About this module
This module introduces deep learning as the next step beyond the classical statistics and geostatistics of Week 3: fully-connected networks as a baseline, LSTMs for river and groundwater time-series forecasting, and a first look at CNNs for image-based coastal monitoring. It also introduces gradients, sensitivity and model calibration in a hands-on way — the conceptual groundwork you'll need to follow the guest webinar on differentiable hydrologic modeling.
How AI tools fit into how you learn here
Deep learning code fails quietly — a shape mismatch, a forgotten normalisation step, a loss that never moves. This module leans hardest on the prompting workflow we've built since Week 1: describing the symptom to an LLM tutor precisely enough to get a useful diagnosis, asking it to explain why a fix works rather than just applying it, and treating its suggestions as a hypothesis to test in the notebook, not a verdict. That same habit of interrogating a model's reasoning — human or artificial — is exactly what the SAGE webinar addresses from the modelling side.
Topics covered
- ✓ Fully-connected neural networks: a primer for readers coming from Weeks 1–3
- ✓ LSTM networks for river-flow and groundwater time-series forecasting
- ✓ A first look at CNNs for image-based coastal monitoring
- ✓ Hands-on with Python, TensorFlow, Jupyter, QGIS and MATLAB
- ✓ Gradients, sensitivity and model calibration — the bridge to Day 5's SAGE webinar
Looking ahead: the SAGE guest webinar
The sensitivity and gradient concepts introduced here return in a guest webinar by Prof. Jasper A. Vrugt (UC Irvine) on SAGE: Sensitivity-Aware Gradient Estimation for Differentiable Hydrologic Modeling. See the webinar details on the main course page for the full abstract and speaker bio.
Optional: MATLAB Self-Paced Track
Provided by our technology partner MathWorks as a free, self-paced complement to this week's session. Budget around 20–30% more time than the stated duration.
Object Detection with Deep Learning — this week's course ~1 hMATLAB Online access and GPU credits are provided free to registered participants during the course period (August 2026).
Where to find the rest of the material
The slide deck and Colab notebooks for this module are being finalized and will be linked directly from this page before the module opens in August 2026.
Questions in the meantime? Reach the course team via the enquiries section on the main course page.