Cause-effect Based Modelling for Reliable Results Under Changing Climatic Conditions

Two-minute introductory script

The script used for the PICO introduction.

I shall begin with my results.

Slide 1
- Show point 1
For reliable predictions, robust models are needed, especially under changing climatic conditions.
- Show figure
- Consider the predictions of a correlation based machine learning model of surface soil moisture.
- The model performs very well in the training with R-square close to 0.9.
- Show figure
- But the performance drops sharply when tested during a period of drought.
- Show point 2
- To build robust models, We need to identify robust and direct drivers of processes.

Slide 2
- To do this we used causality.
- For example a correlation based approach would show air temperature, evaporation and humidity correlated. However only a causation based approach would show that air temperature drives evaporation which drives humidity, which later drives the temperature.
- Show text of 4 CD methods
- In our work we tested four diverse causal discovery algorithms.
- Show figure.
- And the results show that causality-based models are more robust than correlation-based models.

Slide 3
To discuss...
- What are causal discovery methods, how do different causal methods work, how we evaluated them,
Please visit our PICO screen A.6