Thread by @gilmore_H2O

Image-based #water level measurement from fixed ground-based cameras. This method is based on classic machine vision techniques and requires specialized imagery, but provides visual verification not possible with other methods. There are always tradeoffs to achieve accuracy ...

#opensource software, background template, and tips & tricks are available on our website

A stationary background and camera are important for accuracy

Once images are captured, just load the folder in GRIME2

You need to measure the real-world coordinates of bowties in the field for a precise calibration

One click & you have a calibration grid necessary to determine water levels in real-world units.

One click to run the folder of images. Processing is very efficient, with computational requirements on par with @Raspberry_Pi capabilities.

The algorithms are robust & resistant to issues like bio-fouling. There will be some images where a water edge is not detected, though these cases are few in our experience. These "missed" measurements are flagged and can be evaluated visually.

Results are reported in a .csv file.

As noted previously, it is critical to have stationary background, stationary camera, and a vertical (plumb) background. The calibration process can account from some movement (see our website for details) but it's best to minimize this.

Our 2013 paper in Journal of Hydrology showed that under ideal lab conditions we could approach +/- 3 mm in many cases (about the height of water meniscus). Pending publications will flesh out field performance using robust statistical analysis.

Publication can be found here, or on our website:
sciencedirect.com/science/articl…

For those interested, here are some examples of missed water levels (A & B) and poor water level finds (C). These are flagged in the .csv and the overlay images can be saved for inspection. #opensource #hydrology #gaugecam

We also have diagnostics built in to help understand why measurements may fail. #gaugecam

And move detection to adjust calibration (red horizontal line checks location of top two fiducials). In the current version of the software this is done automatically for all images, producing the "Level (adj)" value shown in the image overlay.