Machine learning in hydrometeorology

Topic 1. Technical tools for data analysis and machine learning. Brief review of Python basics. Development tools. Libraries for working with different types of data (numpy, pandas, xarray). Computing resources. Exploratory data analysis, data visualization, and visualization of data distributions.

Topic 2. Probabilistic formulation of ML problems. Maximum likelihood method. Linear regression as an ML method for regression tasks. Loss functions for linear regression. Logistic regression as an ML method for classification tasks. Loss function for logistic regression.

Topic 3. Methodology of the machine learning approach. Feature description of objects and events in machine learning. General workflow for solving supervised learning problems. Normalization of feature descriptions. Evaluation of ML model performance. Cross‑validation approach. Hyperparameters of ML models and their optimization. Uncertainty in ML problems. Estimation of uncertainties in performance metrics, model parameters, and target variables. Bootstrap method. Specific features of hydrometeorological data in the context of ML (autocorrelation, lagged relationships, presence of daily and seasonal variability, etc.). Implementation of the main stages of solving ML problems using the scikit‑learn library.

Topic 4. Overview of main machine learning methods. Parametric and non‑parametric methods. Principles of building different types of models. Optimization of parametric models, gradient optimization methods. Non‑parametric models, optimization of non‑parametric models. Ensemble models and their optimization. Implementation of ensemble ML models using the scikit‑learn library.

Topic 5. Brief introduction to deep learning methods. Fully connected artificial neural network (MLP) as a parametric ML model. Using the MLP implementation from the scikit‑learn package.

Topic 6. Application of ML methods. Formulation of approximation, forecasting, and statistical downscaling tasks. Interpretation of ML results. Feature importance assessment. Application of ML methods for factor analysis in hydrometeorological problems.

Topic 7. ML methods for “unsupervised learning” tasks. Dimensionality reduction, anomaly detection, clustering, and the relationships between these tasks.