Shoreline Assessment Methodology

[kvos]

Discussion of Shoreline Assessment Methodology

This thread is to discuss the methodology to assess the accuracy of the satellite-derived shorelines against the in situ data.

[kvos]

I can propose one option, which works for sites where we have topographic surveys along individual transects (eg, Narrabeen, Duck and Torrey Pines):

1. Transects based approach:

• compute intersections between 2D shorelines and cross-shore transects to obtain time-series of shoreline change.
• apply a water level correction along the transects to obtain tidally-corrected time-series of shoreline change
• compare the tidally-corrected time-series of shoreline change against the in situ time-series of shoreline change for the same contour:
  • compute point-by-point RMSE, bias and STD.
  • compare linear trends

For sites where DEMs or camera-derived shorelines are available, there is the possibility to also compare the shorelines alongshore, as previously done by the UPV team. If anyone could expand on this 'alongshore approach' that would be great.

[kvos]

I have added a new notebook to evaluate the satellite-derived shoreline of time-series of shoreline change against the in situ data:
https://github.com/SatelliteShorelines/SDS_Benchmark/blob/main/2_compare_sat_to_groundtruth.ipynb

You can use this notebook to test your outputs against the in-situ data. Feel free to add to the assessment methodology, for the moment the following metrics are calculated on a transects base as well as grouped into sites:

• RMSE
• STD
• Mean bias
• 90th error percentile
• R2
• Goodness of fit (Chi-squared)

[dbuscombe-usgs]

Could you please expand a little on the 'alongshore approach' - why is this only possible when DEMS or cameras are available? We can extract the shoreline as a continuous vector defined as latitude, longitude ... tides could be applied in 2d, and shorelines could be tracked in 2d, as well. It's just easier to map everything onto transects

[kvos]

at narrabeen for example we only survey 5 cross-shore transects along the 3.5km beach. Duck and Torrey Pines are also transect based. So the comparison between SDS and in situ timeseries can only be done along those transects.

Tidally-correcting the shorelines directly is possible but not that easy, how do you know what is the cross-shore direction? especially on embayed beaches that rotate this is not straight forward and requires some specific transformations (Mitch's log-spiral for example)

[dbuscombe-usgs]

We'll stick with transects, although we are being asked by USGS colleagues to provide continuous shorelines. As you say, that would require knowing the orientation of the coast. I was looking for more specific details about how a DEM could be used here - after all, DEMs are available at some resolution everywhere ...

The challenge with transects is that if the orientation of the coast changes, you need to redraw the transect