We are aiming to prove that:
following the implementation NFM interventions on the Bourne, the gauge rise at Tidmarsh for a given type of rain event, is lower than before,
following the implementation NFM interventions at Elmwood, the gauge rise at Bucklebury for a given type of rain event, is lower than before.
We aim to prove this by statistical analysis of the rise of the river that results from different rain events. Such an analysis requires a model of how the river rises due to a rain event.
The rise on a river gauge, due to a rain event = R, where R is a function of f(a, b, c, d, e, f) where, the parameters are:
a = "Sum of Rain (mm)". The amount of rain in mm that falls in a rain event (this must be the dominant factor, as no rain = no rise).
b = The level of the river already at the start of the event for the gauge under consideration
c = The soil moisture deficit. We don’t have a value for this, but I’m approximating this by computing the “amount of rain that fell in the preceding 7 days”. Thus c = The amount of rain in the preceding 7 days up to the start of the ‘rain event’. In addition to this factor, season probably plays a factor and thus for both Tidmarsh and Bucklebury we plot Winter (Oct-April) and Summer (May-Sept) season events.
d = The length of the rain event in hours (= end of event timestamp – the start of event timestamp). This is probably not a factor really, but the next one is:
e = Average rain intensity during the event (= a / d in mm/hr)
f = Peak rain intensity during the event (= the highest recorded rain fall in a 15min period during the event in mm/0.25 hr).
Selecting Rain Events
To be a ‘rain event’ it must have been dry for 8 hours before the start of the event, and the end of the event is deemed to have then occurred when it is then dry for 8 hours. All time between the start and the end is ‘the event’, and we consider the rain amounts etc within that period. Events with a rain total < 5mm have been excluded from the analysis. Similarly events in which the Tidmarsh gauge has a min below 125cm are excluded, and when Bucklebury has a min below 15cm and a max above 70cm are excluded. In both cases the gauges behave badly in the excluded cases.
We have chosen 8 hours, as that is how long it takes following the end of a large rain event, for the 'peak' flow to reach the Tidmarsh gauge (from observation of the graphs at www.floodalleviation.uk/dashboard). We want clean rain events that are not poluted by multiple rain events and river peaks being combined. That would make any analysis too complex.
You can chose which gauge values to plot on the y-axis, and which parameters to have on the x-axis. For each, the relevant corelations are shown. Within each, you can chose which further parameter to use to colour the points - this might aid a human in seeing a pattern, thereby deducing whether a secondary parameter is relevant. If you wish to download the dataset (with rain events with total > 5mm only), but without the gauge specific filter, click the green download icon in the top right of the page. To read this info again, click the blue info icon.
The rise in the river due to a rain event, is a function of many factors. It is difficult to model a function with 6 parameters.
The function is too complex and cannot be analysed with simple 2-axis correlations and will perhaps require a machine learning technique, which is not available the PVFF. There may be other parameters required in the function.
The 'before intervention' dataset has been recorded from 2003-2019 (16 years) and the 'after intervention' dataset has been recorded for 1-2 years. The 'after' dataset is probably too short, even if a model could be found, for any statistically relevant conclusions to be drawn.
Thus, without further more complex analysis, without a longer 'after' dataset, no conclusions can yet be drawn against the hypothesis.
All data used is based on rain, as measured on the EA's Yattendon rain gauge, and using the EA's Bucklebury and Tidmarsh river level gauges.