This repository supports a research study on estimating lake surface water evaporation using satellite-derived water quality (WQ) parameters and in-situ meteorological (MG) data. We develop Bayesian Optimization (BO)-tuned deep learning architectures (BO-LSTM, BO-GRU) and compare them with their non-optimized counterparts (LSTM, GRU) and a physically based Penman-FAO formulation (baseline).
Status: Active research codebase (manuscript in preparation). Structure and APIs may change.
- Focus on open surface water evaporation.
- Integrates remote sensing–derived WQ parameters (CHL, CDOM, TSM, temperature) with MG drivers.
- Hybrid deep learning: BO-LSTM and BO-GRU (Bayesian hyperparameter optimization).
- Benchmark against Penman-FAO physical model.
- Feature attribution using SHAP for interpretability.
- Demonstrates viability of WQ-only predictors where MG data are sparse.
python -m venv .venv
source .venv/bin/activate # or .venv\Scripts\activate on Windows
pip install -r requirements.txtMIT License (see LICENSE).
This repository is a research vehicle. Model outputs should not be treated as operational hydrological guidance without independent verification.
For any inquiries, please contact:
