Validating our API metric against the Thames Water data

The figure below, generated in Steven's 10_analysis_final.ipynb notebook, shows the smoothed daily discharge events (orange) against the smoothed daily API (red) between March 2022 and March 2026.

As explained in more detail on the homepage, our API metric discounts rainfall over a 7 day period, by a discount factor of k=0.85 (chosen as it seems a reasonable estimate from the relevant literature, cited on the homepage).

Note that the number of EDM monitors increased signifiantly until 2023 (when it became mandatory for all sites to have them fitted). We've smoothed both the API and daily discharge events to account for anomalies that cause sudden spikes in the graph (and hence make it difficult to see the overall trend).

Visually, this graph shows that there is a moderate association between the 7 day API metric and discharge events, which motivates our usage of this metric for trying to estimate how much of the sewage spills at other locations in the country can be attributed to rainfall.