficiency increases of up to 9.1% (Tabriz) and 4.5% (Sahand) compared to WLSSVM.
• Compared to LSSVM, ANN, and MLR, WANN demonstrated substantial improvements, up to 22% and 21.1% (vs ANN), 20% and 30.2% (vs LSSVM), and 50% and 53.3% (vs MLR) respectively.

Conclusions:

• Hybrid models integrating wavelet decomposition with LSSVM and ANN are effective for disaggregating rainfall time series in complex mountainous terrains.
• The WANN model offers a robust and accurate approach for generating daily rainfall data from monthly observations, outperforming other tested methods.
• These findings support the application of advanced data-driven techniques for improving hydro-environmental modeling in regions with limited rainfall data.