search

USGS-R / protoloads

Prototyping and exploring options for broad-scale load forecasting
0 stars 4 forks source link

fig: exceedance probabilities #58

Closed aappling-usgs closed 6 years ago

aappling-usgs commented 6 years ago

x axis is days ranked by flow

img_20180510_162020071

jzwart commented 6 years ago

I was making this fig and thought I was going down the right route, but then questioned what I was understanding this correctly.

I interpreted the y-axis as the probability that the forecasts exceeded a predetermined flux threshold at a give observed load (x-axis) - I calculate 'probability' as the the sum of forecasts exceeding a threshold divided by the total number of forecasts for a given load. Figure would look something like this, where red dashed line is threshold:
image

aappling-usgs commented 6 years ago

Hmm, this is a very useful prototype. The plot is noisier than what I had imagined, but I also didn't have a super detailed image of how it would look. Let's keep thinking about it...maybe we can do something with a more integrated measure (e.g., average concentration or exceedance rate over a week or month), or maybe we can turn the data in this plot into points and then provide a smoother. And let's keep in mind our goal for this plot, which I think is to direct attention to derived information that's potentially important to stakeholders. Maybe we can find some related plots in stakeholder docs (TMDL reports?) that help us choose a cleaner and still relevant visualization.

jzwart commented 6 years ago

Yeah it's not quite there and I think your suggestions will definitely improve the figure. While looking around for TMDL reports, I found these https://www.epa.gov/tmdl , https://nepis.epa.gov/Exe/ZyPDF.cgi/20004PB2.PDF?Dockey=20004PB2.PDF , https://www.epa.gov/sites/production/files/2015-10/documents/crow_wing_river_mn.pdf which might be useful resources

image

jzwart commented 6 years ago

turning into points and only using lead time of 0 makes it a little cleaner: image

Or having all the points and a smoother applied: image

aappling-usgs commented 6 years ago

Now using this space to show examples of what we mean by exceedance, plus switching to a criterion of >=3 daily exceedances in each 10-day window:

fig_exceedance