Loading...

Water Quality Monitoring and Forecasting Based on the Integrated Approach of Deep Learning and Process-Based Model in Stratified Lakes

Moradi, Alireza | 2024

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57527 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Safaie, Ammar
  7. Abstract:
  8. The purpose of this research is to investigate the capability of process-oriented modeling and deep learning for monitoring and forecasting the water quality of stratified lakes and to provide a unified framework integrating these two models to improve the monitoring and prediction of water quality. In this thesis, the process-oriented General Ocean Turbulence Model (GOTM) has been developed to simulate the hydrodynamic and thermodynamic processes of Gull Lake from 2008 to 2017. Subsequently, the MLP deep learning model has been employed to study the lake's key water quality indicators during this period. Meteorological data serves as input for the GOTM. Calibration and validation of the model have been performed using observed lake water temperature data. To monitor the lake's quality parameters—water temperature, dissolved oxygen concentration, and chlorophyll-a concentration—along its depth, meteorological data and thermal structure obtained from the GOTM are input into the deep learning model. The results of the GOTM demonstrate that considering the role of groundwater significantly improves the simulation of lake temperature. For instance, the lake temperature was simulated with a Root Mean Square Error (RMSE) of 1.00 degrees Celsius and a coefficient of determination (R²) of 0.92 when considering groundwater-lake exchange, compared to an RMSE of 3.70 degrees Celsius and an R² of 0.89 when ignoring it. The role of groundwater in the lake bottom temperature is particularly significant, as incorporating groundwater improves the RMSE of lake bottom temperature from 4.27 to 0.69 degrees Celsius. Additionally, defining time-dependent light attenuation coefficients in GOTM has improved the model's performance in simulating the lake's thermal structure. The results of the deep learning model indicate that incorporating the depth and thickness of the thermocline and the thermal structure from the GOTM into the inputs of the deep learning model enhances its performance in monitoring the lake's quality parameters. Consequently, the integrated model is capable of simulating lake temperature, dissolved oxygen concentration, and chlorophyll-a concentration with RMSE values of 0.94 degrees Celsius, 1.46 mg/L, and 0.57 mg/L, respectively, and R² values of 0.97, 0.87, and 0.67, respectively. Furthermore, the deep learning model's accuracy in predicting hypoxia and anoxia days was 91.07% and 80.36%, respectively. A sensitivity analysis of the deep learning model was conducted using the partial derivatives method to assess and identify the relative sensitivity of predictors, including the lake's thermal structure, on the predicted water quality variables. Finally, The results of the projection indicate that stratification will strengthen in spring in the long-term under the SSP5-8.5 scenario using the GFDL-ESM4 model, leading to significant rising trends in temperature, DO, and Chl-a during spring, with trends statistically significant at a 90-95% confidence level based on the Mann-Kendall test and Sen's slope estimator
  9. Keywords:
  10. Deep Learning ; Numerical Modeling ; Generalized Ocean Turbulence Model (GOTM)Turbulent Mixing Model ; Integrated Model ; Climate Change ; Lake Water Quality ; Water Quality

 Digital Object List

 Bookmark

No TOC