Review - Analysis

[jwagemann]

• What exactly is the EventDetector doing? The notebook says that it identifies any pixel that exceeds a threshold. How do you define the thresholds?

• Could you please elaborate more on the interpretation of your results?

[tommylees112]

What exactly is the EventDetector doing? The notebook says that it identifies any pixel that exceeds a threshold. How do you define the thresholds?

The EventDetector allows a user to flexibly identify consecutive periods of above/below threshold timesteps. The thresholds can either be supplied or inferred from the data (Q10, Q90, mean +- 1std etc.). It’s purpose in the pipeline is for prior and posterior analysis of severity and it is relatively self-contained. If necessary we can move it into a experimental location but its role is to help interpret the input/output variables.

Could you please elaborate more on the interpretation of your results?

Of course! When you say interpretation are you referring mostly to the feature importances and the times/places where the model performs well or badly?

We are currently trying to push for a highly performant model given the data that we already have. From this we will be able to demonstrate the interpretation of our results more fully. We are also working on a class that calculates the performance of the model by different groupings (e.g. landcover classes or administrative regions).