Loading...

Exploring the Efficiency of Machine Learning Algorithms in Estimating Time-Variant Travel Time Distributions in River Basins

Alizadeh Attar, Mehdi | 2022

173 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 55648 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Danesh Yazdi, Mohammad
  7. Abstract:
  8. This study aimed to investigate the efficiency of machine learning algorithms in capturing the dynamics of water particles' age distribution and in estimating the time series of median travel time (MTT) in river basins. The mechanism of solute transport in a catchment does not follow the hydrograph shape of the catchment and depends on the spatial and temporal distribution of solute resident time. The equations governing the age dynamics of water particles discharged from a catchment have been presented analytically in recent research. These equations showed that the transfer of solutes depends on various factors, including climatic conditions, basin topography, and basin vegetation. Also, the analytical solution of these equations depends on the StorAge selection function, representing the ratio between the travel time probability distribution function (TTD) and the residence time probability distribution function of water particles. These functions depend on the physical characteristics of the basin, such as climatic conditions, vegetation, and topological conditions. Therefore, quantifying the StorAge selection function using numerical and field studies has many challenges. The most important is obtaining a comprehensive model for estimating the TTD of water particle. Furthermore, research on rainfall-runoff models has shown that machine learning algorithms have shown a promising performance in reducing computational costs, estimating nonlinear functions, and finding complex relationships between governing variables that control the rainfall-runoff process. Due to the similarity of the variables affecting TTD with those affecting the rainfall-runoff process, we imposed the hypothesis that that machine learning algorithms may be able to estimate MTT with acceptable accuracy. To examine this hypothesis, we first developed a coupled hydrological and particle-tracking model to simulate TTD in a virtual catchment with homogeneous soil. We conducted the simulations under four climatic scenarios of Af, Aw, CSa, and CWa, and three suface cover types of bare soil, grassland, and shrubland. We then simulated the MTT time series of water particles with the aid of different machine learning algorithms. The results showed that the neural network model could accurately predict the MTT time series of the water particles that leave the catchment as river discharge. Also, this study showed that the storage time series of stored water in the basin and the evapotranspiration time series have the most and the least impact on the prediction of MTT time series by the neural network model, respectively.
  9. Keywords:
  10. Machine Learning ; Hydrological Modeling ; StorAge Selection Function ; Life Time Adsorber ; Particles Tracking ; Probability Distribution Function ; River Basin Simulation ; Rainfall-Runoff Modeling

 Digital Object List

 Bookmark

No TOC