Machine Learning for Weather & Climate

Built interpretable machine-learning models for climate ensembles, remote-sensing imagery and environmental time series.

In practice

Models designed around the structure of environmental data.

I use gradient boosting, deep learning and dimensionality reduction for multivariate environmental problems, with attention to lag structure, generalisation, interpretability and the difference between correlation and physical explanation.

Built XGBoost models with lagged EOF/PCA predictors across a 40-member climate ensemble.
Used SHAP analysis to connect model behaviour with interpretable ocean–atmosphere patterns.
Developed a U-Net workflow to reconstruct storm-scale radar structure from visible and infrared satellite imagery.
Applied LSTM and transformer architectures to incomplete climate records and forecasting problems.

Methods & tools

XGBoostRandom forestPyTorchTensorFlowCNN/U-NetLSTMTransformersSHAPPCA/EOF

Selected evidence